Major depressive episodes and treatment-resistant depression (TRD) remain major public health concerns with incompletely characterized network-level neurobiology. This study aimed to identify resting-state electrophysiological connectivity markers within canonical brain networks that distinguish individuals with current major depression and those with TRD from healthy volunteers (HVs). The authors sought candidate biomarkers using magnetoencephalography (MEG) and multivariate integration techniques.
The sample included 168 participants: 80 healthy volunteers and 88 individuals experiencing a current major depressive episode. Among the depressed participants, 74 met criteria for treatment-resistant depression (TRD) and 14 were classified as noTRD (depressed without treatment resistance). Group composition and counts were reported by the authors and form the basis for subsequent analyses.
Resting-state electrophysiological data were acquired with MEG. Analyses targeted within-network connectivity across six canonical networks: the default mode network (DMN), executive control network (ECN), salience network (SN), dorsal attention network (DAN), motor network (MN), and visual network (VN). Connectivity was assessed across frequency bandwidths; specific frequency bands implicated in the results include alpha and beta.
The study applied Data Integration Analysis for Biomarker Discovery using Latent Variable Approaches for Omics Studies (DIABLO), a multivariate classification and feature-selection framework, to differentiate diagnostic subgroups and to identify network connectivity markers associated with depression and treatment resistance. DIABLO was used to compare a triple-network model (DMN, ECN, SN) with a broader six-network model across bandwidths and to derive features contributing to group discrimination.
When discriminating the depression group from healthy volunteers, the authors report that the triple network model (DMN, ECN, SN) outperformed the six-network model across bandwidths. Reported area under the receiver operating characteristic curve (AUROC) ranges for the triple-network model were 0.759–0.787, compared with 0.747–0.762 for the six-network model. These values indicate better separation with the triple-network feature set in the present sample.
For distinguishing participants with TRD from healthy volunteers, the triple-network model also demonstrated reasonable predictive performance across different bandwidths, with reported AUROC values ranging from 0.737 to 0.807. The authors highlight that within-network connectivity features contributed to this discriminative capacity.
A specific within-network connectivity difference that distinguished TRD from HVs was identified in the DMN: connectivity between the inferior parietal lobule and the precuneus in the beta band. This DMN within-network connectivity difference reached statistical significance after false discovery rate correction (FDR-corrected p < .05), as reported by the authors.
The study reports that increased number of treatment failures was associated with hyperconnectivity in two connections. First, within the SN—between the superior parietal lobule and the frontal operculum—in the alpha band. Second, within the DMN—between the inferior parietal lobule and lateral prefrontal cortex—in the beta band. These associations were reported as statistically significant (ps < .05), indicating a relationship between spectral-band within-network hyperconnectivity and clinical severity or treatment nonresponse history.
The findings emphasize the importance of the triple network (DMN, ECN, SN) in the electrophysiological signature of major depression and TRD. In particular, DMN beta-band connectivity between parietal and precuneus nodes, and SN alpha-band hyperconnectivity, emerged as candidate biomarkers associated with treatment resistance and with the number of treatment failures. These results suggest that spectral-band–specific within-network connectivity in particular regions may advance mechanistic understanding and support biomarker development for depression and TRD. The authors frame these connectivity patterns as potential electrophysiological markers that warrant further validation.
The work received ethical approval from the NIH institutional review board. Participant consent procedures were completed and archived. The authors declared no competing interests. The manuscript states that data produced in the study are available upon reasonable request to the authors. No additional external data or analytic code details were reported in the source beyond this availability statement.
The source text did not present additional methodological limitations, longitudinal data, or external validation results beyond the reported cross-sectional DIABLO analyses and AUROC values. Details such as preprocessing steps, exact MEG acquisition parameters, or full lists of connectivity features selected by DIABLO were not reported in the summarized source text and therefore are not included here.