Major depressive disorder (MDD) constitutes a debilitating psychiatric affliction, impacting approximately 7% of the global population (Mathers and Loncar, 2006). The condition is typified by persistent emotions of sadness, guilt, and worthlessness, as well as heightened suicide risk (Gotlib and Joormann, 2010). Despite the unidentified neurological substrates underpinning MDD, contemporary neuroimaging investigations have revealed functional connectivity (FC) anomalies across extensive brain networks (Otte et al., 2016; Li et al., 2018; Yan et al., 2019; Peng et al., 2020; Yang et al., 2021). These findings have reconceptualized MDD as a disorder of brain network dysfunction, offering novel perspectives for understanding its pathophysiology.

The triple-network model serves as a notable approach for investigating brain dysconnectivity in psychiatric disorders, delineating a core connectivity pattern that underlies cognitive, perceptual, affective, and social functions, encompassing the default mode network (DMN), salience network (SN), and central executive network (CEN) (Menon, 2011). In MDD patients, FC disruptions within these triple networks have been recurrently observed (Balaev et al., 2018; Cheng et al., 2018; Yan et al., 2019). Notably, the bulk of prior studies gauged FC by calculating Pearson’s correlation between time series of two given brain regions, hindering exploration of the influence of one brain region over another. In contrast, more recent studies have begun to probe the effective connectivity (EC) within MDD’s triple networks, examining the causal or directed influence of one brain region upon another. For example, research employing spectral dynamic causal modeling uncovered weakened connection strength from the SN to the CEN region in MDD patients (Kandilarova et al., 2018). Another study involving 336 MDD patients revealed both increased and reduced ECs from the SN regions (e.g., temporal pole) to other brain regions (Rolls et al., 2018). These findings collectively suggest that abnormal directed influences between triple network regions may be pivotal in MDD etiology. However, these studies primarily focused on MDD-induced EC changes within a broad frequency band (0.01–0.08 Hz or 0.01–0.15 Hz), potentially obscuring information regarding physiological fluctuations at specific frequencies.

The human brain, a biologically intricate system, features myriad oscillatory waves working in concert (Buzsáki and Draguhn, 2004; Samaha et al., 2020). Blood oxygen level-dependent (BOLD) signals at distinct frequency bands can partially reflect these neural processes and corresponding physiological functions (Zuo et al., 2010; Cole and Voytek, 2017; Hu et al., 2021). Previous MDD research has often identified frequency-specific alterations in spontaneous brain activity and connectivity. For instance, a study examined resting-state signal amplitude variability across two discrete frequency bands (slow-5: 0.01–0.027 Hz and slow-4: 0.027–0.073 Hz), revealing that the balance between the DMN and sensorimotor network favored the DMN in slow-5 and correlated with clinical depression symptom scores (Martino et al., 2016). Similarly, a study analyzing FC patterns in bipolar disorder depression across slow-5 and slow-4 found increased long-range FC density in the left lingual gyrus in slow-5 and decreased density in slow-4 (Yang et al., 2021). These results suggest that analyzing functional abnormalities of MDD at multiple frequencies is more rational than examining the routine band. To date, no prior study has explored EC at various low-frequency bands in MDD patients, and the impact of different rhythms on triple network causal processes in the disorder remains uncertain.

This study endeavors to assess EC changes of the triple networks at disparate frequency bands in MDD patients. By employing a data-driven method called complete ensemble empirical mode decomposition with adaptive noise (CEEMDAN) (Colominas et al., 2014), we initially decomposed BOLD oscillations into five distinct frequency bands. Subsequently, we quantified frequency-specific EC patterns among triple network components by integrating group independent component analysis (GICA) with multivariate Granger causality analysis (mGCA). We also conducted a correlation analysis to evaluate the association between EC changes and clinical measures in patients. Moreover, we utilized a support vector machine (SVM) to ascertain whether frequency-specific EC features of the triple networks could facilitate the differentiation of MDD patients from healthy controls (HCs). Based on prior evidence indicating disrupted triple networks in MDD, we hypothesized that (a) the EC in the triple networks would exhibit alterations in patients across various frequency bands; and (b) frequency-specific EC could serve as a biomarker for distinguishing patients from controls.

In this study, we explored the anomalies in EC of triple networks across different frequency bands in MDD by combining the CEEMDAN and mGCA methodologies. Our analysis not only revealed an enhanced mean EC within the CEN in the conventional frequency band, but also a decrease in the mean EC from the SN to the DMN in FOI-1, as well as an increase in the mean EC from the CEN to the SN in FOI-2 in MDD patients. Moreover, a significant association between the mean EC from the CEN to the SN and the HAMD scores was identified in FOI-2 for individuals with MDD. By incorporating EC features across all the three frequency bands, optimal classification performance was achieved. These results reveal frequency-specific alterations in causal influences among triple networks for patients with MDD and highlight the importance of considering multiple frequency bands when developing more precise and dependable biomarkers for disease diagnosis.

Our findings demonstrated that the EC within the triple networks is altered in a frequency-specific manner in patients with MDD. This observation aligns with previous research that has emphasized the importance of examining connectivity patterns across multiple frequency bands in order to fully understand the neurobiological underpinnings of MDD (Wang et al., 2016; Yang et al., 2021). Specifically, we observed a significant increase in the mean EC within the CEN in MDD patients compared to unaffected controls in FOI-N ranging from 0.01 to 0.08 Hz. This finding is consistent with prior studies that have reported altered FC within the CEN in MDD patients (Mulders et al., 2015; Shen et al., 2015; Zhang et al., 2016). The increased EC within the CEN may reflect a compensatory mechanism in response to the disrupted network communications, as the CEN is responsible for higher-order cognitive processes, such as working memory and executive control (Menon, 2011). Alternatively, this alteration might be indicative of a maladaptive change contributing to the cognitive deficits frequently observed in MDD (Mulders et al., 2015; Otte et al., 2016; Kandilarova et al., 2018). Moreover, we found significant decreases in the mean EC from the SN to the DMN in patients compared to controls in FOI-1 (0.12–0.18 Hz). This result aligns with prior findings of disrupted connectivity between the SN and DMN in MDD (Balaev et al., 2018; Fettes et al., 2018; Gong et al., 2019). The SN plays a crucial role in detecting and processing emotionally salient stimuli (Etkin et al., 2011; Seo et al., 2018), while the DMN is implicated in self-referential thinking and rumination (Scheibner et al., 2017). A reduced causal influence from the SN to the DMN might indicate an impaired ability to regulate internal emotional states and a propensity for excessive rumination in MDD patients (Gandelman et al., 2019). In addition, our study demonstrated significant decreases in the mean EC from the CEN to the SN in MDD patients compared to controls in FOI-2 (0.04–0.08 Hz). This finding expands upon previous research that has reported disrupted connectivity between the CEN and SN in MDD (Kaiser et al., 2015). A decreased causal influence from the CEN to the SN might suggest an impaired top-down regulation of emotional processing in MDD patients, potentially contributing to the affective symptom characteristic of the disorder (Kennis et al., 2020). Importantly, our study expands upon existing research by exploring the multi-frequency domain, revealing that EC alterations within the triple networks in MDD may be governed by specific frequency bands. Frequency-specific alterations can reflect distinct biological processes or brain features. BOLD signals at different frequencies may signify differing neuronal activities or interactions. Thus, our finding suggests that varying sensitivities to different frequency bands may exist in the causal interactions among core network structure.

Our findings also revealed that the altered mean EC from CEN to SN in FOI-2 exhibited a significant correlation with HAMD scores in MDD patients, indicating that lower directed interactions correspond to increased disease severity. This observation aligns with prior research demonstrating associations between brain dysconnectivity and depressive symptoms (Kang et al., 2018; Yang et al., 2018; Gandelman et al., 2019). A recent investigation reported correlations between abnormal amygdala connectivity and symptom severity in MDD (Ye et al., 2023), lending further credence to the clinical relevance of our results. Our previous work also indicated that diminished static and dynamic FCs were associated with greater MDD severity (Yao et al., 2019a,b). Importantly, the negative correlations observed exclusively in FOI-2 may represent frequency-specific symptoms of MDD, corroborating earlier findings that Slow-4 (0.027–0.073 Hz) (Yang et al., 2021), overlapping with FOI-2, may hold a crucial role in MDD diagnosis and progression monitoring. These results also suggest that FOI-2 might serve as a specific frequency band reflecting clinical symptoms in MDD patients. Furthermore, our investigation demonstrated that the highest classification accuracy was attained when combining EC features across all three frequency bands (FOI-N, FOI-1, and FOI-2), consistent with other studies reporting enhanced classification accuracy upon considering multiple frequency bands in neuropsychiatric disorders (Chen et al., 2016; Hu et al., 2021). Collectively, these findings underscore the importance of incorporating multiple frequency bands when examining the pathophysiology of MDD and indicate that a comprehensive, multi-frequency approach may yield more precise and reliable biomarkers for the diagnosis and differentiation of patients from controls.

The implications of our findings may extend to two aspects of future MDD therapy. First, this study illuminates the neural pathophysiology underpinning MDD and offers a fresh perspective on frequency-specific dysconnectivity patterns, potentially revealing treatment markers associated with disease severity. The frequency-specific EC alterations identified provide intricate insights into how these functional connections fluctuate across different frequencies. These findings could potentially be harnessed for precise therapeutic interventions, such as neurofeedback or transcranial magnetic stimulation, which can be used to modulate aberrant connectivity patterns in MDD patients (Drysdale et al., 2017). Second, our investigation supplies critical information in the pursuit of clinically valuable diagnostic markers for MDD. Numerous researchers have recently explored the potential of brain connectivity to differentiate MDD patients from unaffected controls (Zhong et al., 2017; Geng et al., 2018; Zhang et al., 2020). Consequently, the identified frequency-specific EC features capable of distinguishing patients from controls with notable accuracy could contribute to the development of more dependable and objective diagnostic instruments, assisting clinicians in the early detection of MDD (Guo et al., 2020). Nevertheless, given the limited sample sizes in this study, the high classification performance warrants validation in future research with larger samples.

There are some limitations that should be noted. First, the resting-state fMRI data acquisition employed a relatively lower repetition time (2 s), constraining the detection of dynamic fluctuations in higher frequency bands (>0.25 Hz). Future research would benefit from utilizing a higher sampling frequency. Second, while Granger causality analysis (GCA) is regarded as an effective method for evaluating EC in resting-state fMRI data, it has been postulated that directional changes might result from hemodynamic coupling differences among distinct brain regions (Pervaiz et al., 2020). Recently, alternative models, specifically the dynamic causal model (DCM)—a hemodynamic model (Friston et al., 2014), have been proposed to detect directed connectivity among hidden neuronal states (Park et al., 2018; Zarghami and Friston, 2020). Consequently, future studies employing DCM to explore frequency-specific reorganizations of EC in MDD patients would be of considerable interest. Third, the patient cohort in this study had prolonged exposure to various antidepressant medications. Prior research has assessed the impact of antidepressants on brain connectivity (Korgaonkar et al., 2019), and it cannot be ruled out that medication effects may have influenced our findings. Nonetheless, previous FC investigations involving high-risk MDD individuals have indicated that altered connectivity between triple networks occurs in the absence of antidepressant treatment (Pawlak et al., 2022). A future study with a never-medicated sample is required to corroborate our findings.

In conclusion, our study revealed frequency-specific alterations in the causal influences among the DMN, SN, and CEN in MDD, with potential ramifications for diagnosis and treatment. These findings enhance our comprehension of the neurobiological underpinnings of MDD and stress the significance of investigating EC patterns within the triple networks across multiple frequency bands. Future research endeavors should build upon these insights to further elucidate the role of frequency-specific EC patterns in MDD pathophysiology, examine their potential as therapeutic targets, and assess their applicability as objective biomarkers for MDD diagnosis.

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

The studies involving human participants were reviewed and approved by Gansu Provincial Hospital. The patients/participants provided their written informed consent to participate in this study.

YL designed the study and performed statistical analyses, drafted the manuscript, and approved the final manuscript as submitted. LQ and GL coordinated and carried out the data collection, revised the manuscript, and approved the final manuscript as submitted. ZZ conceptualized the study, critically reviewed the manuscript, and approved the final manuscript as submitted. All authors contributed to the article and approved the submitted version.

This work was supported by the National Natural Science Foundation of China (No. 82001918) and by the China Postdoctoral Science Foundation (No. 2020 M681865).

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.

The reviewer ZY declared a past co-authorship with the author GL to the handling editor.

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.