Mood disorder [major depressive disorder (MDD) and bipolar disorder (BD)] is a heterogeneous and complex psychiatric condition. It is a major public health issue, ranking as the leading cause of disability worldwide, and the burden of mood disorders continues to grow despite the availability of validated interventions (1). MDD and BD both exhibit similar severe depressive symptoms (major depression, MD) (2).

The primary approaches to deal with MD include pharmacotherapy and psychotherapy. Although these approaches are effective, they still leave a significant proportion of patients with incomplete remission (3). This frequently results in TRD, which is associated with significant morbidity and high suicide risk (4).

As a result, several alternative treatments have been developed to target TRD, one of which is repetitive transcranial magnetic stimulation (rTMS) (5, 6). TMS is a non-invasive brain stimulation procedure that applies repeated magnetic pulses over the scalp to generate an electrical current in the cortex, provoking electrophysiological effects that modify the neural excitability in the target area and correlated brain networks (7). Safe profile (particularly the lack of systemic side effects associated with pharmacotherapy) and improved focality are some of its advantages over other neuromodulation techniques, such as electroconvulsive therapy (ECT).

Over 150 randomized controlled trials (RCTs) in unipolar and bipolar depression have been carried out, and their efficacy has been confirmed in multiple meta-analyses (8, 9). Moreover, real-world data have also confirmed the effectiveness of rTMS for major depressive disorder in clinical practice (10), with the most recent literature indicating response rates of 40–50% and remission rates of 25–30% (11).

The rTMS is effective in major depressive disorder but presents a high interindividual heterogeneity of clinical effectiveness (12). Moreover, this technique is costly in the real world and requires significant financial and time commitment from the patient and the practitioner. These elements highlight the pressing need for clinical and biological markers to predict treatment outcomes.

Clinical factors associated with rTMS response are divided into three main categories (13): patient-related factors (e.g., age, gender, and treatment resistance); illness-related factors (e.g., bipolar depression, duration and severity of depression, therapeutic resistance, previous response to rTMS or ECT), and TMS procedure-related factors (e.g., TMS intensity, number of pulses per session, and number of sessions) (14). Despite the extensive literature, data remain heterogeneous and contradictory with the need to be pursued.

Recent studies suggest that neuroimaging markers may achieve higher predictive accuracy than clinical or demographic variables [for review, see (15)]. Earlier studies showed promising results using methods derived from resting-state functional MRI or diffusion-weighted MRI (16). However, these biomarkers involve complex imaging protocols with few patients and a very specific patient selection. These imaging methods are rarely available in clinical settings, need specialized data processing, and are costly. Therefore, as suggested by Baeken et al., (17), simpler biomarkers like cortical thickness measures, as derived from anatomical MRI data, could be more feasible in current clinical practice. Indeed, to date, two studies have explored cortical thickness before stimulation as a predictor of rTMS response in patients with MDD (17, 18).

We conducted a retrospective naturalistic study to evaluate whether clinical factors or cerebral cortex thickness and volume may be a potential biomarker of rTMS treatment response in drug-resistant patients with depression of a large dataset from patients that received rTMS for the treatment of depression in a real-world clinical setting of a tertiary referral hospital.

Between September 2014 and February 2018, we revised and analyzed the records of 131 patients with mood disorders who received rTMS treatment in the Neurostimulation Department of Henri Laborit Psychiatric Hospital. Each patient was treated only one time with rTMS. Non-opposition to the use of the participants' research data was obtained retrospectively. All patients provided informed written consent, and the study was registered at the Health Data Hub platform (F20210128152411).

All patients in our study population were taking antidepressants and/or thymoregulators, with 3.2% on lithium and 22.1% on anti-epileptic drugs prescribed for thymoregulatory purposes (clonazepam, lamotrigine, oxcarbazepine, and valpromide), with the exception of pregabalin, which was prescribed as an anxiolytic. The majority of patients were treated with one or more antidepressants: 40% were on SSRIs, 27.4% on SNRIs, 24.2% on tricyclic antidepressants, 5.3% on tetracyclic antidepressants, 3.2 % on MonoAmine Oxidase Inhibitors (MAOI), and 8.4 % were prescribed other types of antidepressants. Benzodiazepines or related drugs were prescribed for a majority of patients (80%) and antipsychotics for approximately half of the patients (53.7%).

Of the 131 patients, 20 refused to participate in the study. The final assessment of the MADRS score was missing for 16 patients. For one patient, the MADRS assessment on day 14 was missing. For 94 patients with neither of these variables missing, the evolution between day MADRS0 and MADRS14, as well as the evolution of MADRS0 and MADRS45, was computed and found to be moderately correlated (Pearson's r, r = 0.475, p < 0.001). As a result, we decided to exclude from the following analyses any patient whose final assessment was missing. We, therefore, included 95 patients. Some clinical factors, such as depression type (p = 0.034), sex (p = 0.022), and Thase and Rush resistance stage (p = 0.053), play an important role (Table 1). Table 2 highlights the differences between unipolar and bipolar patients.

If the duration of illness or current episodes was missing from subsequent analyses, it was replaced by the mean of the entire population. Five additional patients were excluded due to a failed parcellation or a faulty MRI acquisition, resulting in a sample size of 90 subjects for the analyses relying on MRI data.

None of the patients participating in the study reported having seizures or shifting between hypomanic/manic states. Patients did not complain about local pain or dizziness after stimulation. Moreover, none of the patients discontinued treatment due to adverse effects.

In this retrospective and naturalistic study, we aimed to identify clinical and neuroimaging factors associated with the efficacy of rTMS evaluated 1 month after the beginning of treatment in 95 patients with drug-resistant depression who were treated with high-frequency rTMS (20 Hz) over the left DLPFC.

The response rate at 1 month after rTMS treatment in our study is 40%. This is consistent with previous studies reporting a 40–50% response rate to rTMS treatment in MDD (33, 34) and BDD (35–37).

Our analysis of clinical variables (type of depression, illness severity evaluated by MADRS and Beck, resistance stage, associated treatment, gender, and resting state) demonstrated that the type of depression, gender, and resistance stage are associated variables with response to rTMS.

First, we found that bipolar depression responds better to HF rTMS over the left DLPFC. To our knowledge, no naturalistic study to date has highlighted bipolar illness as a factor in better response to rTMS or has established a correlation between the type of depression and the clinical effect of rTMS (12, 38). Treating bipolar depression is clinically challenging as antidepressants can worsen the outcome for this category of patients, which is why rTMS has been suggested as a treatment option for bipolar depression (39). Patients with bipolar depression were enrolled in studies focused primarily on unipolar major depression or in dedicated sham-controlled studies examining the efficacy of rTMS in bipolar depression. Three meta-analyses and quantitative syntheses have been conducted to date (9, 40, 41). According to Nguyen et al., active-rTMS is associated with a higher response rate than sham-rTMS in bipolar depression, but subgroup analyses testing differences based on stimulation target and site revealed no significant differences. However, when analyzed separately, HF over the left DLPFC stimulation was associated with a statistically significant greater response than sham treatment. In contrast, bilateral stimulation and low-frequency rTMS over the right DLPFC were not.

We identified the effectiveness of rTMS in bipolar depression in our study. This result differs from that of Yang et al. (42). In fact, in their naturalistic study with an adequately large cohort of participants, they suggest that patients with BD are less likely to achieve clinical response with high-frequency L-DLPFC rTMS than those with unipolar depression (10 Hz). The antidepressant response to rTMS might vary with stimulation frequency. In our study, we used 20 Hz stimulation in seven sham-controlled studies (43–49) in patients with mixed depression and one naturalistic study in patients with BDD (50). Apart from the same frequency of stimulation, there is still significant methodological heterogeneity between studies, including trial duration, stimulation intensity, and several pulses per session, which makes it difficult to compare different findings. Therefore, the stimulation parameters used in our study could account for the improved response in patients with BDD. This difference in response to treatment could be explained by differences in the clinical expression of bipolar and unipolar depressive episodes (51) and indeed by the potential differences in the neurophysiological mechanisms that cause them.

Second, in our study, patients with a high treatment resistance stage respond less to rTMS. The refractoriness of depressive episodes appears to be one of the best-supported predictors of rTMS response. Many studies have suggested that a higher degree of medication resistance may be tied to worse rTMS outcomes in depression (13). Most findings from rTMS response predictor studies suggest that a lower degree of drug resistance is one of the more robust predictors of superior outcomes for rTMS therapy using standard stimulation parameters and targeting methods. While definitive prospective studies are still needed, the existing literature appears to support the use of standard rTMS therapy relatively early during the treatment, prior to the occurrence of numerous medication treatment failures.

Third, we found that women respond better to rTMS than men do. The association between female gender and response to treatment is debatable. A recent meta-analysis, which included 54 sham-controlled trials between 1997 and 2013, revealed that gender might be a positive predictor of response, as studies showing good antidepressant response to rTMS had more female patients (52). In fact, in our study, women (56.8 %) outnumber men (43.2 %). In a second meta-analysis, the same authors show that the antidepressant effect of specifically HF rTMS was higher in RCTs with a greater proportion of female patients (53). They suggested that women's profiles, rather than their sex, might have influenced their response to treatment.

Other clinical and demographic variables, such as age, associated treatment, and resting motor threshold, had no effect on treatment response in our study. The link between treatment response and these variables is debated in the literature. Some studies revealed that younger patients respond better to rTMS (54, 55), whereas other studies found no correlation between age and response (56). Regarding associated treatment, there is growing evidence that concomitant use of medication can impair the clinical effectiveness of rTMS, especially for benzodiazepines (57). However, the impact of concomitant medication on rTMS effectiveness is still debatable (58, 59). Finally, we did not find a link between baseline RMT and response. The correlation between clinical efficiency and stimulation intensity is not precisely known (60, 61). Some studies suggest a dose efficiency correlation. Another hypothesis assumes a more complex and non-linear correlation.

We also investigated the relationship between structural neuroimaging variables (whole-brain analysis, thickness of the left DLPFC and ACC, volumes described in the literature, and distance between scalp and cortical target) and response to rTMS. Neuroanatomical predictors may be particularly useful for brain stimulation interventions, which directly modify the activity within neural circuits, contrary to indirect reorganization caused by psychological or pharmacological interventions.

We initially performed an exploratory analysis of thickness and volume across the brain, in addition to the main cortex-wide analysis that was unconstrained by a priori hypotheses. We also performed ROI analyses to further evaluate the a priori hypotheses and then explored specifically the link between the left DLPFC and rostral ACC thickness at baseline and treatment response.

We did not find a statistically significant link between cortical thickness and treatment response. The link between cortical thickness and the responses to ECT (62), tDCS (63), and rTMS (17, 18) has been reported before. Boes et al. (18) describe that cortical thickness in the left rostral anterior cingulate cortex region correlates with rTMS treatment response in 48 TRD patients treated with HF (10 Hz) rTMS over the left DLPFC baseline. In fact, patients with thinner cortex before treatment tended to have the most clinical improvement. Baeken et al. (17) recently suggested that baseline cortical thickness in the right caudal part of the anterior cingulate cortex was significantly correlated with direct clinical responses in the subgroup that received active aiTBS (21 patients) over the left DLPFC during the first stimulation week, but no correlation was found with delayed response. In this study, we did not confirm the results of these two preceding studies. In Boes and Baeken study's, no accurate correction for multiple comparison testing was performed, increasing the risk of type I error. This can partly explain why we did not find the same results. Different stimulation parameters and the number of patients are also potential explanations for this discrepancy. Moreover, increased cortical thickness after rTMS treatment has been described in longitudinal studies (18, 64).

Response to rTMS was also evaluated in terms of the cortex-scalp distance. According to Lee et al. (65), the non-invasive brain stimulation scalp-to-cortex distance has been reported to critically influence the focality and strength of the electric field induced by rTMS. Our result indicated that there is no difference inefficacy related to this distance. Kozel et al. (66) discovered that these distances do not directly correlate with antidepressant clinical response in 29 depressed patients, but a correlation was established between the motor threshold measurement and the distance from the cortex to the skull under the TMS coil. In our study, the scalp-motor cortex distance was not possible because the stimulation point was not recorded during each session. Moreover, the lack of statistical correlation between response and distance in our study could have been influenced by the position of the stimulation coil. This could be in part due to the fact that the exact targeting zone for every session is unknown. One could also question the validity of the chosen region of interest (10 mm radius sphere around the average theoretical cortical target.

Gray matter volume (GMV) at baseline has been previously described as a predictor of treatment response in mood disorders. We first performed a whole-brain mapping without an a priori hypothesis for a specific brain volume. Interestingly, when taking into account depression type, gender, age, resistance stage, duration of illness and episode, and baseline MADRS score, this whole-brain analysis associated with clinical factors in a regression model shows that the average volume of the left part of the superior frontal gyrus and the caudal middle frontal gyrus was associated with the status of the response, where responders present a lower gray matter volume. The same result could not be reproduced in the right hemisphere.

The superior frontal gyrus and the middle frontal gyrus are usually defined as a part of the DLPFC (67). The superior frontal gyrus contributes to higher cognitive functions (68). It is part of the “hate circuit,” which is involved the pathogenesis of depression symptoms, risk and action responses, attention, reward, and emotion (69). The middle frontal gyrus is critical for higher-order executive functions related to stress perception and appraisal, including attention, working memory, planning, executive cognition, and emotion regulation (70–72), and may confer vulnerability to depression. The GMV deficits of the frontal cortex have been reported in several studies on MDD. Abe et al. (16) found that patients with MDD might have GMV deficits in frontal-temporal-limbic regions, which also included the middle frontal gyrus. Leung et al. (73) found that attention biases toward negative stimuli are associated with a reduced gray-matter concentration in the right superior frontal gyrus. Lai et al. (74) found a GMV increase in the frontal lobe after treatment with aripiprazole in patients with depression and deficits in the superior and medial frontal gyrus for patients with MDD at baseline status. Moreover, Lai et al. (74) compared structural differences between patients who were able to achieve remission and those who responded poorly to antidepressants. The remitting MDD patients showed a bilaterally smaller superior frontal gyrus volume. Yuan et al. (75) found that geriatric patients with depression in remission from their first episode of depression had reduced GMV in the right superior frontal gyrus in comparison with well-matched healthy controls. Although the nature of the involvement of the superior frontal gyrus and the caudal middle frontal gyrus in mood disorders remains a matter of debate, in our study, a greater volume in the left part of the superior and the caudal middle frontal gyrus was observed in non-responder patients. The association of this region with response to rTMS was not previously described in structural MRI studies that investigated response factors.

No statistically significant differences in baseline structural volumes were found between treatment responders and non-responders. Few studies have investigated the association between GMV and treatment response in MDD. Treatment response was evaluated for cognitive behavioral therapy (CBT) (76), antidepressant (77), ECT (78, 79), and rTMS (80). The study found a link between clinical response and the volumes of the left and right hippocampus, the left and right amygdala, the left and right ACC, and the left and right insula (28). In our study, none of these regions were found to be associated with response to rTMS.

Despite the strengths of this study (larger number of patients than already described in the literature, correction for multiple comparison testing with independent logistic regression models, naturalistic design), several limitations must be considered. First, all of the patients included in this study were under medication; prior exposure to medication is a strong confounding factor as it may affect the brain structure. In fact, Hoexter et al. (81) found that the thickness of the orbitofrontal cortex in patients with OCD can serve as a predictive biomarker of treatment response, exclusively in treatment-naive patients. Second, our investigation does not have a placebo-controlled group, which means that any predictive biomarker of treatment response could be confounded by the placebo effect. Third, our patients have multiple comorbidities, which may have confounded our results. Finally, MRI was performed only at baseline. In the future, recording structural MRI data at multiple time points to retrieve information about structural changes after rTMS is recommended.

There remains significant interest in understanding how to optimize the application of rTMS for each patient in order to achieve greater remission rates and provide more efficient symptom relief. The use of structural MRI is an essential tool to achieve this objective. Besides, this type of imaging is easy to perform, and collaborations between several centers can be envisioned to allow for the acquisition of a sufficient volume of imaging and clinical data in the future to establish solid correlations. Moreover, it would be interesting to characterize the structural covariance networks (SCN) to better understand the response to rTMS. SCN analyses aim to identify network patterns of common influences and characterizations within the brain across the population rather than differences in the structure of isolated brain regions in individuals (82). In addition, MDD is associated with deregulation of neural networks rather than a disruption of individual brain regions in isolation (80). Recently, preliminary evidence suggested that gray matter could be used to distinguish rTMS responders and non-responders, particularly in the fronto-parietal network (83, 84).

In conclusion, we investigated the clinical and neuroimaging biomarker associated with the response to high-frequency rTMS over the left DLPFC in a large sample of patients with TRD depression. Women, patients with BDD, and patients who are less resistant were found to respond better to HF rTMS. Responders present a lower volume of the left part of the superior frontal and the caudal middle frontal gyri. The thickness of the DLPFC and ACC, the volumes of the amygdala, hippocampus, ACC, insula, DLPFC, and the distance from the scalp to the target were not associated with the clinical response. Our results reinforce the need to identify accurate and reliable clinical and neuroimaging biomarkers of treatment response. This biomarker that can be translated into clinical practice holds promise for the advancement of precision medicine. Our findings may serve as a guide to future studies with larger datasets to investigate specific neuroimaging biomarkers (the distance between scalp and target and specific volume) and clinical biomarkers (sociodemographic and clinical characteristics), with the ultimate aim of defining a multimodal biomarker profile that predicts rTMS treatment response.

The raw data supporting the conclusions of this article will be made available by the authors, without undue reservation.

Ethical review and approval was not required for the study on human participants in accordance with the local legislation and institutional requirements. Written informed consent for participation was not required for this study in accordance with the national legislation and the institutional requirements.

GH-G, IW, DD, and NJ conducted the patient assessments. RG provided MRI images of the patients. GH-G, NL, and TL and performed analyses. The manuscript was written by GH-G and TL and all authors revised and proofread the manuscript.

This study was supported by recurrent internal funds from the Center Hospitalier Henri Laborit, Poitiers, France.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.