Repetitive transcranial magnetic stimulation (rTMS) is an effective treatment for major depressive disorder (MDD), with a modest response rate of 44 % (Cao et al., 2018). This response rate means that a large proportion of participants receiving this treatment will not respond (Cassidy and Fitzgerald, 2019; O'Reardon et al., 2007). rTMS treatment protocols typically involve daily sessions for several weeks and a high level of commitment from patients and clinical staff. High non-response rates can therefore be costly in terms of both time and money, for both the participants and clinics offering the treatment (George and Post, 2011). Being able to predict a person's likelihood of response to rTMS before treatment could help to avoid unnecessary costs. Furthermore, determining how rTMS changes neural activity to exert its antidepressant effects may aid in the search for new treatment options by improving our understanding of the neurobiology of the antidepressant response.

Resting-state electroencephalography (EEG) functional connectivity has been extensively studied across a range of conventional antidepressant medications as a biomarker for antidepressant response (de Aguiar Neto and Rosa, 2019; Khodayari-Rostamabad et al., 2013; Lee et al., 2011). While there are currently no reliable biomarkers of response to treatments for depression (Widge et al., 2018), there is some evidence that EEG measures that predict response to medications for clinical depression do not predict response to rTMS (Widge et al., 2013). For example, Wu et al. (2020) developed a resting-state EEG-based tool for predicting response to the antidepressant medication sertraline. When this predictive tool was used on a large sample of 152 patients with MDD treated with rTMS, it was found that those participants with a low predicted response to sertraline showed high response rates to rTMS, while those predicted to respond well to sertraline did not respond to rTMS. These findings indicate that patients with MDD may be predisposed to respond to different treatment modalities, and therefore independent predictive markers of response to rTMS should be investigated.

A small number of previous studies have investigated early treatment-emergent changes in EEG functional connectivity as a predictors of response to rTMS. Connectivity quantified using phase synchronisation between electrodes in 42 participants prior to and following the first week of a 5-8-week treatment protocol demonstrated some predictive value (Bailey et al., 2019). However, when the same measure was examined in an independent dataset with a larger sample size (n = 193), this finding was not replicated (Bailey et al., 2021). Another study demonstrated predictive validity of treatment-emergent changes in 109 patients with MDD, a finding that was consistent across a range of connectivity metrics including coherence, envelope correlation and a novel measure of spectral correlation in the alpha frequency band (Corlier et al., 2019). However, pre-treatment baseline connectivity metrics alone were insufficient to predict treatment outcome (Corlier et al., 2019).

The neurobiological effects underlying the observed therapeutic response to rTMS remains ambiguous. There is evidence that a complete course of treatment may disrupt resting-state network connectivity measured using functional magnetic resonance imaging (fMRI) (Kotoula et al., 2023), with decreased salience network connectivity found to correlate with antidepressant response in the current study sample (Godfrey et al., 2022). EEG provides an alternative method to quantifying the effects of a complete course of rTMS on functional connectivity. Kito et al. (2017) demonstrated an increase in source space functional connectivity in the beta frequency band between the left dorsolateral prefrontal cortex (DLPFC) stimulation site and limbic regions in 14 participants following rTMS treatment. Another study also demonstrated increased functional connectivity induced by rTMS treatment in the delta and gamma bands between left frontal and right parieto-occipital areas in 18 participants (Zuchowicz et al., 2018). Furthermore, Mitoma et al. (2022) reported an increase in theta band oscillatory functional connectivity between the left fronto-temporal and occipital regions following rTMS treatment in 15 participants. Together these studies provide initial evidence of modulatory effects of rTMS treatment on EEG connectivity.

This study used EEG to investigate source localised functional connectivity in 28 patients with treatment-resistant depression before and after a course of daily rTMS therapy administered to the left DLPFC. EEG provides good temporal resolution; however, the spatial resolution is typically poor, with the signal recorded at the channel site representing a linear combination of source-level signals from all over the cortex (Nunez et al., 1994). Therefore, connectivity inferences based on these signals can be erroneous, and it is preferable in some applications to analyse signals in the source-space where the signals are at least partially unmixed (Koutlis et al., 2021; Schoffelen and Gross, 2009). To overcome signal leakage at the source-level, an orthogonalised amplitude analysis approach was implemented, using 38 source-level cortical regions of interest (ROI) across frequency bands spanning 1-67 Hz, to remove instantaneous cross-correlations between ROIs (Colclough et al., 2015; Hipp et al., 2012). Any remaining correlations are thought to represent true functional connectivity between ROIs (Colclough et al., 2015; Hipp et al., 2012). Symmetric orthogonalisation methods are used as an alternative to pair-wise orthogonalisation for overcoming signal leakage at the source level. Symmetric orthogonalisation removes instantaneous cross-correlations between all ROIs simultaneously while maintaining better sensitivity to connectivity than pair-wise approaches (Colclough et al., 2015). Amplitude based connectivity was measured, with amplitude of different frequency bands across time correlated between nodes. This method was chosen as it has demonstrated the closest relationship with functional magnetic resonance imaging (fMRI) connectivity, for which treatment effects were found within the same patient group as the current study (Godfrey et al., 2022; Hall et al., 2014). Source-space power was also calculated to investigate potential changes in frequency power, as these fluctuations can influence amplitude-based connectivity across analyses, potentially confounding the results (Cohen, 2014; Muthukumaraswamy and Singh, 2011).

The current study aimed to determine if source localised functional connectivity could be a useful method for identifying correlates of response to rTMS treatment. Pre-post differences in EEG functional connectivity were assessed, as well as correlations between changes in connectivity and the antidepressant response to rTMS treatment. We also examined the relationship between baseline EEG connectivity and changes in Montgomery-Asberg Depression Rating Scale (MADRS) scores to test if baseline connectivity features were associated with greater response to treatment (Montgomery and Asberg, 1979). All these analyses were considered to be exploratory.