Dysregulated connectivity configuration of functional network model in first-episode, treatment-naive adolescents with major depressive disorder Authors: Chan Zhang, Wenjie Zhang, Jinji Bai, Xuan Deng, Jinyuan Zhao, Jiajing Cui, Junjun Ji, Ting Li, Yu Wang, Kefeng Li, Junfeng Li Published: 2026 ABSTRACT AbstractBackgroundMajor depressive disorder (MDD) is a highly prevalent psychiatric condition that frequently emerges during adolescence, a critical developmental stage characterized by heightened vulnerability to emotional dysregulation. Despite increasing evidence of large-scale brain network dysfunction in adult MDD, the static and dynamic connectivity alterations underlying adolescent MDD remain poorly understood.MethodsWe recruited 29 first-episode, treatment-naïve adolescents with MDD and 29 age- and sex-matched healthy controls (HCs). Resting-state functional magnetic resonance imaging (rs-fMRI) data were analyzed using group independent component analysis (ICA) combined with sliding-window clustering to evaluate both static functional network connectivity (sFNC) and dynamic functional network connectivity (dFNC) across the default mode network (DMN), salience network (SN), central executive network (CEN), and dorsal attention network (DAN). Correlation analyses were performed between connectivity metrics and clinical severity assessed by the 17-item Hamilton Depression Rating Scale (HAMD-17).ResultsCompared with HCs, adolescents with MDD exhibited significantly reduced intra- and inter-network connectivity within the DMN, SN, and CEN, alongside a trend toward increased DAN–SN connectivity. Dynamic analyses revealed reduced state transition frequency, shorter dwell time in low-connectivity states (e.g., DMN–CEN–SN interactions), and longer dwell time in high-connectivity states (e.g., DAN–SN coupling). Clinical analyses demonstrated that weaker intra-DMN and intra-DAN connectivity, as well as reduced DMN–CEN and DMN–SN connectivity, were negatively correlated with HAMD-17 scores. Conversely, prolonged dwell time in hyperconnected states positively correlated with greater symptom severity.ConclusionOur findings highlight distinct static and dynamic network abnormalities in adolescent MDD, including disrupted DMN–CEN competitive balance and maladaptive DAN–SN hyperconnectivity. These alterations may hint at developmental-stage-specific neuropathological mechanisms that could differ from adult depression, although direct comparison with adult cohorts is lacking. Integrating static and dynamic FNC analyses may provide preliminary insights into candidate neuroimaging markers for early detection and intervention strategies in adolescent MDD, though these findings require validation in independent, larger cohorts.Supplementary InformationThe online version contains supplementary material available at https://doi.org/10.1186/s12888-026-08332-2.Keywords: Major depressive disorder, Adolescence, Resting-state fMRI, Functional network connectivity, Dynamic functional connectivity Introduction Major depressive disorder (MDD) is a prevalent psychiatric disorder characterized by persistent sadness, irritability, diminished concentration, and impaired social functioning, affecting more than 350 million individuals worldwide [1]. In China, MDD has become the second leading cause of disability, underscoring its substantial public health burden [2]. Adolescence represents a critical developmental window for emotional regulation and cognitive maturation, and is also a peak period for the onset of MDD [3–5]. Compared with adults, adolescents with MDD exhibit higher recurrence rates and distinctive clinical manifestations, including pronounced mood fluctuations and heightened sensitivity to social evaluation [6]. Furthermore, adolescent MDD is associated with increased risks of substance abuse, cognitive impairment, and academic difficulties [7], highlighting the urgent need to clarify its underlying neurobiological mechanisms. Currently, the diagnosis of adolescent depression primarily relies on depression scale screenings, which is a symptom-based approach. According to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-V), a diagnosis of depression can be made if a patient exhibits five or more of the following symptoms persistently over a two-week period: depressed mood, diminished interest or pleasure, significant weight or appetite changes, insomnia or hypersomnia, psychomotor retardation or agitation, fatigue, feelings of worthlessness or guilt, diminished concentration, or recurrent suicidal ideation or attempts [8]. However, these subjective approaches are vulnerable to inter-rater variability and limited diagnostic precision. Accordingly, there is a pressing need to identify objective biomarkers that can improve diagnostic accuracy and guide personalized treatment strategies. Resting-state functional magnetic resonance imaging (rs-fMRI) provides a non-invasive, radiation-free approach to explore large-scale brain network interactions [9]. This neuroimaging approach, which reflects the relationship between brain circuitry and human behavior, holds potential as a source of biological markers for psychiatric disorders [10]. Blood oxygen level-dependent fMRI (BOLD-fMRI) detects spontaneous low-frequency fluctuations in neural activity, which can be used to infer functional connectivity (FC) across distributed brain regions [11]. Functional connectivity has been demonstrated to reliably map reproducible resting-state brain networks at both individual and group levels [12]. The human brain, estimated to contain 100 to 1,000 trillion synapses, exhibits resting-state networks (RSNs) that play critical roles in brain function and disorders such as depression. Modern network theory highlights the suitability of studying this complex neural system from a network perspective, given its intricate organization [13]. Previous studies have consistently identified four core resting-state networks relevant to MDD pathophysiology: the default mode network (DMN), salience network (SN), central executive network (CEN), and dorsal attention network (DAN) [14]. However, the developmental specificity and dynamic characteristics of these functional network abnormalities in adolescents—a unique population—remain poorly understood. The Default Mode Network (DMN), dominant during rest and associated with self-referential thinking and episodic memory, shows hyperactivation in depression, strongly correlating with rumination and fixation on negative emotions [15]. Key DMN regions include the medial prefrontal cortex (mPFC), posterior cingulate cortex (PCC), precuneus, and angular gyrus [16]. The Salience Network (SN), composed of the anterior insula (aINS) and dorsal anterior cingulate cortex (dACC), acts as a “switch” for internal and external stimuli, dynamically allocating attentional resources between the DMN (self-focused processing) and the CEN/DAN (task-oriented processing) [17, 18]. Meanwhile, the CEN and DAN support higher-order cognitive control and spatial attention maintenance, respectively. Their weakened functionality may contribute to impaired executive function and attentional deficits [19]. Studies in adults with depression often report a “competitive imbalance” between the DMN and CEN, where DMN hyperactivation suppresses CEN-mediated cognitive control [20]. However, adolescence is characterized by ongoing neural maturation, particularly within prefrontal–limbic circuits, which may yield developmental-stage–specific network abnormalities [21]. Most prior studies have focused on static FNC (sFNC), reflecting averaged connectivity patterns across the entire scanning session. There remains a lack of systematic exploration into how dynamic functional network connectivity (dFNC) —such as state transition frequency (NT) and mean dwell time (MDT) of specific network states—influences depressive symptoms in adolescents. Dynamic functional network connectivity (dFNC) [22], typically measured using sliding-window and clustering methods, captures temporal variations in network states and provides complementary insights into large-scale brain interactions. This approach enables the study of functional interactions between brain networks and their underlying architecture without restricting analysis to predefined regional connections. Research suggests that functional connectivity (FC) dynamics may become more pronounced during unrestricted mental processes in the resting state [23]. Some researchers propose that individuals freely engage in spontaneous mental activities during conscious rest [24]. From this perspective, brain functional connectivity can vary over short time scales rather than remaining static. Alterations in dynamic FC have been implicated in multiple psychiatric conditions, including Alzheimer’s disease [25], schizophrenia [26], and depression [27]. Investigating whole-brain dynamic network connectivity patterns could offer novel insights into abnormal brain communication in MDD. Static and dynamic FC processes are complementary, and both are indispensable for understanding cognitive function and its development. However, systematic investigations of dFC abnormalities in adolescents with MDD remain limited, leaving critical questions unanswered. For example, do adolescents with MDD exhibit unique temporal patterns of network dysfunction distinct from adults? How do these dynamic features relate to symptom severity? Therefore, this study aimed to investigate both static and dynamic functional network connectivity (FNC) in first-episode, treatment-naïve adolescents with MDD. We hypothesized that [1] adolescents with MDD would exhibit disrupted intra- and inter-network connectivity within the DMN, SN, CEN, and DAN; [2] dynamic functional connectivity would show altered state transitions and mean dwell times; and [3] these abnormalities would correlate with depressive symptom severity, raising the possibility of developmental-stage–specific neural mechanisms that might differ from adult MDD, which requires verification in future studies with adult comparison groups. Methods Participants We recruited 29 treatment-naïve adolescents (aged 10–19 years) with first-episode MDD from the Changzhi Mental Health Center between January 2020 and December 2022. 29 age- and sex-matched healthy controls (HCs) were enrolled through community advertisements and online platforms (e.g., WeChat). The study was approved by the Ethics Committee of Changzhi Medical College (Approval No. ChiCTR2000038210) according to the standards of the Declaration of Helsinki. Written informed consent was obtained from the parents or legal guardians of all participants, and written informed assent was provided by the adolescents themselves. Inclusion and exclusion criteria Inclusion criteria for the MDD group were: [1] diagnosis of MDD based on the Chinese Classification of Mental Disorders, Version 3 (CCMD-3) and DSM-5 criteria; [2] HAMD-17 score ≥ 7; [3] first depressive episode and no prior treatment; [4] right-handedness; [5] no history of psychiatric disorders in first-degree relatives; [6] no neurological disorders, substance/alcohol abuse, or MRI contraindications. Exclusion criteria included: [1] comorbid psychiatric disorders (e.g., bipolar disorder, epilepsy); [2] history of significant head trauma or brain surgery; [3] pregnancy; and [4] left-handedness. The clinical data collection Demographic and clinical data were systematically gathered by trained psychiatrists, encompassing variables such as age, sex, anthropometric measures (height and weight), handedness, educational attainment, prior medical records, personal and family psychiatric history, occupation, and ethnicity. Diagnosis of MDD was established independently by two senior clinical psychiatrists based on DSM-V criteria. To further ensure diagnostic accuracy and rule out other psychiatric conditions, all adolescent participants and their parents underwent the semi-structured Kiddie Schedule for Affective Disorders and Schizophrenia for School-Age Children-Present and Lifetime Version (K-SADS-PL), which integrates reports from both the child and a parent to evaluate lifetime and current mental disorders. Depression severity was quantified using the 17-item Hamilton Depression Rating Scale (HAMD-17). MRI data acquisition All imaging was performed on a 3.0 T MAGNETOM Skyra scanner (Siemens Healthcare, Erlangen, Germany) equipped with a 32-channel head coil at the Peace Hospital affiliated to Changzhi Medical College. For both the adolescent MDD group and the HC group, scans were acquired using T1-weighted imaging (T1WI), T2-FLAIR, echo planar imaging (EPI) sequence and 3D-T1 sequence. First, using T1WI, T2-FLAIR excluding organic lesions. Functional MRI data was acquired using an echo-planar imaging sequence with the following parameters: repetition time/echo time = 2000 ms / 30 ms. flip angle = 90°, field of view = 224 × 224 mm2, matrix size = 64 × 64, isotropic voxel size = 3.5 × 3.5 × 3.5 mm, phase encoding direction = A ≫ P, 240 slices at 3.5 mm thickness covering the whole brain. For each participant, with a scanning duration of 8 min and 6 s. Structural MRI data was acquired through a 3D T1-weighted gradient echo sequence with repetition time/echo time = 25 ms / 2.98 ms; phase encoding direction = A ≫ P, field of view = 256 × 256mm2, flip angle = 90°, 192 sagittal slices covering the whole brain with isotropic voxel size of 1 mm × 1 mm × 1 mm and an acquisition time of 6 min and 3 s. Participants lay flat on the MRI scanning bed with their eyes closed and their head fixed with a foam head pad. Noise-canceling earplugs were given to the participants to reduce the MRI noise, and they were instructed to avoid thinking as much as possible. Data preprocessing Data preprocessing was performed using GRETNA 2.0 (Graph Theoretical Network Analysis). For each subject, the conversion of the original DICOM data file into NIFTI format. The first 10 volumes of the rs-fMRI dataset were discarded to allow for MR signal equilibrium; then we performed slice-timing correction before using rigid body motion correction to adjust each subject’s head motion. All of the subjects in this study satisfied our criteria for head motion with displacement < 2.0 mm in any plane and rotation < 1.5° in any direction. All subjects met the head motion criteria, and no participants were excluded due to excessive motion, resulting in 0% motion-excluded scans. After that, we resampled to 3 × 3 × 3 mm3 using an echo planar imaging template in the standard Montreal Neurological Institute (MNI) space. Finally, we smoothed the fMRI images using a Gaussian kernel having a full-width at half-maximum (FWHM) of 8 mm; linear detrending, and band-pass filter (0.01–0.08 HZ). Nuisance covariates including global mean signals, white matter, and cerebrospinal fluid (CSF) signals were regressed out from the blood oxygen level-dependent (BOLD) signals. Independent component analysis and State function connectivity networks analysis Functional network decomposition was carried out using group-level spatial independent component analysis (ICA) implemented in the GIFT software (v3.0b; http://mialab.mrn.org/software/gift/). To reduce computational load, a two-stage principal component analysis was applied: first, each subject’s preprocessed fMRI data underwent temporal dimension reduction; second, the concatenated data from all subjects were further reduced in dimension. Subsequently, ICA with the Infomax algorithm was applied to the group-level data to extract 46 independent components (ICs), yielding both spatial maps and corresponding time courses for each IC. The stability and reliability of the IC decomposition were ensured by repeating the ICA 100 times via the ICASSO procedure. Subject-specific ICs were then obtained through a back-reconstruction approach and converted to z-score maps, which formed the basis for voxel-wise one-sample t-tests at the group level. The resulting group-level t-maps enabled identification of brain regions significantly associated with each IC. Then we identified ICs of interest by the visual screening based on previously reported template-matching. First, the spatial and temporal profiles of each component were visually examined and labelled as DMN, CEN, DAN and SN. Next, these labels were further determined by calculating spatial correlations between each component and the reference templates [28, 29]. For each subject, we extracted the time courses of all ICs belonging to a given network. To obtain a single network-level representative time course, we applied principal component analysis (PCA) to the matrix of IC time courses (e.g., 4 ICs for DMN) and retained the first principal component, which explained > 60% of the variance in all networks. Pairwise Pearson’s correlations were then computed between the four network-level time courses (DMN, CEN, DAN, SN), resulting in six unique connectivity pairs for static FNC analysis. The same network-level time courses were used for dynamic FNC analysis with sliding-window correlation. These correlation coefficients were then entered into a multivariate analysis of covariance (MANCOVAN) within the GIFT toolbox to test for group differences in static functional network connectivity (sFNC). Dynamic functional network connectivity analysis Sliding window analysis We estimated the dynamic FNC by computing Pearson’s correlations between the network-level time courses (derived via PCA as described in Sect. 2.4) using the sliding window method. We set a window size of 20 TRs (40 s) with a step size of 1 TR (2 s) for each participant in accordance with previous studies. The window length of 20 TRs was chosen because previous studies [24, 30] have shown that a window length of 20–30 TRs may provide a good trade off between the quality of the FNC estimate and the temporal resolution. After excluding the first 10 volumes, 230 TRs remained per participant, yielding 211 windows per participant (230 − 20 + 1 = 211). These time windows were convolved with a Gaussian kernel (σ = 3 TRs). Due to edge effects caused by the kernel, the first and last several windows were excluded, resulting in 210 effective windows per participant. Across 58 participants, this yielded a total of 12,180 windows (58 × 210) for clustering. Fisher’s z-transformation was applied to stabilize variance. We regressed out age, sex and mean FD values as covariates. All clustering and visualizations were performed on Fisher’s z-transformed correlation coefficients. Thus, all dFNC analyses (clustering, state identification, and temporal metrics) were performed on the four network-level time courses (DMN, CEN, DAN, SN), not on individual IC time courses. Clustering analysis To identify recurring patterns of dynamic functional connectivity, we employed a k-means clustering algorithm (squared Euclidean distance) on all sliding-window connectivity matrices. The analysis was repeated 500 times with 150 replicates to ensure robust clustering. The optimal number of clusters was determined using the elbow criterion, which indicated four distinct connectivity states, This 4-state solution was used as the primary clustering result for all subsequent dynamic analyses reported in the main manuscript (Figs. 3, 4 and 5; Table 2). Validation analyses using 5 and 6 clusters were also performed and are presented in the Supplementary Materials. Dynamic functional connectivity states among the four large-scale networks (DMN, CEN, SN, DAN). Each matrix displays the centroid (average Fisher’s z transformed correlation) of all windows in that state. Positive values (warm colors) indicate positive functional connectivity; negative values (cool colors) indicate anticorrelation. Numbers below each matrix indicate the total count and percentage of sliding windows assigned to that state Group differences in dynamic functional connectivity within State 3 (MDD vs HC). The matrix displays the mean Fisher’s z transformed correlation coefficients for each network pair (DMN CEN, DMN SN, CEN SN, etc.) in State 3. Warm colors indicate positive connectivity; cool colors indicate negative connectivity. Asterisks indicate significant group differences (pFDR < 0.05, Mann Whitney U test) Temporal properties of dynamic states (MDD vs HC). (A) The fraction of time the occurrence of FC state 2, state 3 and state 4 has significant between group difference (pFDR < 0.05). (B) The mean dwell time of FC state 2, state 3 and state 4 has significant between group difference (pFDR < 0.05). (C) The number of transitions between states has significant difference (pFDR < 0.05) between group. (D) cluster number. FT1, FT2, FT3, FT4 indicates the fraction of time in state1, state2, state3, state4, respectively. MDT1, MDT2, MDT3, MDT4 indicates the mean dwell time in state1, state2, state3, state4, respectively Table 2Comparison in temporal properties between the two groupsz valuep valueFraction timeState 1-10.32State 2-3.140.002State 3-2.860.004State 4-4.56<0.001*Mean dwell timeState 1-10.32State 2-2.440.02State 3-2.730.01State 4-4.3<0.001*Number of transitions-3.020.003When windows TR = 20, clusters = 4, Dynamic connectivity feature analysis for the HC and MDD groups. The test method is Mann-Whitney U test. pFDR <0.05 was considered statistically significant. *Significant pFDR<0.001 Comparison in temporal properties between the two groups When windows TR = 20, clusters = 4, Dynamic connectivity feature analysis for the HC and MDD groups. The test method is Mann-Whitney U test. pFDR <0.05 was considered statistically significant. *Significant pFDR<0.001 Temporal properties and connectivity strength of dynamics states To characterize the temporal dynamics of functional connectivity, we computed three metrics for each state: fractional time (the proportion of total windows assigned to that state), mean dwell time (the average duration that a state was maintained before switching to another), and number of transitions (total switches between states across the scanning session). Age and sex were included as covariates in subsequent group comparisons. Validation analysis Two additional window lengths (22 and 25 TRs) were applied to test the robustness of the sliding window analysis. Moreover, the numbers of clusters were set at 5 and 6 also used to verify the stability of our results. Statistical analyses Demographic and clinical characteristics were analyzed using SPSS 26.0 software (IBM Corp., Armonk, NY, USA). Continuous variables including age, BMI, and HAMD-17 scores were compared between groups using two-sample independent t-tests and reported as mean ± SD. Categorical data were compared using the chi-square test. To compare static FNC between groups across all network pairs, we first performed multivariate analysis of covariance (MANCOVAN) within the GIFT toolbox, with age, sex, and mean FD as covariates. Given that the connectivity values deviated from normality (Shapiro–Wilk test, pFDR < 0.05), post-hoc pairwise comparisons for each individual FNC edge were conducted using Mann–Whitney U tests. The resulting test statistics are reported as z values. Dynamic functional network connectivity (dFNC): For each dynamic metric, group comparisons between MDD and HC were performed using Mann–Whitney U tests (non-parametric) due to non-normal distributions. These results are also reported as z values. Pearson’s correlation was conducted to assess the relationship between the strength of FNC, dynamic FNC measures and HAMD-17 scores, controlling for age, sex, and mean FD values were treated as covariates. To evaluate the statistical power of our main findings, post-hoc power analyses were conducted using G*Power 3.1. Based on the observed effect sizes (Cohen’s d ranging from 0.40 to 0.50 for key static and dynamic functional connectivity comparisons), with an alpha level of 0.05 (two-tailed) and a final sample size of 29 per group, the achieved power ranged from 0.79 to 0.81. These results indicate that our sample size was adequately powered to detect moderate-to-large effect sizes. Results Demographics and clinical characteristics The details of demographic and clinical characteristics are shown in Table 1. We did not observe any significant differences in age, education level, gender between MDD patients and HCs. HAMD-17 scores were significantly higher in the adolescent MDD group compared to the HC group (p < 0.001). Although BMI differed significantly between groups (p = 0.02), this difference was driven by a single healthy control participant with an outlier BMI value (31.8). After excluding this participant, the BMI difference remained significant (p = 0.05). However, all main functional connectivity findings were unchanged, confirming that the original results were not confounded by BMI (Supplementary Materials for sensitivity analysis). Therefore, BMI was not included as a covariate in the primary imaging analyses. Table 1Demographics and clinical characteristics of all participantsVariablesMDD (n=29)HC (n = 29)t/c2valuep valueAge(years)15.07±1.4915.79±2.231.460.15Sex(male/female)May-24Nov-18-3.110.08Education(years)9.07±1.499.79±2.231.460.15BMI19.20±2.5921.53±4.262.520.02HAMD-17(score)18.54±6.001.03±1.52-14.21<0.001MDD, Major depressive disorder; HC, Healthy controls, BMI, Body mass index; HAMD-17, the 17-items Hamilton Depression Scale; The data are presented as the mean ± standard deviation (SD) or percentages. Group differences were assessed using either chi-square analysis or independent t-tests. A p value < 0.05 was considered to indicate statistical significance. Demographics and clinical characteristics of all participants MDD, Major depressive disorder; HC, Healthy controls, BMI, Body mass index; HAMD-17, the 17-items Hamilton Depression Scale; The data are presented as the mean ± standard deviation (SD) or percentages. Group differences were assessed using either chi-square analysis or independent t-tests. A p value < 0.05 was considered to indicate statistical significance. Among the 29 MDD patients, 23 (79.3%) had mild to moderate depression (HAMD-17 score 7–24). only 6 patients had severe depression (HAMD-17 ≥ 25). Due to the limited sample size within each subgroup, further stratified analyses by depression severity were not conducted. ICs of interest Of the 46 ICs identified by the group ICA, 28 ICs were identified as noise components and then discarded. The 18 ICs of interest, DMN, CEN, DAN and SN, which were identified and categorized into four networks using group ICA (Fig. 1). Table S1 lists the ICs’ labels and peak activation coordinates. Spatial map of 4 functional networks. Spatial maps of the 18 selected independent components have been identified and grouped into four functional networks: default-mode network (DMN), control executive network (CEN), salience network (SN), Dorsal attention network (DAN). FNC differences between MDD and HC Figure 1 illustrates the ICs that matched to the four brain networks. IC14, IC27, IC38, IC45 belong to DMN; IC5, IC10, IC15, IC19, IC21, IC36, IC34, IC41 belong to CEN; IC3, IC9, IC30, IC39 belong to DAN; IC33, IC44 belong to SN. (Table S2 and Fig. 2) illustrate differences in connectivity between MDD and HC. Compared with HCs, The MDD group exhibited decreased connectivity in the following pairs: intra-DMN; DMN_CEN; DMN_SN; CEN_SN; intra-SN_SN( pFDR < 0.05 ). On the other hand, MDD exhibited increased connectivity compared to HC in the following: DAN_SN (z = 1.925, p = 0.048, pFDR = 0.054). Therefore, it is reported as a trend rather than a significant finding. (A) Group differences in the selected FNC. Red lines indicate significantly increased connectivity, and blue lines indicate significantly decreased connectivity (MDD vs HC, p < 0.05). (B) The comparison of FNC between the MDD group and the HC group ( p < 0.05). Dynamic FNC states and properties Using the 4-state clustering solution (determined by the elbow criterion), we identified four recurring dynamic functional connectivity states. Figure 3 shows the four recurring dynamic functional connectivity states identified by k-means clustering. Each submatrix in Fig. 3 represents a state centroid, i.e., the average Fisher’s z-transformed connectivity pattern across all sliding windows assigned to that state, capturing transient configurations of inter-network interactions. The numbers below each matrix indicate the total count and percentage of windows assigned to that state. Figure 4 further focuses on State 3—one of the four identified states—and displays the group differences (MDD vs. HC) in functional connectivity strength specifically within this state. The matrix in Fig. 4 compares the mean Fisher’s z-transformed correlation coefficients for each network pair (e.g., DMN-CEN, DMN-SN, CEN-SN) between the two groups. Thus, while Fig. 3 visualizes the characteristic connectivity patterns of each dynamic state, Fig. 4 illustrates the between-group comparisons within a single state of interest. The proportions of the four states are 1% (178), 31% (3786), 8% (996) and 59% (7220) respectively. In state 3, compared with HCs, the MDD group showed increased connectivity in DMN-SN and DMN-CEN; at the same time, the CEN with DAN and SN had decreased functional connectivity; SN had decreased connectivity with DAN in MDD group (Fig. 4). In State 1,2,4, the FNC within the four networks was very sparse. Moreover, we found positive connections intra SN; negative connections intra DAN (Fig. 3). For dynamic FNC properties (Mann-Whitney U tests, pFDR < 0.05 ), Compared with the HC group, we found significant decreased fraction time ( FT ), mean dwell time ( MDT ) in state2, state3 and number of transitions ( NT ) in MDD patients. However, we found increased FT, MDT in state4 (Table 2; Fig. 5). Correlation analysis FNC difference index and HAMD-17 scores There is a negative correlation between the FNC strength of intra/inter-network and the HAMD-17 score (Fig. 6). The strength of intra-DMN with HAMD-17 score had negative correlation (r = − 0.296, p = 0.024). The strength of intra-DAN with HAMD-17 score had negative correlation (r = − 0.270, p = 0.040). At the same time, The strength of DMN_CEN ( r =-0.386, p = 0.003),DMN_SN( r =-0.503, p <0.001 ),CEN_SN ( r =-0.259, p = 0.049)with HAMD-17 score had negative correlation. Correlations between connectivity metrics and depressive severity (HAMD-17) in the MDD group. (A) The correlations between the strength of functional connectivity (FC) between networks, and the HAMD-17 scores. (B) The correlations between the dynamic FNC difference index and the HAMD-17 scores. Pearson’s correlation was employed, pFDR < 0.05 was considered significant. with the correlation coefficient r values denoted using the color bar. * indicates p ≤ 0.05, ** indicates p ≤ 0.01, *** indicates p ≤ 0.001, **** indicates p ≤ 0.0001. “_” indicates the strength of functional connectivity (FC). Abbreviations: HAMD-17: Hamilton Depression Rating Scale Dynamic FNC properties and HAMD-17 scores The mean dwell time was negatively correlated with HAMD-17 score in state3 ( r = − 0.276, p = 0.036); positive correlated with HAMD-17 score in state4( r = 0.292, p = 0.026) The fraction time was negatively correlated with HAMD-17 score in state2 ( r = − 0.301, p = 0.022), state3 ( r = − 0.357, p = 0.006); positive correlated with HAMD-17 score in state4 ( r = 0.479, p< 0.001). The number of transitions was negatively correlated with HAMD-17 score (r = − 0.318, p = 0.015) (Table 3; Fig. 1 and Figure S6). Table 3Dynamic state features correlated with HAMD-17 scores (MDD group)r valuep valueFraction time of State 2-0.30.02Fraction time of State 3-0.360.01Fraction time of State 40.48<0.001*Mean dwell time State 3-0.280.04Mean dwell time State 40.30.03Number of transitions-0.320.02When using the 4‑state clustering solution (TR = 20), the correlations between the dynamic FNC difference index and the HAMD-17 scores. pFDR <0.05 was considered statistically significant. *Significant pFDR<0.001 Dynamic state features correlated with HAMD-17 scores (MDD group) When using the 4‑state clustering solution (TR = 20), the correlations between the dynamic FNC difference index and the HAMD-17 scores. pFDR <0.05 was considered statistically significant. *Significant pFDR<0.001 Post-hoc power analysis revealed that for the primary between-group comparisons (MDD vs. HC), with an observed effect size of Cohen’s d = 0.45 (averaged from significant sFNC/dFNC findings), α = 0.05 (two-tailed), and final sample sizes of 29 per group, the achieved power was 0.81. For the correlation analyses between connectivity metrics and HAMD-17 scores (r ≈ 0.40), the power was 0.79. These findings suggest that our study was sufficiently powered to detect the observed moderate-to-large effects. Validation analysis In this study, we used two different sliding window lengths to verify our main results of dynamic FNC. The results of the sliding window lengths of 22 and 25 TRs were similar to the results of the 20 TR we found. To test the robustness of our primary 4-state results, we also performed clustering with k = 5 and k = 6. As shown in Supplementary Figures S2–S5 and Tables S3–S5, the overall patterns of group differences remained consistent with the 4-state findings, though not all metrics reached significance, as expected given the influence of cluster number on temporal measures. Discussion In this study of first-episode, treatment-naïve adolescents with major depressive disorder (MDD), we identified distinct patterns of both static and dynamic functional network dysconnectivity. Statically, adolescents with MDD showed reduced connectivity within and between the default mode network (DMN), central executive network (CEN), and salience network (SN), alongside a trend toward increased between the dorsal attention network (DAN) and SN. Dynamically, they exhibited reduced state flexibility and spent more time in a hyperconnected state that correlated with symptom severity. Crucially, these static and dynamic alterations were significantly correlated with depressive symptoms, highlighting their clinical relevance and offering insights into the neurobiology of early-onset depression. Disrupted static network architecture in adolescent MDD The weakened static connectivity within the DMN and between the DMN and SN likely reflects a core deficit in regulating self-referential thought. In healthy states, the SN helps disengage the DMN’s introspective processing; its failure to do so in MDD may trap adolescents in ruminative, negative self-evaluative loops, which aligns with clinical presentations of irritability and social withdrawal [15, 31].Furthermore, the reduced DMN-CEN connectivity suggests a disruption of the typical antagonistic balance required to shift between internal thought and external, task-focused cognition [32]. This breakdown may directly contribute to the attentional and executive function deficits, such as academic difficulties, commonly observed in depressed adolescents [33].Conversely, the observed DAN-SN a trend toward increased points toward a state of maladaptive hypervigilance. Although this finding did not survive multiple-comparison correction and should be interpreted cautiously, it may suggest the SN persistently biases the dorsal attention system toward emotionally salient, often negative, stimuli, creating a cycle that reinforces attention to and memory of adverse information [34]. Altered network dynamics suggest cognitive and emotional inflexibility Our dynamic analysis complements these static findings by revealing a loss of network flexibility. The significantly lower number of transitions between brain states in the MDD group suggests a “neurodynamic rigidity,” where the brain is less able to fluidly shift between different information processing configurations in response to changing demands [35]. This inflexibility may be a neural correlate of the cognitive rigidity and emotional perseveration characteristic of depression. This rigidity was further characterized by altered temporal properties. Patients spent less time in sparsely connected states (e.g., State 3), which correlated with milder symptoms and may represent an adaptive or restful mode. Conversely, they spent significantly more time in a hyperconnected state (State 4), and this prolonged dwelling was positively correlated with greater symptom severity. This suggests that adolescent MDD involves getting “stuck” in a maladaptive, resource-intensive network configuration that reinforces pathological mood. Clinical correlations and developmental specificity The clinical relevance of these network disruptions is underscored by their direct correlations with HAMD-17 scores. Weaker static connectivity within the DMN and between the DMN, CEN, and SN was associated with more severe depression, supporting the Triple Network Model’s premise that impaired coordination between self-referential, executive, and salience-detecting networks is central to MDD pathophysiology [14]. The link between prolonged dwell time in a hyperconnected dynamic state and higher symptom severity further solidifies the connection between neurodynamic rigidity and the clinical expression of depression. It is tempting to interpret these findings through the developmental lens of adolescence, as this period is characterized by ongoing maturation of prefrontal-limbic circuits [36, 37]. However, given the lack of an adult comparison group in our study, such interpretations remain speculative. The reduced DMN-CEN connectivity we observed may not just be a symptom of depression, but a sign of a disrupted or delayed developmental trajectory, as this anticorrelation typically strengthens with age [38]. If this speculation holds true, then such neural immaturity might amplify the impact of adolescent-specific stressors, such as social evaluation, potentially contributing to the SN hypersensitivity and maladaptive DAN-SN hyperconnectivity seen in our cohort [20]. Nevertheless, direct evidence comparing adolescents and adults with MDD is needed to confirm this developmental hypothesis. Therefore, the dysfunctions identified here may represent a critical developmental derailment, offering a unique window for early intervention before these pathological network patterns become consolidated in adulthood. It is also important to note that our study cohort consists exclusively of Chinese adolescents of Han ethnicity; therefore, the extent to which these network dysfunctions generalize to other ethnic or cultural populations remains to be determined. Our identification of neurodynamic rigidity (reduced state transitions and prolonged dwell time in hyperconnected states) has direct clinical implications. It is tempting to speculate that neuromodulation techniques such as repetitive transcranial magnetic stimulation (rTMS) or transcranial direct current stimulation (tDCS) might eventually be applied to restore network flexibility, for example, by targeting salience network nodes to facilitate state switching [39]. However, such applications remain highly speculative at this stage and require empirical testing. Moreover, dynamic connectivity metrics (e.g., dwell time in State 4) warrant further investigation as potential treatment response markers for adolescent MDD, but this remains to be tested in prospective studies [40]. Limitations Our study has several limitations. First, the relatively small sample size may limit the generalizability of our findings and precluded more complex subgroup analyses. Future studies with larger cohorts are needed to validate these results. Due to time constraints during clinical enrollment, we did not collect Tanner staging. Future studies will include self- or clinician-rated Tanner scales. Our cohort consists exclusively of Chinese adolescents of Han ethnicity, which may limit the cross-cultural generalizability of our findings. Cultural factors, including differences in emotional expression, social stressors, and help-seeking behaviors, could influence both the clinical presentation of depression and the underlying neural correlates. Moreover, genetic and environmental backgrounds may differ across populations. Therefore, our results should be interpreted as specific to Chinese adolescent MDD patients. Future studies with multi-ethnic, cross-cultural cohorts are needed to validate and extend these findings. Second, the cross-sectional nature of this study precludes any causal inferences regarding the relationship between network rigidity (e.g., reduced state transitions, prolonged dwell time in hyperconnected states) and the onset of depressive symptoms. Longitudinal studies are essential to establish causality and to track these network changes over time. Third, our analysis was restricted to four major networks; finer-grained analyses of subnetworks may reveal more specific pathophysiological mechanisms. Finally, this study relied exclusively on neuroimaging; future work should integrate multi-omics approaches to provide a more comprehensive understanding of the pathogenesis of adolescent MDD. Conclusion In conclusion, by integrating static and dynamic fMRI analyses, this study demonstrates that first-episode adolescent MDD is characterized by both stable network imbalances and a loss of dynamic flexibility. The disrupted interplay between the DMN, CEN, and SN, coupled with maladaptive DAN-SN hyperconnectivity, is directly linked to symptom severity and could reflect a pathological deviation in the brain’s neurodevelopmental trajectory, although longitudinal and cross-developmental studies are required to test this possibility. These findings suggest that integrated neuroimaging metrics could be further explored as candidate biomarkers for early detection in future hypothesis-driven studies. They also raise the possibility that interventions aimed at restoring network balance and flexibility during the critical neuroplastic window of adolescence might eventually offer therapeutic promise, though this remains to be established. Supplementary Information Below is the link to the electronic supplementary material. Supplementary Material 1 Acknowledgements None. Abbreviations Major depressive disorder Static functional network connectivity Dynamic functional network connectivity Default mode network Control Executive network Salience network Dorsal attention network 17-item Hamilton Depression Rating Scale The Diagnostic and Statistical Manual of Mental Disorders Functional connectivity Independent component analysis Author contributions All authors provided have made strong, direct, effective contributions to this research, and agree to publish it. Conceptualization: K. Li, and J. Li ; Literature search:C. Zhang, W. Zhang, J. Bai, X. Deng, J. Ji; IRB approvals: J. Ji, T. Li, Y. Wang, and J. Li; Participant enrollment: C. Zhang, W. Zhang, J. Zhao, J. Cui, J. Bai, X. Deng, J. Ji, T. Li, and Y. Wang; Data analysis and visualization: C. Zhang, W. Zhang, J. Cui, J. Ji, T. Li, and Y. Wang; Project supervision: J. Li. Funding: J. Li, and K. Li; Manuscript-first draft: C. Zhang, W. Zhang, Manuscript-review & editing: All the authors. All the authors have approved the final manuscript. Funding This study was supported by the Natural Science Foundation of Shanxi Province (201901D221113), the Four “Batches” Innovation Project of Invigorating Medical through Science and Technology of Shanxi Province (2023XM016), and the Macao Polytechnic University fund (RP/FCA-14/2023). Data availability The data that supports the findings of this study are available from the corresponding authors upon reasonable request. Declarations Ethics approval and consent to participate The study was approved by the Ethics Committee of Changzhi Medical College (Approval No. ChiCTR2000038210) according to the standards of the Declaration of Helsinki. Written informed consent was obtained from the parents or legal guardians of all participants, and written informed assent was provided by the adolescents themselves. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Footnotes Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. Chan Zhang and Wenjie Zhang contributed equally co-first authors. Contributor Information Kefeng Li, Email: kefengl@mpu.edu.mo. Junfeng Li, Email: lijunfengcz@163.com. References Associated Data Supplementary Materials Supplementary Material 1 Data Availability Statement The data that supports the findings of this study are available from the corresponding authors upon reasonable request.