Disruption of prefrontal and superior temporal gyrus white matter networks in antipsychotic-treated schizophrenia with comorbid metabolic syndrome Authors: Ying Li, Qianqian Liu, Xinyue Chen, Shiji Peng, Hangyu Li, Kaike Liao, Rui Yu, Nian Liu Published: 2026 ABSTRACT BackgroundMetabolic syndrome (MS) is a frequent comorbidity in schizophrenia, yet the characteristics of white matter (WM) network alterations in this subgroup remain largely unknown. This study examined distinctive WM structural network disruptions in schizophrenia patients with comorbid MS and explored whether these alterations mediate the relationship between schizophrenia and metabolic abnormalities.MethodsTwenty-six schizophrenia patients with MS (SZ-MS), twenty-two without MS (SZ-nMS), and twenty-four age- and sex-matched healthy controls (HC) underwent diffusion tensor imaging. Individual WM structural networks were constructed, and group differences in connectivity and topological organization were analyzed using network-based statistics and graph theory, controlling for age and sex.ResultsBoth SZ-MS and SZ-nMS patients exhibited reduced FA-weighted connectivity in prefrontal, parietal, and temporal cortices, thalamus, and hippocampus compared to HC. Relative to SZ-nMS, the SZ-MS group exhibited significant reduction in prefrontal FA-weighted connectivity. Topological metrics revealed global and nodal disruptions in both patient groups, notably in prefrontal cortex, bilateral thalamus, supramarginal gyrus, left precuneus, and right superior temporal gyrus. The SZ-MS group demonstrated greater nodal efficiency impairments in prefrontal and right superior temporal gyrus regions. Further exploratory analyses revealed that WM network metrics correlated with triglycerides, BMI, waist circumference, hip circumference, and fasting glucose. Additionally, reduced nodal efficiency in the right superior temporal gyrus exhibited a potential mediating pathway between SZ-MS and elevated BMI.ConclusionThese findings indicate that SZ-MS patients display characteristic WM network disruptions, predominantly in prefrontal and right superior temporal areas. These alterations are associated with metabolic abnormalities, providing neuroimaging evidence that may clarify the neuropathological mechanisms connecting schizophrenia and metabolic dysfunction. Introduction As a chronic psychiatric disorder, schizophrenia is defined by profoundly disturbed cognition, emotion, and behavior, leading to substantial social and occupational dysfunction. Its high recurrence rate, associated disability, and suboptimal treatment adherence contribute to a considerable public health burden (1, 2). Antipsychotic medications, particularly second-generation agents, are the cornerstone of treatment and are generally effective for symptom control (3, 4). However, prolonged use is frequently accompanied by adverse metabolic changes, such as increased weight, dyslipidemia, and insulin resistance, which collectively elevate metabolic syndrome (MS) risk (5). The reported prevalence of MS among individuals with schizophrenia ranges from 40% to 43%, significantly exceeding that of the general population (6). MS aggravates psychiatric symptoms, impairs cognition, reduces medication compliance, and increases relapse risk, thereby contributing to poorer clinical outcomes and higher discontinuation rates (7–9). Despite these clinical implications, the neural correlates and imaging biomarkers underlying schizophrenia comorbid with MS remain poorly characterized. Previous investigations have primarily focused on the systemic metabolic and physiological aspects of MS in schizophrenia, while neuroimaging evidence concerning brain structural and functional alterations is comparatively limited. Structural MRI studies have shown reductions in total brain volume, orbitofrontal cortical thickness, and reward circuitry volumes in schizophrenia patients with MS relative to those without MS (10). Functional MRI (fMRI) research has revealed abnormal functional coupling in the bilateral middle frontal gyrus (11) and reduced regional homogeneity in the inferior orbitofrontal cortex (12). Other findings include hypoperfusion in the orbitofrontal cortex and greater insula and middle/superior frontal gyri connectivity (13). Nevertheless, these studies primarily emphasize gray matter and functional activity, leaving the microstructural properties of white matter (WM) networks largely unexplored. The WM connectome forms the anatomical backbone for efficient neural communication, and graph-theoretical approaches provide a powerful framework for quantifying its organization and efficiency (14, 15). Wu and colleagues first employed fiber number (FN)-based graph analysis to reveal disrupted topological architecture within the WM networks of schizophrenia patients with comorbid MS (16). Although informative, FN-based measures primarily reflect global tract count and are susceptible to biases related to tractography seeding and fiber length, which may limit their sensitivity to localized microstructural abnormalities. In contrast, network-based statistics (NBS) can identify specific connectivity subnetworks exhibiting significant alterations while effectively controlling the family-wise error rate (FWER) (17, 18), thereby complementing conventional graph-theoretical approaches. Furthermore, fractional anisotropy (FA), derived from diffusion tensor imaging (DTI), is sensitive to microstructural tissue properties such as myelination and axonal density (19, 20). Although FA is influenced by multiple biological and methodological factors and is not a direct or specific measure of axonal integrity, it provides a useful summary of WM microstructure along reconstructed pathways. Compared with FN-based measures, FA-weighted network analysis may offer a more refined characterization of microstructural alterations in SZ-MS patients. Therefore, integrating FA-weighted NBS with graph-theoretical metrics may provide complementary insights into the WM network disruptions associated with schizophrenia and metabolic comorbidity. Building upon this foundation, the present study enrolled schizophrenia patients with MS (SZ-MS), schizophrenia patients without MS (SZ-nMS), and demographically matched healthy controls (HC). By integrating graph-theoretical analysis with NBS using FA-weighted WM connectomes, we aimed to (1) characterize WM network alterations unique to SZ-MS and (2) examine their role in mediating the relationship between schizophrenia and metabolic abnormalities. This study is expected to identify key neuroimaging markers of schizophrenia with comorbid metabolic syndrome. Materials and methods This research received approval from the Ethics Committee of the Affiliated Hospital of North Sichuan Medical College (approval number: 2022ER416-1) and adhered to the principles of the Declaration of Helsinki. All participants were fully informed of the study’s procedures, and written informed consent was obtained prior to participation. Participants From psychiatric hospitals in Guang’an City, Sichuan Province, China (specifically, Wusheng Bashu Psychiatric Hospital and Wusheng Ruikang Psychiatric Specialist Hospital), patients diagnosed with schizophrenia were enrolled. Participants received no financial compensation but were reimbursed a fixed travel allowance for their participation. All participants were receiving standard clinical care during hospitalization. Inclusion criteria for schizophrenia were based on the DSM-5 (21), and included: ① age ≥ 18 years; ② right-handedness; ③ ongoing treatment with second-generation antipsychotic medications; ④ absence of comorbid psychiatric disorders or substance dependence; ⑤schizophrenia onset ascertained from patient/family reports, medical records, and the Nottingham Onset Schedule (22). MS was diagnosed in accordance with the race-adjusted criteria outlined by NCEP ATP-III (23). A diagnosis required a minimum of three of the following conditions: ① central obesity, representing a waist circumference ≥80 cm in women or ≥90 cm in men; ② reduced high-density lipoprotein cholesterol (HDL-C), defined as <1.29 mmol/L (50 mg/dL) in women or <1.03 mmol/L (40 mg/dL) in men, or pharmacological management for low HDL-C; ③ elevated triglycerides (TG) ≥1.7 mmol/L (150 mg/dL) or current treatment for hypertriglyceridemia; ④ hypertension, indicated by systolic/diastolic blood pressure ≥130/85 mmHg, or a prior diagnosis and ongoing treatment for hypertension; and ⑤ raised fasting blood glucose (FBG) ≥5.6 mmol/L (100 mg/dL) or previously diagnosed type 2 diabetes mellitus. HC participants were recruited through community advertising and matched to the schizophrenia groups on sex, age, and education. All HCs were confirmed to be right-handed and denied any history of psychiatric or neurological disorders. As part of the screening process, the SCID-NP was administered with the dual purpose of confirming that participants themselves had no psychiatric illness and ensuring that none of them had a first-degree relative with a psychiatric diagnosis. For both schizophrenia patients and HC, exclusion criteria were as follows: ① contraindications to MRI scanning (e.g., claustrophobia, metallic implants); ② history of alcohol or substance abuse; ③ previous organic brain lesions or significant head trauma; ④ severe systemic illnesses such as hepatic cirrhosis or cardiovascular disease; and ⑤ poor-quality DTI data or incomplete brain coverage. Medical record reviews confirmed that all schizophrenia participants were free of MS prior to their most recent hospitalization. A priori sample size estimation was performed using G*Power 3.1.9.7 (24). Assuming an effect size of f = 0.40, a significance level (α = 0.05), and a desired power (1-β = 0.80) for a one-way analysis of variance with three groups, the required total sample size was calculated to be 66 participants. Ultimately, the final analysis comprised 72 participants, including 26 SZ-MS (13 males and 13 females; mean age 41.69 ± 10.34 years), 22 SZ-nMS (15 males and 7 females; mean age 38.18 ± 10.00 years), and 24 HC (9 males and 15 females; mean age 40.38 ± 9.77 years). Among SZ-MS patients, 2 received clozapine monotherapy, 9 were on risperidone monotherapy, and 15 received combination therapy; the SZ-nMS group comprised 3, 7, and 12 patients under the corresponding regimens. Daily antipsychotic dosages were standardized to chlorpromazine (CPZ) equivalents for comparison across medication types (22, 25). Clinical data collection Data collection for clinical, demographic, and metabolic data coincided with the MRI scan date. Demographic information included sex, age, height, weight, illness duration, and years of education. Metabolic parameters included waist circumference, blood pressure, body mass index (BMI), low-density lipoprotein cholesterol (LDL-C), HDL-C, total cholesterol (TC), TG, and FBG. All clinical scales were administered by trained psychiatrists with at least five years of clinical experience, who underwent standardized training. Scoring and clinical interpretation were supervised by a researcher with neuropsychological qualifications. The Positive and Negative Syndrome Scale (PANSS) was used to evaluate psychiatric symptom severity, with higher scores indicating greater psychopathology (26). In healthy controls, the PANSS was administered solely to screen for and exclude the presence of psychotic-spectrum symptoms and to maintain procedural uniformity across all participants. Cognitive function was assessed using the Brief Assessment of Cognition in Schizophrenia (BACS), which evaluates six cognitive domains: working memory, verbal memory, verbal fluency, motor speed, attention, and executive functioning (27). Lower BACS scores reflect poorer cognitive performance. Adverse effects of treatment were assessed using the Treatment Emergent Symptom Scale (TESS) and the Rating Scale for Extrapyramidal Side Effects (RSESE), where higher scores indicate more severe side effects (28). Image data acquisition Imaging data were acquired using a 3.0 Tesla MRI system (Discovery 750, GE Healthcare) at the Affiliated Hospital of North Sichuan Medical College. Standardized instructions required participants to maintain stillness, keep their eyes closed, stay awake, and avoid focused mental activity throughout the scan. To mitigate motion artifacts and acoustic interference, foam padding and earplugs were utilized. DTI data were acquired using a diffusion-weighted spin-echo echo-planar imaging sequence with the following parameters: repetition time (TR) = 7900 ms, echo time (TE) = 86 ms, flip angle = 90°, slice thickness = 2 mm, voxel size = 2 × 2 × 2 mm3, field of view (FOV) = 208 × 208 mm2, acquisition matrix = 104 × 104, and diffusion weighting (b-values) = 0 and 1000 s/mm2 with 30 non-collinear diffusion gradient directions. High-resolution T1-weighted anatomical images were obtained using a spoiled gradient-echo sequence with the following settings: TR = 7 ms, TE = 3 ms, flip angle = 9°, slice thickness = 1 mm, voxel size = 1 × 1 × 1 mm3, FOV = 256 × 256 mm2, and acquisition matrix = 256 × 256. Image data processing and network construction DTI preprocessing and network construction were undertaken using FSL (https://fsl.fmrib.ox.ac.uk/fsl) and the Pipeline for Analyzing Brain Connectivity Images (PANDA) toolbox. DTI and T1-weighted images were first converted from DICOM to NIFTI format. Head motion and eddy current distortions were corrected using FSL’s eddy tool, with all images aligned to the first b0 reference image. Susceptibility-induced distortions were corrected using FSL’s topup based on reverse-phase-encoded b0 images. Non-brain tissue was then removed and FA maps were computed. Quality control was performed at key preprocessing steps. After eddy correction, automated QC metrics were generated using FSL’s tool to assess residual motion and outlier slices. Additionally, all diffusion-weighted images were visually inspected for severe artifacts prior to tractography. Participants with mean frame-wise displacement greater than 0.3 mm or absolute rotation greater than 2° were excluded. Deterministic fiber tracking was conducted over the whole brain with the following parameters: FA threshold = 0.2, angular threshold = 45°, step length = 1 mm, and seed density = 1 seed per voxel. Tracking was initiated from all WM voxels with FA > 0.2 and terminated when FA dropped below 0.2 or when the turning angle exceeded 45°. Node Definition: The brain was parcellated into 90 cortical and subcortical regions using the Automated Anatomical Labeling (AAL) atlas, with each region defined as a network node. Anatomical correspondence was achieved by spatially normalizing the T1-weighted images to the Montreal Neurological Institute (MNI) ICBM152 standard template. The transformation parameters derived from this step were subsequently utilized to align and parcellate each subject’s DTI data. The accuracy of spatial normalization was visually verified for each subject by overlaying the transformed T1-weighted images onto the MNI template. Subjects with obvious registration errors were excluded from further analysis. Edge Definition: Edges between any two nodes were defined by the mean FA value of all fiber tracts connecting the respective regions. This edge weight serves as a streamline-derived summary of the underlying WM microstructure, specifically the mean diffusion anisotropy along reconstructed pathways. It is not a direct measure of anatomical connection strength or tract count. FA is sensitive to microstructural tissue properties, such as myelination and axonal density. This sensitivity allows FA-based measures to complement fiber count-based metrics. Fiber count-based metrics are more susceptible to biases related to tractography seeding and fiber length. For any pair of brain regions, we identified all fiber tracts with one endpoint in each region, calculated the mean FA value across these tracts, and defined this mean as the edge weight between the two regions, thereby generating a symmetric 90 × 90 FA-weighted adjacency matrix for each participant. White matter network analysis Because there is no universally accepted thresholding approach for structural networks, sparsity thresholds were applied to ensure network stability and comparability across participants. A network was retained if it exhibited a mean degree greater than log(N/2) and a small-worldness index exceeding 1.1 (29, 30). Sparsity thresholds ranged from 0.10 to 0.40 (in 0.01 increments) to assess consistency and mitigate spurious connections. For each global and nodal metric, the area under the curve across the sparsity range was computed for each participant. Global and regional topological metrics of the WM network were computed using the GRETNA toolbox (https://github.com/sandywang/GRETNA). Global metrics included normalized clustering coefficient (Gamma), characteristic path length (Lp), normalized characteristic path length (Lambda), small-worldness (Sigma), clustering coefficient (Cp), global efficiency (Eg), and local efficiency (Eloc). Cp and Gamma reflect local segregation. Lp and Lambda reflect global integration. Sigma captures the balance between segregation and integration. Eg and Eloc quantify information transfer efficiency at the whole-network and regional levels, respectively. Nodal metrics encompassed nodal degree centrality (DC), nodal efficiency (NE), and nodal shortest path length (NLP). DC represents the number of direct links to a node, reflecting its centrality within the network. NE quantifies the efficiency of information exchange between nodes, whereas NLP characterizes the extent to which a node facilitates network-wide integration (15). Statistical analysis Data are shown as means ± standard deviation or median (interquartile range), depending on distributional characteristics. Data normality and variance homogeneity were assessed using the Shapiro-Wilk and Levene tests, respectively. For variables following a normal distribution, between-group differences were examined using independent-samples t-tests (two groups) or one-way analyses of variance with post hoc comparisons (three groups). Skewed data were assessed using Mann-Whitney U tests (two groups) or Kruskal-Wallis tests (three groups) as appropriate. Group differences in sex distribution were assessed with the chi-square test. To minimize the influence of potential confounders, age and sex were included as covariates in all relevant analyses. To assess the robustness of our findings, we performed sensitivity analyses by additionally including years of education as a covariate in the same statistical models. The results remained essentially unchanged, with all significant group differences retaining the same direction and significance (Supplementary Tables S1–S5). White matter FA-weighted connectivity NBS were applied to examine group differences in FA-weighted connectivity between network nodes (18, 31). Initially, two-sample t-tests were performed on each edge of the intergroup brain networks to generate a t-value map. An edge-level threshold of p < 0.01 was used to identify candidate connections. The significance of any interconnected suprathreshold components was then assessed via permutation testing (5,000 iterations), specifically by determining how often the size of the largest component in permuted data exceeded that in the actual data. FWER correction was applied, and subnetworks with corrected p < 0.05 were considered significant. Network topological properties Analysis of covariance (ANCOVA) was conducted to assess group differences in both global and nodal network measures across the SZ-MS, SZ-nMS, and HC groups, controlling for covariates. Bonferroni correction was applied to adjust for multiple comparisons of nodal metrics. Post hoc pairwise contrasts were subsequently performed via Fisher’s least significant difference (LSD) test. Correlation analysis Within the patient cohort, Pearson correlation coefficients (normally distributed) or Spearman rank correlations (skewed) were computed to explore the associations between altered FA-weighted connectivity or network topology metrics and metabolic indicators of MS. Bonferroni correction was applied for multiple comparisons (α = 0.05/20/10), with a corrected p value < 0.00025 considered statistically significant. Mediation model analysis Mediation analysis was performed using the SPSS PROCESS macro (v 4.2) (32). The independent variable (X) represented MS comorbidity status (SZ-MS vs. SZ-nMS), the mediator (M) corresponded to WM network nodes or connections exhibiting significant group differences, and the dependent variable (Y) was the metabolic parameter that differed significantly between groups. The significance of the mediation (indirect) effect was established through a nonparametric bootstrap procedure (5,000 resamples), which produced 95% bias-corrected confidence intervals. The statistical significance of the mediation effect was determined by whether the bootstrap confidence interval excluded zero. Results Participant characteristics The demographic and clinical characteristics of the study sample are presented in Table 1. No significant group differences were observed in age, sex, education, LDL-C, or HDL-C among the SZ-MS, SZ-nMS, and HC groups. Likewise, the two groups did not differ significantly in terms of illness duration, CPZ equivalent, TESS, or RSESE scores (p > 0.05). Table 1Demographic information and clinical characteristics of the participants.CharacteristicsSZ-MS group(n=26)SZ-nMS group(n=22)HC group(n=24)Test statistic (F/t/U/H/χ²)p valuePost hoc analysisSZ-MS vs. SZ-nMSSZ-MS vs. HCSZ-nMS vs. HCMale/female13/1315/79/154.3570.113———Age (years)41.69 ± 10.3438.18 ± 10.0040.38 ± 9.770.7330.484———Education (years)7 (6, 9)9 (6, 10)9 (7.50, 11.50)3.1500.207———Illness duration (years)9.81 ± 5.8611.68 ± 6.62—-1.0390.304———CPZ equivalent (mg/day)219.55 (187.50, 409.37)387.5 (228.12, 475)—207.0000.100———BMI (kg/m2)25.24 ± 2.8520.89 ± 2.4722.23 ± 2.7516.501<0.001***<0.001***<0.001***0.292LDL-C (mmol/L)3.01 ± 0.752.69 ± 0.552.55 ± 0.872.6820.076———HDL-C (mmol/L)1.26 ± 0.571.57 ± 0.551.52 (0.90, 3.16)4.8340.089———TC (mmol/L)4.68 ± 1.043.48 (2.91, 4.63)4.45 (2.00, 5.06)8.2310.016*0.019*0.1201.000TG (mmol/L)2.19 (1.91, 2.61)1.58 (1.32, 2.01)1.55 (1.38, 2.39)13.5720.001**0.002**0.013*1.000FBG (mmol/L)5.80 (5.07, 6.45)4.85 (4.37, 5.40)—151.5000.005**———Waist circumference (cm)91.01 ± 5.2778.72 ± 6.79—7.056<0.001***———Hip circumference (cm)96.44 ± 3.8089.81 ± 4.91—5.260<0.001***———Systolic blood pressure125.31 ± 18.49111.95 ± 11.57—2.9330.005**———Diastolic blood pressure82.81 ± 11.3375.55 ± 8.78—2.4450.018*———PANSS total score53 ± 12.4255.82 ± 14.634 (33, 35)47.561<0.001***1.000<0.001***<0.001***RSESE0.5 (0, 5)0 (0, 4)—202.5000.081———TESS4 (2.75, 9.25)2.5 (1.5, 7.5)—246.0000.361———BACS total score136.12 ± 53.06120.86 ± 53.59237.08 ± 45.1536.757<0.001***0.909<0.001***<0.001***Working memory14 ± 8.2510 (3, 17)25 (23.25, 27)33.887<0.001***0.558<0.001***<0.001***Verbal memory18.5 ± 9.4514.14 ± 9.4240.54 ± 13.4739.493<0.001***0.520<0.001***<0.001***Verbal fluency17.42 ± 8.0214.45 ± 7.0127.54 ± 8.0118.535<0.001***0.567<0.001***<0.001***Movement speed66 (37, 76)67 (43, 74.5)82 (78, 88)20.392<0.001***1.000<0.001***<0.001***Executive function 9.5 ± 6.778.45 ± 6.0519 (15, 20.75)24.752<0.001***1.000<0.001***<0.001***Attention19 (7.5, 34)18.5 (3.75, 25.5)46.21 ± 19.7827.473<0.001***1.000<0.001***<0.001***“—” not applicable; *p < 0.05, **p < 0.01, ***p < 0.001.Data are presented as mean ± SD for normally distributed variables or median (interquartile range) for skewed variables. Statistical tests performed: group differences in age, BMI, LDL-C, BACS total score, verbal memory, and verbal fluency, one-way ANOVA test and post hoc test; gender, chi-square test; education, PANSS total score, HDL-C, TC, TG, working memory, attention, executive function, and movement speed, Kruskal-Wallis test and post hoc test; illness duration, waist circumference, hip circumference, systolic blood pressure, and diastolic blood pressure, independent sample t-test; chlorpromazine equivalent, TESS, RSESE, and FBG, Mann-Whitney U test. Demographic information and clinical characteristics of the participants. “—” not applicable; *p < 0.05, **p < 0.01, ***p < 0.001. Data are presented as mean ± SD for normally distributed variables or median (interquartile range) for skewed variables. Statistical tests performed: group differences in age, BMI, LDL-C, BACS total score, verbal memory, and verbal fluency, one-way ANOVA test and post hoc test; gender, chi-square test; education, PANSS total score, HDL-C, TC, TG, working memory, attention, executive function, and movement speed, Kruskal-Wallis test and post hoc test; illness duration, waist circumference, hip circumference, systolic blood pressure, and diastolic blood pressure, independent sample t-test; chlorpromazine equivalent, TESS, RSESE, and FBG, Mann-Whitney U test. Both SZ-MS and SZ-nMS patients exhibited significantly lower total BACS and higher PANSS scores compared with the HC group (p < 0.001), with no difference between the two patient groups. BMI and TG levels were significantly elevated in the SZ-MS group relative to both HC and SZ-nMS participants (p < 0.05). Additionally, TC was significantly higher in SZ-MS group compared with SZ-nMS group (p < 0.05). The SZ-MS group also demonstrated greater FBG, blood pressure, waist circumference, and hip circumference than the SZ-nMS group (p < 0.05). White matter FA-weighted connectivity NBS analysis revealed distinct patterns of reduced FA-weighted connectivity among three groups (Figure 1). Compared with HC, the SZ-MS group exhibited significantly decreased FA-weighted connectivity encompassing 16 nodes and 15 edges (p < 0.001), spanning both cortical and subcortical (thalamic) regions. The SZ-nMS group also showed a diminished subnetwork relative to HC, involving six nodes and five edges (p < 0.05), primarily located in the right postcentral gyrus, left inferior frontal gyrus (triangular part), left superior parietal gyrus, left hippocampus, left thalamus, and left superior temporal gyrus. Compared with SZ-nMS patients, the SZ-MS group demonstrated further reductions in FA-weighted connectivity comprising five nodes and four edges (p < 0.05). These alterations predominantly involved the left dorsolateral superior frontal gyrus, left inferior frontal gyrus (triangular and orbital parts), left medial superior frontal gyrus, and left supplementary motor area. Differences in white matter FA-weighted connectivity among the SZ-MS, SZ-nMS, and HC groups. The blue curves indicate the lower FA values between the two regions. HC, healthy controls; SZ-MS, schizophrenia patients with comorbid metabolic syndrome; SZ-nMS, schizophrenia patients without metabolic syndrome. Global topological properties ANCOVA results indicated significant group differences in global network topology (Figure 2). Post hoc analyses revealed that, relative to HC, both schizophrenia groups displayed increased Gamma and Sigma values but significantly decreased Eg (p < 0.05). Moreover, the SZ-MS group demonstrated elevated Lp and Lambda as well as reduced Eloc compared with HC (p < 0.001 and p < 0.05, respectively). A comparison of the two patient groups indicated that SZ-MS individuals exhibited higher Lp, Gamma, Lambda, and Sigma, along with lower Eg and Eloc than SZ-nMS patients (p < 0.05). Differences in global topological properties among the SZ-MS, SZ-nMS, and HC groups. *p < 0.05, ***p < 0.001. HC, healthy controls; SZ-MS, schizophrenia patients with comorbid metabolic syndrome; SZ-nMS, schizophrenia patients without metabolic syndrome; Gamma, normalized clustering coefficient; Lp, characteristic path length; Lambda, normalized characteristic path length; Sigma, small-worldness; Cp, clustering coefficient; Eg, global efficiency; Eloc, local efficiency. Topological properties of nodes ANCOVA with Bonferroni correction revealed significant group differences in nodal-level metrics across multiple regions (Figure 3). Post hoc analyses showed that, compared with HC, both schizophrenia groups exhibited reduced DC and NE, along with increased NLP in several areas including the supramarginal gyrus, left medial superior frontal gyrus, left precuneus, right calcarine fissure, right superior temporal gyrus, and bilateral thalamus (p < 0.05). In addition, SZ-MS patients demonstrated further nodal disruption in the prefrontal cortex, left postcentral gyrus, and right superior parietal gyrus (p < 0.05). Relative to SZ-nMS patients, the SZ-MS group exhibited more pronounced nodal impairments primarily in the left inferior frontal gyrus (triangular and orbital parts), left medial superior frontal gyrus, right dorsolateral superior frontal gyrus, right superior temporal gyrus, right superior parietal gyrus, and postcentral gyrus (p < 0.05). Differences in nodal network metrics (a) degree centrality, (b) efficiency, and (c) shortest path length among the SZ-MS, SZ-nMS, and HC groups. HC, healthy controls; SZ-MS, schizophrenia patients with comorbid metabolic syndrome; SZ-nMS, schizophrenia patients without metabolic syndrome; R, right; L, left; IFGtriang, the triangular part of the inferior frontal gyrus; ORBinf, the orbital inferior frontal gyrus; SFGdor, the dorsolateral superior frontal gyrus; SFGmed, the medial superior frontal gyrus; CAL, the calcarine fissure and surrounding cortex; PoCG, the postcentral gyrus; SPG, the superior parietal gyrus; SMG, the supramarginal gyrus; PCUN, the precuneus; THA, the thalamus; STG, the superior temporal gyrus. Correlation between white matter network alterations and MS-related indicators In the uncorrected analyses, several significant associations were observed between altered WM metrics and metabolic parameters within the schizophrenia cohort (Table 2). TG levels were inversely correlated with FA-weighted connectivity between the left dorsolateral superior frontal gyrus and the left inferior frontal gyrus (triangular part) (r = −0.395). Reduced NE in the right dorsolateral superior frontal gyrus, left inferior frontal gyrus (triangular part), and right superior temporal gyrus correlated negatively with BMI (r = −0.311, −0.330, and −0.435, respectively) and waist circumference (r = −0.361, −0.337, and −0.372, respectively). Decreased NE in the left inferior frontal gyrus (triangular part) also correlated with hip circumference (r = −0.381), while reduced NE in the left medial superior frontal gyrus correlated with FBG (r = −0.351). Conversely, increased NLP in the right dorsolateral superior frontal gyrus, left inferior frontal gyrus (triangular part), and right superior temporal gyrus showed positive correlations with BMI (r = 0.332, 0.349, 0.424, respectively) and waist circumference (r = 0.357, 0.369, 0.476, respectively). Elevated NLP in the left inferior frontal gyrus (triangular part) was also associated with hip circumference (r = 0.380), and higher NLP in the left medial superior frontal gyrus correlated positively with FBG (r = 0.351). No other significant associations between metabolic variables and brain regions were detected at the uncorrected level. However, after applying Bonferroni correction for multiple comparisons, none of these associations remained statistically significant (all corrected p > 0.05). Table 2Correlations between alterations in white matter networks and TG, BMI, waist circumference, hip circumference, or FBG.Network metricsTGBMIWaist circumferenceHip circumferenceFBGr valuep valuer valuep valuer valuep valuer valuep valuer valuep valueFA-weighted connectivitySFGdor.L - IFGtriang.L-0.3950.005**————————Nodal efficiencyIFGtriang.L——-0.3300.005**-0.3370.019*-0.3810.008**——SFGdor.R——-0.3110.032*-0.3610.012*————STG.R——-0.4350.002**-0.3720.009**————SFGmed.L————————-0.3510.014*Nodal shortest path lengthIFGtriang.L——0.3490.015*0.3690.010**0.3800.008**——SFGdor.R——0.3320.021*0.3570.013*————STG.R——0.4240.003**0.4760.001**————SFGmed.L————————0.3510.014*“—” not applicable; *p < 0.05, **p < 0.01.R, right; L, left; IFGtriang, the triangular part of the inferior frontal gyrus; SFGdor, the dorsolateral superior frontal gyrus; STG, the superior temporal gyrus; SFGmed, the medial superior frontal gyrus. Correlations between alterations in white matter networks and TG, BMI, waist circumference, hip circumference, or FBG. “—” not applicable; *p < 0.05, **p < 0.01. R, right; L, left; IFGtriang, the triangular part of the inferior frontal gyrus; SFGdor, the dorsolateral superior frontal gyrus; STG, the superior temporal gyrus; SFGmed, the medial superior frontal gyrus. White matter networks as mediators in SZ-MS and metabolic abnormalities Mediation analysis revealed that NE in the right superior temporal gyrus showed a significant indirect effect in the relationship between SZ-MS and BMI (Figure 4). The total effect of MS comorbidity status (SZ-MS vs. SZ-nMS) on BMI was significant (β = 1.537, p < 0.001). After introducing right superior temporal gyrus NE as a mediator, the direct effect of MS comorbidity status on BMI became non-significant (β = 1.045, p = 0.100), whereas the indirect pathway via NE remained statistically significant (β = 0.581, BootSE = 0.293, 95% BootCI [0.025, 1.184]). When the same mediation analysis was performed with other significantly different brain regions as mediators, no statistically significant indirect effects were observed. Exploratory mediation pathway of node efficiency in the right superior temporal gyrus linking metabolic syndrome comorbidity status to BMI. *p < 0.05, ***p < 0.001. c, total effect of MS comorbidity status (SZ-MS vs. SZ-nMS) on BMI; c’, direct effect of MS comorbidity status on BMI after introducing the mediator; a, effect of MS comorbidity status on nodal efficiency of the right superior temporal gyrus; b, effect of nodal efficiency in the right superior temporal gyrus on BMI; STG.R, the right superior temporal gyrus; MS, metabolic syndrome. Discussion This study integrated NBS and graph-theoretical analyses to examine WM structural network alterations in SZ-MS patients after antipsychotic exposure, as well as to explore the neural mechanisms linking schizophrenia with metabolic dysfunction. At the edge level, both schizophrenia groups exhibited diffuse reductions in FA-weighted connectivity relative to HC. Compared with the SZ-nMS subgroup, patients with SZ-MS displayed additional reductions in prefrontal FA-weighted connectivity. At the topological level, SZ-MS patients demonstrated more extensive alterations in nodal metrics, particularly within the prefrontal cortex and right superior temporal gyrus. These WM disruptions correlated with TG, waist circumference, hip circumference, BMI, and FBG. In addition, exploratory mediation analysis indicated that reduced NE in the right superior temporal gyrus was statistically consistent with a mediating role between SZ-MS and BMI. By employing FA-weighted edge analysis, these findings further extend the prior FN-based graph-theoretical study. Collectively, they provide complementary neuroimaging evidence for WM network alterations in SZ-MS and offer new insights into the neuroimaging mechanisms underlying the comorbidity of schizophrenia and metabolic abnormalities. At the connectivity level, both schizophrenia groups showed markedly reduced FA-weighted connectivity linking the prefrontal, parietal, temporal, and thalamic regions, as well as connections between the parietal cortex and hippocampus, when compared with HC. This pattern is consistent with prior work showing widespread WM microstructural deficits in schizophrenia (33–35), which are believed to underlie impairments in higher cognitive and affective functions (34, 36, 37). Notably, compared with SZ-nMS patients, the SZ-MS group showed further reductions in connectivity involving the prefrontal cortex and supplementary motor area, which are key components of the default mode network and frontoparietal control network. These systems are essential for decision-making, executive control, and cognitive flexibility (38–40). Functional MRI studies have reported similar connectivity disruptions within these networks in patients with MS (41, 42). Moreover, individuals with abdominal obesity demonstrate impaired frontoparietal connectivity (43) and reduced coupling in the supplementary motor area (44). Because abdominal obesity is a major contributor to the onset and progression of MS (45, 46), these findings suggest that metabolic disturbances may exacerbate schizophrenia-related brain network disorganization by further compromising prefrontal WM microstructural integrity. At the global network level, schizophrenia patients exhibited decreased integration and efficiency of WM networks relative to controls, a finding consistent with prior reports (16, 47, 48). The brain’s topological organization reflects its capacity for both local specialization and global information integration (49, 50). Interestingly, the SZ-MS group displayed significantly higher clustering coefficients and small-worldness indices than both SZ-nMS and HC groups, indicating an excessively clustered yet inefficient network organization. This pattern could reflect the impact of MS-related chronic metabolic stress, such as systemic inflammation and oxidative damage, impairs WM networks, ultimately impairing global communication efficiency (51, 52). Consequently, information transmission across distributed brain regions may be less efficient in SZ-MS patients. At the nodal level, compared to the SZ-nMS group, the SZ-MS group demonstrated lower DC and NE, and higher NLP values within the dorsolateral and medial superior frontal gyri, as well as the left inferior frontal gyrus (triangular and orbital parts). The prefrontal cortex serves as a central hub for executive control, impulse regulation, and decision-making (53, 54). Damage in these regions may lead to disinhibited eating behavior and a preference for high-calorie foods, promoting obesity, insulin resistance, and, ultimately, the development of MS (55, 56). Wu and colleagues previously reported alterations in the medial prefrontal cortex of SZ-MS patients (16). In contrast, our study mainly observed nodal disruptions in the dorsolateral and medial superior frontal gyri, as well as in left inferior frontal gyrus (triangular part). This discrepancy may reflect methodological differences, as FN reflects fiber tract quantity, whereas FA is more sensitive to microstructural tissue properties. Nonetheless, both studies suggest that prefrontal network topological disruptions may play an important role in schizophrenia with comorbid MS. Supporting this, de Nijs et al. reported reduced gray matter volumes in reward-related circuits encompassing the orbitofrontal and medial prefrontal cortices (10), while Chen et al. observed decreased regional homogeneity in the inferior orbitofrontal gyrus in SZ-MS relative to SZ-nMS patients (12). Together, these findings indicate that both structural and functional disturbances within prefrontal regions, which are key nodes for reward processing and self-regulation, may constitute a shared neuropathological substrate linking schizophrenia and metabolic dysregulation. In uncorrected exploratory analyses, elevated TG levels were linked to diminished FA-weighted connectivity between the dorsolateral superior frontal gyrus and the inferior frontal gyrus (triangular part) in the patient cohort. High TG levels have been shown to induce endothelial dysfunction in patients with MS (57), a process closely associated with cerebral tissue injury. Previous research has reported that schizophrenia patients with hypertriglyceridemia or central obesity exhibit abnormal WM microstructure (58). Consistent with these findings, our results suggest that elevated TG in SZ-MS patients may be associated with microstructural alterations in prefrontal circuits involved in executive function. Moreover, disruptions within prefrontal and right superior temporal gyrus nodes were correlated with higher BMI, larger waist and hip circumferences, and increased fasting blood glucose levels. These regions represent critical hubs within the neural networks supporting higher-order cognitive processes (59, 60). It is plausible that these metabolic disturbances may affect information processing efficiency in these regions, thereby influencing the regulation of eating behavior in schizophrenia patients. A notable exploratory finding of this study is that node efficiency in the right superior temporal gyrus was statistically consistent with a mediating role in the relationship between SZ-MS and increased BMI. However, given the cross-sectional design, this mediation analysis does not demonstrate temporal precedence or causality. It remains equally plausible that metabolic changes influence WM microstructure, or that antipsychotic exposure drives both phenomena. Therefore, this finding should be interpreted as hypothesis-generating rather than as evidence of a mechanistic pathway. Although this region is classically associated with auditory perception and language comprehension (59, 61), accumulating evidence indicates its involvement in broader neural circuits related to obesity (62). Through its anatomical and functional connections with the insula, prefrontal cortex, and hypothalamus, the superior temporal gyrus contributes to the interoceptive perception of hunger and satiety (63, 64). Consequently, reduced NE in this area may reflect network dysfunction that disrupts energy homeostasis and metabolic regulation. Interestingly, while the prefrontal WM network exhibited the most pronounced alterations in the SZ-MS group, mediation analysis did not identify it as a significant intermediary between schizophrenia and metabolic indices. This suggests that the prefrontal and superior temporal networks may modulate metabolic function through distinct neurobiological mechanisms. The reasons underlying this dissociation remain uncertain and represent an important direction for future research. This research is subject to several limitations. First, it was a preliminary exploratory investigation with a relatively small sample size. Although a priori sample size estimation was performed based on the effect size, the relatively small sample size may limit statistical power and estimation stability for graph-theoretical and mediation analyses. Replication in larger, multicenter, longitudinal cohorts is warranted. Second, although metabolic indicators were systematically assessed in schizophrenia patients, the influence of antipsychotic type, dosage, and treatment duration on brain structural networks could not be fully controlled. The present study primarily included patients receiving clozapine or risperidone, which reduces variability due to mixed pharmacotherapy but precludes conclusions regarding the effects of other antipsychotic agents. In addition, only partial lipid profiles were collected in the HC group, and comprehensive metabolic parameters were not assessed. Therefore, direct comparisons of metabolic profiles between the HC and patient groups were not feasible. Future studies should systematically collect complete metabolic data in control groups to more comprehensively control for confounding factors affecting WM networks. Third, given the cross-sectional design, the causal relationships between MS and WM alterations cannot be determined. Therefore, the mediating effects reported herein should be interpreted as exploratory findings rather than evidence of causal mechanisms. Furthermore, the cross-sectional design also limits the ability to track dynamic changes in metabolic parameters and brain network metrics during antipsychotic treatment. Future studies combining longitudinal follow-up with causal mediation analysis are warranted to further elucidate the temporal relationships and underlying mechanisms linking metabolic abnormalities to brain structural and functional impairments in schizophrenia patients. Fourth, Bonferroni correction was applied to the correlation analyses to control for multiple comparisons. However, none of the associations remained statistically significant after correction. Therefore, these correlation results should be interpreted as exploratory findings. Fifth, our DTI protocol used a single-shell acquisition with only 30 diffusion-encoding directions, which offers limited sensitivity to crossing fibers. This methodological constraint may affect tractography performance in regions with complex WM architecture. Future studies employing multi-shell acquisitions with higher angular resolution or probabilistic tractography would provide more robust characterization of white matter network organization. Conclusion In summary, this integrated analysis using network-based statistics and graph-theoretical modeling identified characteristic disruptions in both FA-weighted connectivity and network topology of WM in schizophrenia patients with metabolic syndrome. These abnormalities predominantly involve the prefrontal cortex and right superior temporal gyrus and are associated with metabolic dysfunction. The findings provide neuroimaging evidence that may elucidate the neurobiological mechanisms linking schizophrenia and metabolic syndrome.