Dynamic functional connectivity and coupling analysis of triple networks and white matter functional networks in first-episode schizophrenia patients: mechanisms revealed by follow-up studies

Abstract

Background

Patients with schizophrenia exhibit widespread disruptions in large-scale brain network communication, particularly within the default mode network (DMN), central executive network (CEN), and salience network (SN) — collectively termed the triple networks. While existing research has established the association between functional abnormalities in these networks and the pathophysiological mechanisms of schizophrenia, investigations into white matter’s functional role remain limited. Specifically, the involvement of white matter in dynamic interactions among the triple networks remains unclear. Dynamic functional connectivity (DFC), capable of capturing time-varying characteristics of brain activity, may offer novel insights into both triple network dysfunction and white matter abnormalities in schizophrenia.

Methods

This study enrolled 93 schizophrenia patients and 92 healthy controls. To assess white matter function, we extracted 48 white matter networks based on the Johns Hopkins University (JHU) white matter atlas and employed sliding window analysis to examine DFC patterns and global coupling properties within both triple networks and white matter networks. Additionally, a longitudinal observational follow-up study (approximately 5 months) was conducted with 39 patients to evaluate treatment-related changes in DFC and coupling dynamics.

Results

Compared with healthy controls, schizophrenia patients exhibited significant alterations in both intra-network DFC and global coupling across the triple networks and white matter networks. Longitudinal observations revealed DFC changes and temporal characteristics, particularly within white matter networks. Patients exhibited higher baseline scores for fractional window and mean dwell time than healthy individuals, which decreased during treatment follow-up. Additionally, the PANSS scores of the patients were significantly lower compared to before treatment. Brain regions showing significant global coupling changes included anterior/posterior DMN, corpus callosum, and the left crus of cerebellum.

Conclusion

Our findings highlight the pathophysiological significance of functional and coupling abnormalities between triple networks and white matter networks in schizophrenia. These results provide novel insights into dynamic alterations within brain networks of schizophrenia patients and may suggest potential neuroimaging biomarkers for future therapeutic strategies.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12888-025-07455-2.

Introduction

Schizophrenia is a complex psychiatric disorder characterized by diverse clinical manifestations including hallucinations, delusions, cognitive impairment, and affective blunting [1]. Although its etiology remains incompletely understood, accumulating evidence suggests that the pathophysiological mechanisms of schizophrenia are closely associated with aberrant functional brain networks [2–5]. Notably, large-scale brain networks—particularly the Default Mode Network (DMN), Central Executive Network (CEN), and Salience Network (SN) (collectively termed the triple networks)—are associated with impaired communication in schizophrenia [6–8]. Functional abnormalities in these networks correlate strongly with patients’ cognitive deficits and emotional symptoms [7, 9]. Critically, white matter networks provide the connecting foundation for the triple networks, facilitating efficient information transfer and integration among these functional systems [10]. In psychiatric disorders, white matter abnormalities may disrupt functional coordination within the triple networks, ultimately contributing to clinical symptomatology [11].

In recent years, with the rapid development of neuroimaging techniques, researchers have extensively explored functional connectivity (FC) abnormalities in large-scale brain networks in schizophrenia patients. Numerous studies have confirmed significant changes in functional activity in both gray and white matter in schizophrenia patients [12–15]. In particular, white matter functional signals, which were previously regarded as noise, have been increasingly shown in recent studies to reflect neural activity during both resting states and functional tasks [16–18]. Building on converging evidence, Peer et al. [19] first delineated white-matter functional networks in large cohorts, revealing interacting modules analogous to those in gray matter. Ding et al. [16] subsequently demonstrated stimulus-synchronous and task-specific white-matter BOLD responses across sensory, motor, and resting-state paradigms, providing robust evidence for their functional relevance. Jiang et al. [20] identified altered white-matter networks in schizophrenia patients via FC analyses, and later integrated DTI with fMRI to reveal white-matter functional-structural decoupling, implicating these alterations in the disorder’s neuropathology [15]. Most recently, Huang et al. [21] employed intracranial stereotactic-electroencephalography and fMRI to show that white-matter BOLD FC reflects underlying electrophysiological synchronization and is structurally constrained by fiber tracts across sensory, motor, and resting-state conditions, underscoring its functional and anatomical validity. Collectively, these studies emphasize white-matter functional abnormalities as a pivotal direction for understanding schizophrenia symptoms.

Traditional static FC analysis typically assumes that brain networks are temporally stable, an assumption that may obscure the complexity and dynamics of brain functional activity [22, 23]. In fact, even during resting states, brain functional networks exhibit significant temporal variability and dynamic reorganization [24]. Static FC analysis may overlook crucial pathophysiological information as it fails to capture these time-varying characteristics. In contrast, DFC analysis can reveal temporal variation characteristics in brain network interactions [25–27], thus providing a more comprehensive reflection of abnormal patterns in brain functional networks in schizophrenia patients. Previous research has demonstrated that patients with schizophrenia exhibit extensive DFC abnormalities across multiple key resting-state networks, including the DMN, executive-control (EXE), and sensorimotor (SM) networks; these anomalies are associated with psychopathological features such as hallucinations and delusions [7]. Additionally, evidence has linked white-matter functional disturbances to cognitive and symptom severity in schizophrenia [20] and epilepsy [28]. Nevertheless, how white-matter function participates in the dynamic interplay among the triple networks and its relationship to cognitive and symptom severity remains to be elucidated.

Antipsychotic medications remain the cornerstone of schizophrenia treatment [29]. Accumulating evidence shows that these drugs markedly reshape both brain structure and functional networks in patients with schizophrenia [12, 27, 30, 31]. Our longitudinal study revealed that post-treatment enhancement of static white-matter FC is associated with symptomatic improvement [2]. In addition, DFC patterns between the DMN and CEN may serve as predictive markers of therapeutic response to antipsychotics [32]. However, other findings indicate that although antipsychotics robustly modulate FC, these alterations may not directly translate into symptom relief [20, 33]. Notably, the mechanisms by which antipsychotics influence DFC among the triple-network system (DMN-CEN-SN) and white-matter networks remain poorly understood. Elucidating drug-induced dynamic changes in brain networks is therefore critical for refining treatment strategies and improving patient outcomes.

This study hypothesizes that patients with schizophrenia exhibit abnormalities in the DFC and coupling between the triple networks and white matter networks, and that antipsychotic treatment may modulate these abnormalities. Using the Johns Hopkins University (JHU) white matter atlas [34] and independent component analysis (ICA), we will extract white matter networks and identify triple networks in both patients and healthy controls. Sliding window approaches and Pearson correlation analysis will be employed to investigate their DFC and coupling characteristics. The data processing pipeline is shown in Fig. 1. Through a 5-month longitudinal follow-up of schizophrenia patients, we aim to elucidate the neural mechanisms underlying these dynamic functional changes, thereby providing novel neuroimaging evidence for the therapeutic effects of antipsychotic drugs.

Fig. 1.

The data processing pipeline and the construction of gray matter-white matter functional coupling. First, independent component analysis (ICA) was employed to identify the triple brain networks. Subsequently, functional signals of the triple networks were extracted, followed by clustering and functional state analysis using the sliding window method. Then, based on the JHU white matter atlas, functional signals from 48 brain regions were extracted, and similarly subjected to clustering and functional state analysis using the sliding window method. Finally, the functional coupling between gray matter and white matter was calculated

Methods

Participants

We enrolled 93 first-episode schizophrenia patients from the Department of Psychiatry at Xijing Hospital and recruited 92 healthy controls through local community advertisements. Inclusion criteria comprised patients experiencing their first hospital admission or outpatient visit for any psychiatric disorder, with a cumulative lifetime exposure to antipsychotic medications not exceeding two weeks. Healthy participants had no history of neurological or psychiatric disorders and were not taking any medications affecting the central nervous system. All patients underwent comprehensive clinical evaluations and met the diagnostic criteria for schizophrenia according to the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) [1]. Among them, 23 patients received antipsychotic medication within two weeks of undergoing baseline MRI scanning and clinical assessments, while the remaining patients were medication-naïve. Clinical and psychological assessments were conducted by two senior clinical psychiatrists. At baseline, psychometric evaluations and MRI scans were performed on the same day, with follow-up clinical assessments and MRI scans conducted approximately five months later. During this period, the treatment plan for the patients was determined by the clinical psychiatrist based on the individual patient’s condition. We collected demographic information from all participants, including age, gender, and education level. Symptom severity was assessed on the scan day using the Positive and Negative Syndrome Scale (PANSS) [35]. Cognitive function was evaluated using the Digit Symbol and Digit Span tests from the China-Revised Wechsler Adult Intelligence Scale [36], along with the semantic (category) fluency subtest from the Beijing version of the Montreal Cognitive Assessment [37]. Current antipsychotic medication doses at the time of MRI scanning were converted to defined daily doses [38].

This study was approved by the Institutional Review Board of the First Affiliated Hospital (Xijing Hospital) of the Fourth Military Medical University, and all participants provided written informed consent in accordance with the principles of the Declaration of Helsinki. Clinical trial number: not applicable.

Imaging acquisition

High-resolution T1-weighted MRI and resting-state fMRI were obtained on General Electric Discovery MR750 3.0 T scanner. The following parameters of T1-weighted MRI were utilized: repetition time (TR) of 8.2 ms, echo time (TE) of 3.2 ms, flip angle of 12°, field of view (FOV) set to 256 mm × 256 mm, matrix size of 256 × 256, slice thickness of 1 mm, section gap of 0 mm, and a total of 196 slices acquired. Regarding the fMRI sequence with a total of 210 volumes of 45 slices covering the whole brain, the parameters were as follows: TR of 2000 ms, TE of 30 ms, flip angle of 90°, FOV of 240 mm × 240 mm, matrix size of 64 × 64, slice thickness of 3.5 mm, section gap of 0 mm. Moreover, a custom-built MRI head coil, fitted with foam padding and earplugs, was employed to reduce participant head movement and to mitigate the noise from the scanner. Throughout the data collection process, participants were instructed to remain motionless with their eyes closed [39], avoiding sleep and conscious thought while inside the scanner.

Image preprocessing

First, T1-weighted images were segmented into gray matter, white matter, and cerebrospinal fluid using SPM12 (https://www.fil.ion.ucl.ac.uk/spm/software/) and spatially normalized to the Montreal Neurological Institute (MNI) template. Functional image preprocessing was carried out with DPABI (http://rfmri.org/dpabi) and included: discarding the first ten volumes to allow signal equilibration; slice-timing correction; realignment of functional images to the mean volume and co-registration to the anatomical image; excluding subjects whose head motion exceeded 2 mm or 2°; removing linear trends; regressing out 24 motion parameters [40] and the mean cerebrospinal-fluid signal (white-matter and global signals were not regressed to avoid interference with the target signals); calculating framewise displacement (FD) and scrubbing volumes with FD >1 mm. Gray-matter BOLD signals were band-pass filtered at 0.01–0.1 Hz, whereas white-matter BOLD signals were filtered at 0.01–0.15 Hz to retain higher-frequency white-matter functional information. Separate 6 mm FWHM Gaussian smoothing kernels were applied to white-matter and gray-matter masks to minimize signal mixing between tissues. Finally, the data were normalized to the standard EPI template and resampled to 3 mm isotropic voxels for subsequent analyses. In addition, the two-sample t-test revealed no significant group difference in mean FD (p = 0.053, T = 1.95).

Triple network construction

Independent component analysis (ICA) was applied to discern large-scale networks across participants, utilizing the GIFT software (version 2.0) [41]. This involved dimensionality reduction to 20 components, determined by the minimum description length criterion, and ICs estimation via the Infomax algorithm with 100 iterations of the ICASSO algorithm for stability. Individual components were reconstructed using a dual-regression method. The resulting ICs’ time courses and spatial maps were standardized to Z-scores. Evaluation of ICs was based on criteria such as peak activations in gray matter, low-frequency fluctuations in time courses, and minimal overlap with white matter structures and artifacts. Based on the widely adopted visual inspection procedure, the five key components corresponding to the triple network model were identified: the Salience Network (SN), anterior Default Mode Network (aDMN), posterior Default Mode Network (pDMN), left Central Executive Network (lCEN), and right Central Executive Network (rCEN) (Fig. 2).

Fig. 2.

Spatial distribution of the triple networks. The five key components corresponding to the triple network model: A the Salience Network (SN), B anterior Default Mode Network (aDMN), C posterior Default Mode Network (pDMN), D left Central Executive Network (lCEN), and E right Central Executive Network (rCEN). The values of X, Y, and Z correspond to the coordinates of the brain network

Dynamic time-varying cross-network interactions

In this study, we investigated DFC patterns within the triple networks and white matter networks. First, we extracted mean time series from five brain networks of the triple networks for each participant, along with 48 tract regions (21 per hemisphere and 6 commissural tracts) from the JHU ICBM-DTI-81 white matter atlas [34] (Supplementary Table 1). Subsequently, we employed a sliding time window approach using the DynamicBC toolbox [42] to analyze these time series. There is currently no formal consensus on window length in the field. Guided by previous work [43], we adopted a sliding-window length of 50 TRs (100 s) with a step size of 1 TR (2 s), yielding 151 windows per participant. In addition, we examined alternative window lengths (15 TR and 30 TR) with a step size of 1 TR each to further investigate the potential effects of different sliding-window durations on dynamic functional connectivity outcomes (Supplementary Tables 2 and 3). For each window, we computed Pearson correlation coefficients between 5 Gra

y matter and 48 white matter networks, generating 5 × 5 and 48 × 48 correlation matrices respectively. Fisher’s r-to-z transformation was applied to standardize the correlation distributions. The variability of DFC, captured through standard deviations of functional connectivity across windows, was measured to assess dynamic interactions within both the triple networks and white matter networks.

Additionally, we computed the Network Interaction Index (NII) for each window as a measure of cross-network interactions [44]. The NII evaluates dynamic functional integration within the triple network framework by quantifying interaction strengths between the SN and DMN, as well as between SN and CEN, following established computational methods [43]. Specifically, the NII is calculated based on functional connectivity strengths between SN-DMN and SN-CEN. Higher NII values indicate more prominent roles of SN in mediating switching and information transfer between DMN and CEN. This index effectively captures dynamic interaction patterns among the triple networks, where NII alterations in psychiatric disorders like schizophrenia may reflect abnormal network integration [32, 44]. Thus, the NII provides a crucial quantitative measure for understanding dynamic functional reorganization of triple networks in disease states. Furthermore, we calculated the Network Interaction Index (NII) as a metric for assessing interaction strength between brain networks. To assess dynamic connectivity and coordinated activity in brain networks, we computed both the mean NII (mNII) and its variability (sNII, calculated as the standard deviation of NII) [45].

Cluster analysis and temporal characteristics

Consistent with previous studies [43, 44, 46], we applied k-means clustering to all windowed functional connectivities (states) characterized by their frequency and structure. The similarity between windowed FC matrices was estimated using L1 distance (Manhattan distance) [47]. Cluster validity analysis (silhouette) was performed across all subjects’ samples to determine the optimal cluster number. Specifically, subsampling analysis was conducted to reduce redundancy between windows and computational demands [22]. The reproducibility of FC states was established through bootstrap resampling and split-half validation. The optimal cluster number for both white matter networks and triple networks was consistently determined as k = 3 based on silhouette criterion analysis of cluster validity index (Fig. 3).

Fig. 3.

Clustering results of triple network functional connectivity states for all subjects. The median clustering matrices for States 1, 2, and 3 are displayed, with the total occurrence rates of each state presented as percentages, specifically 31.38% for state 1, 38.31% for state 2, and 30.31% for state 3

We analyzed the temporal characteristics of DFC states (fractional window, mean dwell time, and number of transitions). The fractional window measures the proportion of time spent in each state, expressed as a percentage [46, 48]. Mean dwell time represents the average duration participants remained in a given state, calculated by averaging the number of consecutive windows belonging to one state before switching to another [46, 48]. The number of transitions indicates how often switches occurred between two states, with higher transition counts reflecting lower stability [46, 48]. These metrics provide a comprehensive evaluation of the temporal evolution of functional network states within the triple network and white matter network.

Functional coupling between white matter and triple networks

In this study, the functional coupling between white matter networks and the triple networks was assessed using a method consistent with previous research [49, 50]. Specifically, we computed Spearman correlation coefficients between each column in the white matter FC matrix and its corresponding column in the triple network FC matrix to represent regional coupling for each individual. During this calculation, self-connections within each column were excluded. This yielded a 5 × 48 matrix per participant, where each element represented the FC coupling value between a given region and all other regions of interest (ROIs) across the whole brain.

Statistical analysis

All clinical and imaging data were statistically analyzed using SPSS version 25.0. Differences in gender between groups were assessed using the Chi-square test with a p value less than 0.05 considered statistically significant. The median matrix for subjects was obtained by estimating the states at the group level, and independent sample t-tests were used to compare the DFC strength, fractional windows, mean dwell time, and number of transitions of each brain region in the triple network and white matter network between patients and healthy controls within each state (p < 0.05 was considered the significance threshold). The false discovery rate (FDR) method was used to correct for multiple comparisons. Specifically, FDR correction was performed separately on the p-values for each DFC state, enabling independent evaluation of significance within the context of each individual state. Paired sample t-tests were used to compare the changes in the aforementioned statistical values before and after treatment in patients (p < 0.05). Age and education level were included as covariates in the intergroup comparisons. The longitudinal follow-up analysis included scan interval and medication dosage as covariates. Correlation analysis was performed to describe the relationship between individual DFC and relevant clinical and neuropsychological characteristics.

Results

Demographic and clinical characteristics

The demographic and clinical characteristics of all participants are presented in Table 1. While no significant gender differences were observed between the patient and healthy control groups, statistically significant differences were found in age and education level (p < 0.05). Treatment regimens were determined by clinicians following standard clinical practice. During follow-up, 39 patients were ultimately included in the longitudinal analysis (Table 2) due to attrition and data quality issues. These patients received approximately five months of antipsychotic treatment, with all follow-up cases prescribed second-generation antipsychotics: paliperidone (n = 5), risperidone (n = 21), olanzapine (n = 13), amisulpride (n = 2), and quetiapine (n = 1). Some patients received combination therapy. Following antipsychotic treatment, patients showed a significant reduction in PANSS scores compared with baseline (p < 0.001), whereas cognitive scores did not change significantly (p > 0.05) (Table 2).

Table 1.

Baseline demographic and clinical characteristics of participants (mean ± standard deviation)

Characteristic	Patients (n = 93)	HC (n = 92)	p-values
Age (y)	23.42 ± 6.97	33.18 ± 11.08	< 0.001
Gender (M/F)	52/41	50/42	0.830
Education level (y)	11.94 ± 3.04	15.20 ± 4.53	< 0.001
Illness duration (mon)	14.35 ± 22.10	NA	NA
PANSS scores
Positive score	21.69 ± 5.27	NA	NA
Negative score	20.42 ± 7.30	NA	NA
General psychopathology score	43.97 ± 8.25	NA	NA
Total score	86.31 ± 14.07	NA	NA
Cognitive measures
Digit symbol	46.46 ± 14.13	NA	NA
Digit Span
Forward	9.03 ± 1.65	NA	NA
Backward	4.93 ± 1.84	NA	NA
Semantic verbal fluency	17.17 ± 5.19	NA	NA

Note: All the data represent mean ± standard deviation unless otherwise indicated. Four patients withdrew from the cognitive testingAbbreviations: y, year; M/F, male/female; mon, month; HC, healthy controls; NA, not applicable; PANSS, Positive and Negative Syndrome Scale.

Table 2.

Demographic and clinical characteristics (mean ± standard deviation) of participants in the follow-up group

Characteristic	Pre-treatment (n = 39)	Post-treatment (n = 39)	p-values
Age (y)	24.08 ± 8.66	NA	NA
Gender (M/F)	22/17	NA	NA
Education level (y)	12.13 ± 3.05	NA	NA
Illness duration (mon)	14.95 ± 21.30	NA	NA
PANSS scores
Positive score	22.92 ± 4.90	10.33 ± 3.80	< 0.001
Negative score	21.41 ± 6.36	14.72 ± 4.76	< 0.001
General psychopathology score	44.77 ± 8.79	27.10 ± 6.61	< 0.001
Total score	89.41 ± 14.23	48.18 ± 17.31	< 0.001
Cognitive measures
Digit symbol	45.74 ± 12.11	47.83 ± 12.58	0.270
Digit Span
Forward	8.62 ± 1.63	8.72 ± 1.67	0.494
Backward	4.77 ± 1.81	4.33 ± 1.39	0.117
Semantic verbal fluency	16.97 ± 4.28	17.50 ± 4.29	0.276

Note: All the data represent mean ± standard deviation unless otherwise indicated. Three patients withdrew from the cognitive testing

Abbreviations: PANSS Positive and Negative Syndrome ScaleAbbreviations: y, year; M/F, male/female; mon, month; HC, healthy controls; NA, not applicable; PANSS, Positive and Negative Syndrome Scale.

Dynamic functional connectivity strength and Temporal characteristics of the triple networks

K-means clustering was applied to classify the DFC matrices across all subjects. Analysis of the DFC states revealed three distinct triple-network FC patterns. As shown in Fig. 3, State 1 (31.38%) and State 3 (30.31%) were relatively similar, whereas State 2 (38.31%) was the most prevalent. In State 1, compared to healthy controls, patients exhibited significantly decreased FC between the aDMN and pDMN (p = 0.002, FDR corrected, T = − 3.12). No significant differences were observed between States 2 and 3. Following a period of antipsychotic treatment, the FC changes between brain regions after treatment were not statistically significant compared to baseline.

After controlling for age and education level, we further compared between-group differences in temporal characteristics, including fractional window, mean dwell time, and number of transitions. Except for the statistically significant difference observed in the fractional window of State 3 (p = 0.045), no other temporal characteristics showed statistically significant differences. In the longitudinal analysis, the fractional window of State 2 showed a statistically significant increase after treatment (p = 0.030), while no significant changes were observed in other temporal characteristics.

Dynamic time-varying cross-network interactions

We compared between-group differences in sNII and mNII values of dynamic triple network interactions. In this study, no statistically significant differences were observed between groups for these two indicators at baseline (p > 0.05). mNII showed a statistically significant decrease after treatment (p = 0.020, t = 2.427), while no other indicators showed statistically significant differences during the entire treatment process (p > 0.05).

Dynamic functional connectivity strength and Temporal characteristics of white matter networks

We identified three white matter FC state patterns and compared the connection strengths between schizophrenia patients and healthy controls across these states. As shown in Fig. 4, the total occurrence rates of these three states were 21.37% for State 1, 42.46% for State 3, and 36.17% for State 2. As shown in Fig. 5, compared with healthy controls, patients exhibited abnormal connectivity in over 90% of brain regions across all three states, primarily characterized by increased functional connectivity (approximately 80%) (p < 0.001, FDR corrected) and decreased functional connectivity in the remaining patients (approximately 20%) (p < 0.001, FDR corrected). After a period of antipsychotic treatment, changes in functional connectivity were observed in all three states (p < 0.001, FDR corrected) (Table 3).

Fig. 4.

Clustering results of white matter network functional connectivity states for all subjects. The median clustering matrices for states 1, 2, and 3 are displayed, with the total occurrence rates of each state presented as percentages, specifically 21.37% for state 1, 42.46% for state 2, and 36.17% for state 3

Fig. 5.

Differences in functional connectivity (FC) among the three states for the first-episode schizophrenia patient group and the healthy control group (p < 0.001). Green lines indicate increased FC, while red lines indicate decreased FC. The green grid represents the right white matter network; yellow grid, white matter network in the middle; red grid, left white matter network

Table 3.

White matter functional connectivity was statistically significant (p < 0.001) after antipsychotic treatment compared to baseline in all three States

	Area 1	Area 2	p value	Statistic
State 1	Body of corpus callosum	Superior longitudinal fasciculus L	< 0.001	−3.75
	Inferior cerebellar peduncle L	Cingulum (hippocampus) L	< 0.001	−4.25
	Sagittal stratum L	Tapetum L	< 0.001	−3.54
State 2	Inferior cerebellar peduncle L	Anterior corona radiata R	< 0.001	3.53
	Cerebral peduncle R	Posterior thalamic radiation R	< 0.001	−3.47
	Posterior limb of internal capsule L	Fornix (cres)/Stria terminalis L	< 0.001	4.15
State 3	Posterior limb of internal capsule R	Sagittal stratum L	< 0.001	3.67
	Posterior limb of internal capsule R	Uncinate fasciculus L	< 0.001	3.78
	Posterior limb of internal capsule L	Sagittal stratum L	< 0.001	3.97
	Superior corona radiata L	Posterior corona radiata R	< 0.001	−3.67

Note: L left, R right

After controlling for age and education level, we further compared between-group differences in temporal characteristics, including fractional window, mean dwell time, and number of transitions. Compared with healthy controls, the patient group showed a larger fractional window in State 2 (p = 0.005, F = 4.42) and a smaller fractional window in State 3 (p < 0.001, F = 10.60). In the longitudinal analysis, compared with pre-treatment, post-treatment patients exhibited a decreased fractional window in State 2 (p = 0.020, T = −2.42), an increased fractional window in State 3 (p = 0.028, T = 2.28), and a shortened mean dwell time in State 2 (p = 0.019, T = −2.44). As shown in Fig. 6.

Fig. 6.

Dynamic functional connectivity characteristics of white matter. Compared to controls, schizophrenia patients had larger fractional windows for white matter state 2 (p = 0.005) and smaller for state 3 (p < 0.001), longer dwell times in state 2 (p = 0.009), and shorter in state 3 (p < 0.001). In longitudinal analysis, post-treatment, patients showed reduced state 2 fractional windows (p = 0.020), increased state 3 fractional windows (p = 0.028), and decreased dwell times in state 2 (p = 0.019)

Functional coupling of triple network and white matter network

Figure 7 demonstrates significant alterations in functional coupling between triple networks and white matter networks in schizophrenia patients compared to healthy controls. Five brain region pairs exhibited significantly changed functional coupling (p < 0.0001). Patients showed reduced functional coupling between the corpus callosum and pDMN relative to healthy controls, along with decreased coupling between: (1) right inferior cerebellar peduncle and aDMN, (2) right inferior cerebellar peduncle and pDMN, (3) right posterior limb of internal capsule and rCEN, and (4) left posterior limb of internal capsule and rCEN. Following antipsychotic treatment, corpus callosum-pDMN coupling significantly increased compared to pre-treatment levels (p < 0.001, FDR-corrected). Conversely, functional coupling between left inferior cerebellar peduncle and aDMN was weakened relative to baseline measurements (p < 0.001, FDR-corrected) (Table 4).

Fig. 7.

T-value mapping of functional coupling between the triple networks and white matter functional networks. (A) represents the comparison between patients at baseline and healthy controls. (B) indicates the comparison between post-treatment and pre-treatment conditions. Cool color tones represent lower T-values, whereas warm color tones indicate higher T-values. The asterisk symbol (*) denotes statistical significance at p < 0.0001

Table 4.

Triple network and white matter network functional coupling were statistically significant (p < 0.001) after antipsychotic treatment compared to pre-treatment in all three States

	White matter network	Triple network	p value	Statistic
Pre-treatment	Body of corpus callosum	pDMN	< 0.0001	−4.19
	Inferior cerebellar peduncle R	aDMN	< 0.0001	4.11
	Inferior cerebellar peduncle R	pDMN	< 0.0001	4.57
	Posterior limb of internal capsule R	rCEN	< 0.0001	4.66
	Posterior limb of internal capsule L	rCEN	< 0.0001	4.50
Post-treatment	Body of corpus callosum	pDMN	< 0.0001	5.06
	Inferior cerebellar peduncle L	aDMN	< 0.0001	−4.47

Note: aDMN anterior Default Mode Network, pDMN posterior Default Mode Network, rCEN right Central Executive Network. R right, L left

Correlation analysis

We further investigated the correlations between temporal characteristics (including both triple networks and white matter networks during pre- and post-treatment phases) and changes in PANSS scores and cognitive function before versus after treatment. After controlling for age and education level, no significant correlations were found. This included the absence of statistically significant correlations between medication dosage and scan intervals (p > 0.05).

Discussion

This study utilized DFC analysis to examine the differences in triple networks and white matter functional networks between first-episode schizophrenia patients and healthy controls, along with their changes in an approximately 5-month follow-up cohort. The investigation focused on temporal characteristics (fractional window, mean dwell time, and transition counts), DFC strength, functional coupling between triple networks and white matter networks, and neuroimaging changes during follow-up. Three distinct connectivity states were identified for both triple networks and white matter functional networks. Between-group differences revealed abnormal DFC patterns in the triple networks and white matter networks of schizophrenia patients, which could be ameliorated by antipsychotic treatment - consistent with our previous findings that white matter structural abnormalities are modifiable by antipsychotics [30]. Our study provides evidence for DFC abnormalities in these networks among schizophrenia patients, corroborating prior research findings [51]. Altered DFC may underlie abnormal brain function and clinical symptoms in schizophrenia [52]. Furthermore, this work offers new evidence extending schizophrenia’s dysconnection hypothesis from static to dynamic brain networks, while highlighting the potential value of abnormal DFC in white matter and triple networks for elucidating the disorder’s pathophysiological mechanisms.

Our triple-network DFC analysis revealed that, compared with healthy controls, patients exhibited a pronounced reduction in FC between the aDMN and pDMN; this deficit remained statistically unchanged after antipsychotic treatment, whereas the mNII was significantly decreased. The persistent aDMN–pDMN hypoconnectivity suggests that this sub-pathway represents a relatively “stubborn” disconnectivity phenotype. Duan et al. [53] reported decreased DFC variance between the dorsal anterior insula and precuneus in first-episode, drug-naïve patients, indicating aberrant DMN–insula interactions; this aligns directionally with our aDMN–pDMN findings and further supports impaired dynamic integration within DMN subsystems as an early, trans-diagnostic feature of first-episode schizophrenia. Notably, Duan et al. [53] observed a significant reversal of this pathway after only 8 weeks of risperidone monotherapy, implying that DMN-related dynamics can be rapidly normalized when the intervention target is specific and the medication regimen is uniform. In contrast, our 5-month follow-up still failed to detect recovery of aDMN–pDMN connectivity, suggesting that this sub-pathway may follow a steeper dose–response curve or exhibit a prolonged plateau phase. The significant reduction in mNII indicates that antipsychotic medication has engaged a “state-switching” mechanism. Unlike the significant mNII increase reported by Wang et al. [32] at the 12-week follow-up, our results suggest that mNII may follow highly individualized nonlinear trajectories. Collectively, these findings indicate that “locally persistent disconnectivity coupled with cross-network switching plasticity” constitutes a dual-window framework for treatment monitoring and long-term prognosis in first-episode schizophrenia.

Our white-matter DFC analysis revealed widespread aberrant connectivity, most notably an elevated baseline DFC between the right cerebral peduncle and the right posterior thalamic radiation in patient state 2 that decreased after treatment, suggesting that this hyper-connectivity may reflect an early compensatory response. Prior work has reported diffuse microstructural white-matter deficits in schizophrenia [54]; the present findings complement these structural observations by demonstrating, at the functional level, that such compensatory alterations are reversible. Furthermore, compared to healthy controls, baseline patients exhibited a longer mean dwell time and a greater fractional window in State 2 of the white matter network; these elevations potentially reflect abnormal DFC in the diseased brain, characterized by more frequent switching between network states (indicating local dynamic instability in non-State 2 regions) but shorter durations (reflected by lower mean dwell time on non-State 2 states), signifying pathological entrapment in State 2 where the brain exhibits excessive reliance and prolonged locking onto this state, which may constitute the neural basis for symptoms such as cognitive rigidity and the persistence of hallucinations/delusions. When the brain is forced to leave State 2, it fails to sustain effectively in other states, manifesting as a rapid escape or return to State 2; this “escape-return” pattern maintains an unchanged overall number of transitions but reflects aberrant state transition strategies (a lack of the stable exploratory capacity seen in the healthy brain), indicating that DFC abnormalities in schizophrenia may not represent a simple global increase or decrease in flexibility but rather involve dysfunction within specific network states accompanied by altered state transition strategies. Such abnormal DFC patterns have been confirmed to correlate with cognitive impairments and clinical symptoms in schizophrenia patients [27, 53]. Following antipsychotic treatment, follow-up patients showed reductions in both mean dwell time and fractional window in State 2 of the white matter network, which can be interpreted as indicative of DFC normalization. This aligns with findings from gray matter studies [27] where first-episode, medication-naïve patients with low baseline DFC variability exhibited increased variability alongside symptom improvement after treatment, suggesting drug-induced DFC normalization. This normalization likely reflects the modulatory effect of medication on large-scale brain networks: by enhancing DFC variability, it increases the flexibility of state switching, thereby improving cognition and alleviating clinical symptoms [27, 32, 53, 55]. In summary, antipsychotic treatment can recalibrate white matter DFC towards a pattern closer to that of healthy controls, with its mechanism potentially lying in enhancing network flexibility and stability, ultimately leading to the mitigation of clinical manifestations.

This study revealed significantly abnormal functional coupling between the triple networks (SN, CEN, and DMN) and the white matter network in schizophrenia patients compared to healthy controls, a finding that strongly aligns with the “dysconnection hypothesis” of schizophrenia. Within Friston’s theoretical framework [56], such coupling impairment can be interpreted as a failure of functional integration between brain regions, directly leading to disrupted coordination in large-scale network dynamics. Notably, this triple network-white matter coupling dysfunction may share intrinsic mechanistic links with white matter DFC abnormalities—specifically, the pathological entrapment in State 2 and local instability during transitions between non-dominant states. Aberrant coupling likely impairs the triple networks’ regulatory capacity over state transitions, preventing the brain from effectively disengaging from maladaptive network states and thereby exacerbating rigid white matter DFC patterns. Specifically, reduced functional coupling between the corpus callosum and the pDMN reflects impaired interhemispheric information integration, which may constitute a key neural substrate for patients’ cognitive dysfunction [57]. Following antipsychotic treatment, significantly enhanced corpus callosum-pDMN coupling indicates that pharmacotherapy can directly optimize triple network-white matter interactions by restoring bilateral hemispheric coordination. This change coincides with the normalization of white matter DFC, collectively forming the neural mechanism underlying symptom alleviation. This phenomenon also resonates with gray matter network research demonstrating that antipsychotics modulate DFC [27], potentially through reshaping cross-hierarchical network coupling to achieve global brain dynamic equilibrium. Furthermore, post-treatment weakening of coupling between the left inferior cerebellar peduncle and the aDMN provides additional evidence for the drug’s targeted recalibration of aberrant functional connections. This suggests that treatment optimizes functional network architecture and improves clinical symptoms by precisely adjusting coupling strength between critical brain regions.

This study has several limitations. First, the sample size was insufficient, particularly for follow-up patients, and cognitive measurements were not comprehensive. Second, structural characteristics were not compared within the same subjects, and potential overlaps between structural and functional abnormalities require further investigation. Third, follow-up patients received different antipsychotic medications, which may have influenced the results. Fourth, significant differences in age and educational level existed between the patient and control groups. Although these variables were controlled for statistically, residual confounding may remain. Since age affects white matter microstructure and education is linked to cognitive reserve, the observed differences may partly reflect demographic influences rather than disease-specific effects. Future studies should use more closely matched controls. Fifth, although we included first-episode schizophrenia patients, some had already received medication within 2 weeks of enrollment, which may have introduced confounding effects. Therefore, future studies should adopt more stringent inclusion criteria for schizophrenia patients and explore both structural and functional mechanisms underlying disease pathogenesis and treatment-induced dynamic changes.

Conclusion

This study systematically reveals for the first time characteristic abnormalities in DFC and functional coupling between triple networks and white matter networks in first-episode schizophrenia patients: persistent aDMN-pDMN connectivity impairment constitutes the core pathology, while compensatory functional enhancement in specific white matter pathways demonstrates reversibility. Antipsychotic treatment effectively induces synchronous normalization of global DFC patterns and key network couplings. These findings substantiate the white matter network basis of schizophrenia’s dysconnection mechanism, deliver quantifiable dynamic biomarkers for early clinical intervention, and confirm pharmacotherapy-induced modifiability of network-level dynamic information processing deficits—thereby establishing the theoretical foundation for precision network-remodeling therapies.

Abbreviations

y

Year

M/F

Male/female

mon

Month

HC

Healthy controls

NA

Not applicable 

PANSS

Positive and negative syndrome scale

CEN

Central executive network

DMN

Default mode network

DFC

Dynamic functional connectivity

ICA

Independent component analysis

NII

Network interaction index

PANSS

Positive and negative syndrome scale

SN

Salience network

Authors’ contributions

Xusha Wu wrote the original draft and prepared formal analysis; Yan Li, Wenzhong Hu, and Yang Zhang prepared writing review and editing; Xiaowei Kang did the validation; Hong Yin and Yibin Xi did the supervision; Xuan Li did the methodology; All authors have read and agreed to the published version of the manuscript.

Funding information

This research was supported by the Natural Science Basic Research Program of Shaanxi Province (2022JM-460), the Research Incubation Fund of Xi’an People’s Hospital (Xi’an Fourth Hospital) (No. ZD-11), the National Natural Science Foundation of China (81601474).

Data availability

The data that support the findings of this study are available from the corresponding author, Yibin Xi, upon reasonable request.

Declarations

Ethics approval and consent to participate

This study was approved by the Institutional Review Board of the First Affiliated Hospital (Xijing Hospital) of the Fourth Military Medical University, and all participants provided written informed consent in accordance with the principles of the Declaration of Helsinki. Clinical trial number: not applicable.

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.