Association of Repetitive Transcranial Magnetic Stimulation Treatment With Subgenual Cingulate Hyperactivity in Patients With Major Depressive Disorder

Key Points Question Is repetitive transcranial magnetic stimulation associated with changes in subgenual cingulate cortex (SGC) activity in patients with major depressive disorder (MDD)? Findings This diagnostic study, which compared 30 patients with MDD and 30 healthy controls, found that using transcranial magnetic stimulation combined with electroencephalography, SGC activity in patients with MDD was significantly higher compared with healthy controls. After active repetitive transcranial magnetic stimulation, SGC hyperactivity in patients with MDD was attenuated toward the levels of healthy controls. Meaning These SGC-localized findings support SGC hyperactivity as a central construct in the pathophysiology of MDD, which future work might develop into a clinically significant biological target.


Introduction
Major depressive disorder (MDD) is a debilitating psychiatric condition with a 16% lifelong prevalence. 1 Antidepressant medication is widely used and has been researched for more than 4 decades. However, the mechanism of antidepressant action remains unclear, and nearly 40% of patients with MDD experience persistent depression even after 2 antidepressant treatments. 2 To improve outcomes, the neurophysiology of MDD and underlying therapeutic mechanisms of treatments need to be better understood.
Hyperactivity of the subgenual cingulate cortex (SGC) is associated with MDD pathophysiology.
Based on positron emission tomography studies, 3 the SGC was shown to be overactive in the depressed state and during transient sadness in healthy controls. Furthermore, higher SGC activity was also found in patients with MDD in both depressed and remitted states. 3 The SGC has also been a primary target for deep brain stimulation in the treatment of MDD. 4 Evidence also implicates dorsolateral prefrontal cortex (DLPFC)-SGC connectivity in MDD pathophysiology. An inverse association of the SGC with right DLPFC activity was found in patients with MDD and healthy patients, 3 implying that there is a functional association of the SGC with the DLPFC. Moreover, functional magnetic resonance imaging showed that DLPFC-SGC activity is anticorrelated, 5 and the magnitude of this anticorrelation can predict the antidepressant efficacy of repetitive transcranial magnetic stimulation (rTMS) targeting the DLPFC. Finally, both positron emission tomography and functional magnetic resonance imaging studies have shown that baseline SGC activation abnormalities were attenuated toward healthy levels using a variety of antidepressant treatments, including conventional antidepressant medications, 6-9 rTMS, 10,11 electroconvulsive therapy, 12 and deep brain stimulation. 4 Combining TMS with electroencephalography (TMS-EEG) is a noninvasive technique that is used to assess neuronal activity, 13,14 connectivity, 15,16 and plasticity. 17 Compared with other neurophysiological methods, TMS-EEG enables a more reliable and causal assessment of the neurophysiology of brain function. This is owing to the fact that TMS-EEG stimulates the cortex directly without dependence on prior activation of upstream brain functional regions (eg, sensory constructs). 16 Significant current density (SCD) is a standardization measure that sums all of the significant TMS-responsive current sources (compared with baseline). Significant current scattering (SCS) is a standardization measure for inferring activation propagation; SCS may also measure the effective connectivity of a source relative to the stimulation site. 13,18 Previously, SCD has been computed to differentiate between patients with Alzheimer disease and healthy controls. 19 The 2 measures have also been used to capture changes in brain network connectivity during loss of consciousness, 18 saccadic movement, 20 and task performance. 21 Significant current scattering has also been effective in distinguishing patients with schizophrenia from healthy controls. 22 We used TMS-EEG measures of SCD to evaluate SGC excitability and TMS-EEG measures of SCS to evaluate DLPFC-SGC effective connectivity in patients with MDD. These measures were taken at baseline and after rTMS treatment. We hypothesized that SCD would demonstrate higher excitability in the SGC of patients with MDD. We also hypothesized that SCS would show a stronger effective connectivity between the left DLPFC and the SGC in patients with MDD. Finally, we hypothesized that increased excitability in (ie, SCD) and connectivity to (ie, SCS) the SGC would be attenuated after applying rTMS over the DLPFC in patients with MDD.

Recruitment and Treatment
Overall, 121 participants with MDD (77 [63.6%] women) were recruited at the Centre for Addiction and Mental Health in Toronto, Ontario, Canada. These patients participated in a randomized clinical trial, as previously described. 23 The demographic and clinical characteristics are described in the eTable in the Supplement. The adverse effects of rTMS on study participants have been previously described. 23

Acquisition and Preprocessing of TMS-EEG Data
In all groups, a figure-of-eight coil connected to Magstim 200 stimulators (Magstim) was used to stimulate the brain over the left DLPFC. Locating the motor cortex was done by eliciting motorevoked potentials at the right abductor pollicis brevis. The left DLPFC was identified using the miniBIRD neuronavigation system (Ascension Technologies). Before each experiment, resting motor threshold was determined, as previously described. 26 Each patient's stimulus intensity was determined as the intensity eliciting peak-to-peak amplitude of 1 mV averaged over 20 trials.
Electroencephalography was performed using a 64-channel Synamps 2 EEG system (Compumedics Neuroscan). It was recorded applying a 200-Hz low-pass filter at a 20-kHz sampling rate. All electrodes (silver/silver chloride ring) impedances were kept less than 5 kΩ throughout the session, Finally, the inverse solution was computed based on the standardized low-resolution brain electromagnetic tomography algorithm, 33 with dipoles constrained normally to the cortex surface.
For each patient, the source localization procedure generated a 15 002-voxel current density map in brain space for every point of the TEP.
Significant current scatter was calculated based on the following equation adapted from methods previously published in the article by Casali et al 13 :

Statistical Analysis
Both bilateral and unilateral rTMS treatment groups were pooled together into the active rTMS treatment group owing to statistical power considerations. To avoid normality assumptions, the Wilcoxon rank sum test was used to examine the differences between current density, SCD, and SCS in patients with MDD and healthy controls and in patients with MDD in the different rTMS treatment arms. Statistical significance was set at P < .05, and all tests were 2-tailed. No correction for multiple comparisons was applied. A Spearman correlation was used to quantify the association of SCS change with HRSD-17 score change before and after treatment. Finally, a receiver operating characteristic analysis was computed for the pre-rTMS sample; a source current density amplitude window from 15 milliseconds to 350 milliseconds was used as the criterion to classify healthy controls vs patients with MDD. This time segment was chosen because significant sources dropped drastically 350 milliseconds after stimulation. The probability of correct prediction was quantified by the area under the receiver operating characteristic curve, 35 while the optimal threshold was determined as the source current density value associated with the maximum Youden index, or height above the diagonal line of no discrimination. 36 All statistical analysis was done using MATLAB version r2017b.

Results
Overall  (Figure 2A). These differences were localized in voxels in the region of the SGC, and their timings aligned with standard TEP temporal components (P30 milliseconds, N100 milliseconds, and P200 milliseconds  (Figure 3).
We then compared patients with MDD after active vs sham rTMS treatment ( Figure 2B).

Differences in mean (SD) current density between unilateral and bilateral rTMS treatment groups
were not statistically significant ( μA/mm 2 ; z = −1.00; P = .31). Therefore, we pooled these treatment groups to achieve better averaged neurophysiological signals and greater statistical power. Marked differences of current density, SCD, and SCS were found between the 2 experimental groups and were also localized in SGC-related voxels. These timing differences were also associated with the known TEP components (P60 milliseconds and P200 milliseconds). Mean (SD) current density in the SGC was higher for the sham group compared with the active rTMS group at 50 milliseconds (9.  This comparison is also presented in Figure 4 as a region of interest  Figure 5A shows the association of SGC source current density with HRSD-17 score in participants with MDD before rTMS treatment (ρ = 0.41; P = .03). The correlation shows that symptoms of depression were more pronounced when SGC current density was higher at 100 milliseconds after the TMS pulse. Figure 5B shows a correlation of the change in HRSD-17 score from baseline to post-rTMS treatment with change in SCS Model quality is demonstrated by the area under the curve (AUC). Based on Youden index, the optimal model (marked by dot) differentiates major depressive disorder from healthy control with 77% accuracy (70% sensitivity and 83% specificity).

Discussion
This study's results show that patients with MDD had higher-amplitude SGC-localized source measures and higher DLPFC-SGC effective connectivity compared with healthy controls. We found that SCD indexing had a high accuracy in discriminating patients with MDD from healthy controls and The SGC-DLPFC interconnectivity may, in part, be governed by γ-aminobutyric acid (GABA)ergic neurotransmission. 45 Because rTMS is capable of modulating deep brain structure activity transsynaptically, 26 the observed reduction of the effective connectivity signal (ie, SCS) between those 2 regions might be associated with the reconfiguration of these long-range connections. The SGC current density at approximately the 100 millisecond peak, its correlation with symptom severity at baseline (Figure 5A), and the correlation between symptom change and SCS signal change at approximately 100 milliseconds ( Figure 5B) may also be associated with GABAergic neurotransmission because the evoked potential at approximately 100 milliseconds post-TMS pulse is associated with neuronal inhibitory processes. 46-49 Using the EEG sensor distribution for localizing electrical generators in the cortex yields different possible spatial arrangements of source current generators. Incorporating some assumptions regarding brain anatomy and electrophysiological energetic constraints into the source localizing computation will restrict the multiplicity of possible solutions, but it will remain nonabsolute. Further, the immense spatial and temporal complexity of brain activity and the noisy nature of the EEG signal increase uncertainties when trying to solve this inverse problem. 50 However, when TMS-EEG is applied, the timing and location of brain activation is constrained by the experimenter, reducing substantially the short latency inaccuracies of the source estimations. Additionally, as TMS was applied over the DLPFC, we can rely on prior knowledge regarding DLPFC connectivity and its association with our main region of interest, the SGC. This supports the potential of SGC activation after longer latencies following the TMS induction.
Apart from the advantages of using TMS-EEG to assess activation of localized generators in the brain mentioned earlier, we also implemented SCD and SCS statistical indexing over our data, which further mitigates erroneous source evaluations. On its own, simple signal averaging is a strong tool for reducing noise in our electrophysiological recordings. When applied over the averaged signal, SCD is far more statistically stringent by taking into account the whole distribution of the activationmaking this indexing robust when dealing with signal outliers. 13 The SCS takes the SCD approach a step further by accumulating the distances between significant sources in the SGC and the site of stimulation. Hence, the SCS computation estimates the effective connectivity between those 2 sites after stimulation. The methodological and computational steps we took to ensure our data quality and the fact that our a priori hypotheses were supported by our results were key in making our source estimations findings more deterministic and valid.

Limitations
This study had some limitations. Our study design allowed only a between-participant rather than within-participant statistical inference, which is not ideal in cases that look at the potential effects of treatment. Despite this issue, each group in this study had an adequate number of participants.
Moreover, the measured associations produced highly comparable values between the independent groups, enabling valid statistical inferences. It is important to note that to achieve adequate neurophysiological signals, the active treatment group was pooled from the bilateral rTMS treatment arm and the unilateral rTMS treatment arm (as depicted in Figure 1). However, the unilateral and bilateral treatment groups did not show statistical differences in EEG activations. As another limitation, patients with MDD in this study were receiving heterogeneous antidepressant pharmacotherapy. This fact may have confounded our clinical symptom assessments and EEG signal observations. However, the use of sham control treatment and the finding that only the active rTMS group demonstrated a correlation of the EEG signal with symptom improvement suggests that these changes may stem from the rTMS treatment itself rather than from concomitant pharmacotherapy.
Moreover, the depression change score had a stronger correlation with SCS change after active treatment compared with the correlation after sham treatment. However, our sample size was likely not large enough to find a significant difference in these 2 correlations.

Conclusions
In conclusion, this study demonstrated the usefulness of TMS-EEG and the SCD/SCS computation when investigating the association of SGC activation and DLPFC-SGC effective connectivity with MDD pathophysiology and clinical improvement. Left DLPFC rTMS may have improved MDD symptoms by altering the connectivity between the DLPFC and the SGC, likely via GABAergic neurotransmission. These findings support the hypothesis of SGC involvement in the