Introduction
Approximately 60%–80% of patients with schizophrenia experience auditory verbal hallucinations (AVH), and about 25% of these patients do not respond to adequate doses and durations of antipsychotic treatment. This condition is referred to as persistent AVH (p-AVH). AVH is often characterized as harsh, commanding, aggressive, and typically conveys negative and malicious content imbued with intense negative emotions. Its chronic presence can lead to high rates of suicide and self-harm, significantly impairing quality of life while imposing substantial burdens on families and society. Therefore, it is crucial to explore the underlying neurobiological mechanisms of p-AVH and develop effective treatment strategies accordingly.
Hypotheses on the possible mechanisms of AVH
Despite extensive research efforts to uncover the underlying neuropathological mechanisms of AVH, no unified theory has been established to date. Existing hypotheses include hyperactivation of the auditory cortex, errors in inner speech generation, misattribution of internal events, and disruptions in episodic memory retrieval, involving a range of brain regions (Mo et al., Reference Mo, Zhao, Li, Cai, Song, Wang, Yu and Zhu2024). Magnetic resonance imaging (MRI) studies have demonstrated structural and functional brain alterations associated with language perception and self-monitoring, such as the prefrontal and temporal cortices and subcortical areas (e.g. superior temporal gyrus, middle temporal gyrus, inferior frontal gyrus, and anterior insula), in schizophrenia patients with AVH (Pinheiro, Schwartze, & Kotz, Reference Pinheiro, Schwartze and Kotz2020). For instance, structural MRI studies have found that changes of the gray matter volume (GMV) in the superior and middle temporal gyri are correlated with the severity of hallucinations (Modinos et al., Reference Modinos, Costafreda, van Tol, McGuire, Aleman and Allen2013; Onitsuka et al., Reference Onitsuka, Shenton, Salisbury, Dickey, Kasai, Toner, Frumin, Kikinis, Jolesz and McCarley2004). Moreover, functional MRI (fMRI) has revealed that abnormal activity in the frontotemporal network (e.g. Broca’s area, anterior insula, superior temporal gyrus, middle temporal gyrus, parietal lobe, and precentral gyrus) is closely associated with the onset of AVH (Jardri, Pouchet, Pins, & Thomas, Reference Jardri, Pouchet, Pins and Thomas2011). Similarly, visual hallucinations have also been localized to distinct brain networks (Li et al., Reference Li, Mo, Liu, Sun, Zhu, Yu and Zhu2026). However, studies of transcranial magnetic stimulation (TMS) targeting these AVH-related brain regions have not clarified the relationship between AVH improvement and the functional changes of these regions, and the efficacy of TMS treatment remains controversial (Hoffman et al., Reference Hoffman, Gueorguieva, Hawkins, Varanko, Boutros, Wu, Carroll and Krystal2005). For instance, low-frequency repetitive TMS (rTMS) targeting the left temporoparietal junction (TPJ) is commonly used, but randomized controlled trials have not consistently shown significant therapeutic benefits for AVH symptoms (Slotema et al., Reference Slotema, Blom, de Weijer, Diederen, Goekoop, Looijestijn, Daalman, Rijkaart, Kahn, Hoek and Sommer2011). This may be attributed to sample heterogeneity, as some patients may experience p-AVH (Xie et al., Reference Xie, Guan, Wang, Ma, Wang and Fang2023).
The crucial role of the cerebellum in p-AVH
Additionally, lesion network mapping studies have shown that damage to various brain regions can be linked through their functional connectivity (FC) to common brain networks. Approximately 90% of lesion sites exhibit positive FC with the cerebellar vermis VI and cerebellar lobule X, and negative FC with the right superior temporal gyrus. Moreover, 95% of lesion sites show FC with the cerebellar dentate nucleus (Kim et al., Reference Kim, Hsu, Talmasov, Joutsa, Soussand, Wu, Rost, Morenas-Rodríguez, Martí-Fàbregas, Pascual-Leone, Corlett and Fox2021). This suggests that the neuropathological mechanisms of AVH may involve changes in neural circuits related to language generation, monitoring, and perception, with the cerebellum playing a key role in language processing and perception (Allen et al., Reference Allen, Modinos, Hubl, Shields, Cachia, Jardri, Thomas, Woodward, Shotbolt, Plaze and Hoffman2012; Filippi et al., Reference Filippi, Richardson, Dick, Leech, Green, Thomas and Price2011). Studies have found that schizophrenia patients with p-AVH show a significant reduction in GMV in the cerebellar region VIIIa compared to both ordinary AVH patients and healthy controls, and this change correlates with AVH symptoms (Cierpka et al., Reference Cierpka, Wolf, Kubera, Schmitgen, Vasic, Frasch and Wolf2017). The cerebellum is not only an important node in the p-AVH network but also appears to differ from the ordinary AVH network in terms of the neural networks involved. The cerebellum dynamically updates internal models through error-driven learning mechanisms to predict behavior and cognitive functions, which is closely linked to the occurrence of p-AVH. When errors in inferring internal events are over-weighted, it may lead to the generation of hallucinations. Furthermore, abnormalities in the cerebellar vermis have been found to be related to hallucination triggers (Okugawa, Sedvall, & Agartz, Reference Okugawa, Sedvall and Agartz2003).
The cerebellum participates in auditory processing and the exchange of information with the cerebral cortex through complex anatomical connections (Andreasen & Pierson, Reference Andreasen and Pierson2008; Ramnani, Reference Ramnani2006). The spinocerebellar tract transmits auditory information to the cerebral cortex via deep nuclei, while the cerebrocerebellar pathway relays information from the cortex, particularly from the parietal lobe, through the pons and thalamus. The dentate nucleus, as the primary output nucleus of the cerebellum, transmits information to the cerebral cortex through the thalamus. Abnormalities in these pathways may constitute the neuropathological mechanism of p-AVH.
The cortical–cerebellar–subcortical–cortical circuit and the neuropathological mechanism of p-AVH
Research indicates that the cortical–cerebellar–subcortical–cortical circuit plays an important role in sensory, motor, and higher cognitive activities. The close connection between the cerebellar Crus I and Crus II regions and the frontal cortex has been further confirmed in monkey animal experiments (Habas et al., Reference Habas, Kamdar, Nguyen, Prater, Beckmann, Menon and Greicius2009). The prefrontal and parietal cortices are interconnected with Crus I and Crus II of the cerebellum through the pontine fibers, and signals are transmitted back to the cortical regions via the cerebellar dentate nucleus (Ramnani, Reference Ramnani2006). This bidirectional feedback mechanism plays a critical role in sensory processing and motor information updating. However, in patients with AVH, errors in the feedback information updating process may lead to the onset of p-AVH. Dynamic fMRI studies have shown that the left cerebellum is activated before the onset of AVH, suggesting that abnormal processing of auditory input in the cerebellum is one of the neuropathological bases of p-AVH. Furthermore, impairments in Purkinje cell synaptic plasticity, such as the inability to effectively modulate input strength through climbing fibers and parallel fibers, further support the central role of the cerebellum in p-AVH.
Research hypothesis and study content
Traditional rTMS targets the left TPJ for the treatment of ordinary AVH (Slotema et al., Reference Slotema, Blom, van Lutterveld, Hoek and Sommer2014), with the proposed mechanism being the regulation of functional abnormalities in the temporoparietal circuit. Continuous theta burst stimulation (cTBS) can more rapidly suppress neural network activity. Studies have shown that cTBS at the left TPJ not only improves AVH symptoms but also causes significant changes in the FC intensity of the left cerebellum, with the improvement in FC intensity negatively correlating with symptom relief (Chen et al., Reference Chen, Ji, Zhu, Bai, Wang, He, Gao, Tao, Yu, Tian and Wang2019). Moreover, based on the dynamic activation patterns and synaptic plasticity characteristics of the left cerebellum, it has been further proposed that the cerebellum could be a target for cTBS treatment of p-AVH. This study hypothesizes that the neuropathological mechanism of p-AVH involves functional abnormalities in the cortical–cerebellar–subcortical–cortical circuit, and that cTBS-assisted therapy improves p-AVH symptoms by modulating the abnormal FC of this loop. The study aims to: (1). Investigate whether cTBS-assisted treatment targeting cerebellum can significantly improve p-AVH symptoms; (2). Explore whether changes in whole-brain FC after treatment can predict improvements in AVH symptoms; (3). Assess whether the FC associated with symptom improvement is primarily concentrated in the cortical–cerebellar–subcortical–cortical circuit.
Methods
Subjects
This study was approved by the Ethics Committee of the Second Xiangya Hospital, Central South University (ClinicalTrials.gov Identifier: NCT05039489). All schizophrenia patients and their families who were included in the study signed an informed consent form.
Schizophrenia patients with p-AVH group: The diagnosis was made by two experienced psychiatrists (above the level of attending physician) based on the Diagnostic and Statistical Manual of Mental Disorders, Fifth Edition (DSM-5) diagnostic criteria for schizophrenia. The specific inclusion criteria were as follows: (1) age 18–45 years, right-handed; (2) meeting the DSM-5 criteria for schizophrenia; (3) after being treated with two or more antipsychotic drugs at adequate doses for at least 8 weeks (with at least 1 second-generation atypical antipsychotic), experiencing auditory hallucinations more than five times a day, having a clinical history of persistent auditory hallucinations for at least 8 weeks prior to enrollment, and a PANSS P3 item score >3 at the screening visit; (4) having stable symptoms for at least 4 weeks before inclusion (except for auditory hallucinations), with the medication regimen unchanged during the study. Exclusion criteria: (1) history of epilepsy, seizures, stroke, or other serious organic brain diseases; (2) history of severe infectious diseases, malignant tumors, etc.; (3) history of intellectual disability, personality disorders, etc.; (4) contraindications for MRI (e.g. presence of nonremovable metal objects in the body); (5) history of marijuana or other psychoactive substance use; history of alcohol abuse or benzodiazepine or antiepileptic drug use; (6) pregnancy; (7) intellectual disability (IQ < 70); (8) received modified electroconvulsive therapy (MECT) or TMS treatment in the past 6 months.
Data of subjects
General data collection and assessment of psychiatric symptoms
Basic personal information of the subjects, such as gender, age, years of education, marital status, duration of illness, types and doses of antipsychotic medications (converted to olanzapine equivalent doses; Leucht, Samara, Heres, & Davis, Reference Leucht, Samara, Heres and Davis2016), etc., was collected, and routine physical and psychiatric examinations were conducted. Clinical data were assessed and collected by two experienced psychiatrists who had undergone scale consistency training. The Positive and Negative Syndrome Scale (PANSS) was used to assess the patients’ positive and negative symptoms, with a four-factor symptom dimension rating. The Clinical Global Impressions (CGI) scale was used to assess the improvement of symptoms. The Treatment Emergent Symptom Scale (TESS) was used to evaluate medication side effects. The auditory hallucination subscale of the Psychotic Symptom Rating Scales (PSYRATS-AH) was used to assess the severity of auditory hallucinations in the patients. Patients’ cognitive function was assessed using the Repeatable Battery for the Assessment of Neuropsychological Status (RBANS) scale.
RS-fMRI data acquisition and preprocessing
All fMRI data were collected using a Siemens 3.0 T MRI scanner at the Radiology Department of the Second Xiangya Hospital, Central South University. After gathering general demographic and clinical symptom data from the participants, MRI scans were performed on the same day. Participants were instructed to lie still, close their eyes, and remain awake during the scan. Each participant underwent an MRI scan with the following parameters: repetition time/echo time = 2000 ms/30 ms, 30 slices, 64 × 64 matrix, 90° flip angle, 24 cm field of view, 4 mm slice thickness, 0.4 mm gap, and 240 volumes. Preprocessing of MRI data was conducted using the DPABI toolbox, including steps such as slice-timing correction, head motion correction, spatial normalization, smoothing, filtering, and linear drift removal. More details are in Supplementary Materials.
FC analysis
Whole-brain FC analysis was performed using the DPABI toolbox, with the Shen268 brain atlas as the template, which divides the brain into three parts: cortex, subcortex, and cerebellum, with a total of 268 nodes (Fan et al., Reference Fan, Li, Zhuo, Zhang, Wang, Chen, Yang, Chu, Xie, Laird, Fox, Eickhoff, Yu and Jiang2016). More details are in Supplementary Materials.
Intervention treatment with cTBS for schizophrenia patients with p-AVH
A 14-day cTBS intervention was conducted for schizophrenia patients with p-AVH. The cTBS protocol included continuous burst stimulation lasting 40 seconds, comprising 200 bursts of three 50 Hz pulses each, with a 200 ms interval between bursts, totaling 600 pulses per session. Stimulation intensity was set at 80–100% of the resting motor threshold, referred to the intensity used in previous cerebellar stimulation studies (Arasanz, Staines, Roy, & Schweizer, Reference Arasanz, Staines, Roy and Schweizer2012; Koch et al., Reference Koch, Mori, Marconi, Codecà, Pecchioli, Salerno, Torriero, Lo Gerfo, Mir, Oliveri and Caltagirone2008), to achieve potential effective neuromodulation while maintaining participant safety and treatment tolerability. Patients received three sessions daily, each spaced 15 min apart, delivering 1800 pulses per day. Targeted stimulation was manually localized to the left cerebellum, identified 3 cm lateral and 1 cm inferior to the occipital protuberance. Before the first session, an electroencephalogram (EEG) screening ensured safety. Daily treatment was followed by hallucination assessments using the PSYRATS-AH scale. After 14 days, clinical data were re-evaluated, and fMRI scan was performed. Follow-ups at weeks 4, 6, and 10 included clinical and cognitive assessments. In this study, if the interruption lasted 3 days or more, the subject was considered a dropout. A final fMRI scan was conducted at week 10. The study model diagram is shown in Figure 1.
The study model diagram. Note: cTBS, ‘continuous theta burst stimulation’; PSYRATS-AH, ‘the psychotic symptom rating scales’; RR, ‘reduction rate’; PANSS, ‘the positive and negative syndrome scale’; CGI, ‘the clinical global impressions scale’; RBANS, ‘the repeatable battery for the assessment of neuropsychological status’; TESS, ‘the treatment emergent symptom Scale’, MRI, ‘Magnetic resonance imaging’.
Figure 1. Long description
The diagram is organized into three main horizontal stages.
Top-left: Screen and baseline V 0. An icon of a head with a circular arrow represents Auditory Verbal Hallucinations. A blue box lists procedures: signed informed consent, inclusion or exclusion criteria, demography, symptom review using P A N S S, P S Y R A T S - A H, C G I, T E S S, and R B A N S, and an M R I scan.
Top-center: 14 days of c T B S treatment V 1 to V 2. An anatomical brain model shows the target area in the cerebellum. Below this, a detailed stimulation protocol shows three sessions labeled 1st, 2nd, and 3rd. Each session consists of 40 seconds or 600 pulses, with 15 minutes rest between sessions. A magnified view of the pulse train shows a frequency of 50 Hertz with 20 millisecond intervals between pulses within a burst, and bursts occurring every 200 milliseconds at 5 Hertz. A blue box indicates symptom review and M R I scan at V 2.
Top-right: 8 week follow-up after treatment V 3 to V 5. A head icon shows a smaller hallucination symbol. A blue box indicates symptom review and M R I scan at V 5.
Bottom: Data analysis. An arrow points from the treatment phase to three colored F C changes matrices labeled 268 times 268 times n. These lead to a scatter plot with a linear regression line. The y-axis is P S Y R A T S - A H score and the x-axis is F C. The equation y equals m x plus b is shown. A blue box at the bottom is labeled build predictive modeling.
Statistical analysis
Analysis of cTBS intervention results
One-way repeated measures ANOVA was used to analyze clinical data before and after cTBS treatment. Corrections were applied for violations of sphericity, and post hoc t-tests were adjusted using Bonferroni correction. Primary outcome: change in PSYRATS-AH from baseline (V0) to V1 (1 week), V2 (2 weeks), V3 (4 weeks), V4 (6 weeks), and V5 (10 weeks). Response was defined as ≥25% reduction rate of PSYRATS-AH (RR_PSYRATS-AH). Sample-size estimation indicated a minimum of 14 participants (α = 0.05, power = 0.9), with a 20% dropout allowance (target N ≥ 18). Secondary outcomes included changes in PANSS and RBANS scores, and FC alterations at V0, V2, and V5. Adverse events were monitored daily.
CPM analysis
Changes in whole-brain FC before and after treatment (baseline vs. post-cTBS) were used as neuroimaging features, with RR_PSYRATS-AH at 2 weeks post-treatment as the clinical feature. A prediction model was constructed using the connectome-based predictive modeling (CPM) method, with age and sex included as covariates. CPM analysis was performed using an open-source MATLAB script. Specifically, for each subject, the FC matrices (268 × 268 × n) were loaded, and a leave-one-out cross-validation (LOOCV) procedure was applied. In each iteration, one subject was left out as the test case, and the remaining subjects formed the training set. Within the training set, partial correlation analyses were conducted between each FC edge and RR_PSYRATS-AH scores, controlling for age, sex, and framewise displacement (FD) as covariates. Edges significantly correlated with symptom improvement (priori p = 0.001–0.01, step = 0.001) were selected to construct positive and negative feature sets. For each subject, the mean FC strength of these selected edges was computed to obtain individual network strength indices. Linear regression models were then fitted between network strengths and RR_PSYRATS-AH values to generate prediction equations (y = b + w pos·X pos + w neg·X neg). The trained model was applied to the left-out subject to predict clinical outcomes, and prediction accuracy was evaluated by correlating predicted and observed RR_PSYRATS-AH scores across all subjects. Statistical significance was assessed using non-parametric permutation tests with 5,000 iterations by randomly shuffling subject labels. The entire prediction pipeline was repeated from scratch within each iteration’s LOOCV loop. A null distribution of prediction performances (r perm) was thus generated. The empirical p value was derived by calculating the proportion of permutations in which the permuted prediction performance was equal to or better than the performance obtained using the actual, unshuffled data (r actual), formulated as p = ∑(r perm ≥ r actual)/5,000. The final CPM models were established to predict clinical improvement in auditory hallucinations following cTBS treatment.
FC analysis based on high-degree nodes
Nodes with the highest number of edges in the prediction model were defined as high-degree nodes. The connections of these nodes were analyzed.
Results
Participant characteristics
Thirty patients with schizophrenia and p-AVH completed baseline fMRI scans. Twenty-seven patients completed the 14-day cTBS treatment, and three dropped out due to COVID-19 infection or personal reasons. After excluding five subjects with excessive head motion, 22 patients were included in the final fMRI analysis. The mean age was 27.83 ± 3.05 years, mean illness duration was 5.97 ± 2.74 years, and mean PANSS total scores was 62.77 ± 11.14 at baseline. The detailed demographic information is presented in Table 1.
Baseline clinical characteristics of the schizophrenia patients with p-AVH
Table 1. Long description
The table is divided into two main sections. The first section lists Mean and S D for various clinical measures. Age is 27.83 years (S D 3.05). Education is 12.43 years (S D 2.06). Illness duration is 5.97 years (S D 2.74). Olanzapine equivalent doses are 26.52 mg (S D 10.12). M E P is 35.05 percent (S D 10.26). P S Y R A T S dash A H is 28.37 (S D 3.09). P 3 is 5.90 (S D 0.40). P A N S S total score is 62.77 (S D 11.14), with sub-scores for Negative (15.20), Positive (11.20), Emotional (15.57), and Cognitive (20.23) symptoms. R B A N S total score is 151.93 (S D 17.15), with sub-scores for Immediate memory (28.17), Visuospatial (14.97), Language (23.97), Attention (59.00), and Delayed memory (25.83).
The second section lists counts (N) and percentages. Gender distribution is 18 males (60.0 percent) and 12 females (40.0 percent). Medication types include Olanzapine plus paliperidone (6, 20.0 percent), Olanzapine plus aripiprazole (6, 20.0 percent), Olanzapine plus blonanserin (4, 13.3 percent), Amisulpride plus risperidone (4, 13.3 percent), Clozapine plus blonanserin (4, 13.3 percent), Clozapine plus aripiprazole (3, 10.0 percent), and Others (3, 10.0 percent).
Note: p-AVH, persistent auditory verbal hallucinations; PSYRATS-AH, the psychotic symptom rating scales; MEP, motor evoked potential; PANSS, the positive and negative syndrome scale; RBANS, the repeatable battery for the assessment of neuropsychological status.
a The figure copy and figure recall subtests were not included in the calculation of the RBANS scores.
Changes in auditory hallucination and PANSS scores
Repeated-measures analysis revealed significant reductions in PSYRATS-AH (F = 6.324, p = 0.001), RR_PSYRATS-AH (F = 6.871, p < 0.001), and PANSS P3 scores (F = 3.744, p = 0.011) following cTBS intervention, indicating a marked improvement in auditory hallucination severity. Total PANSS and subscale scores showed no significant changes. Post hoc tests confirmed that PSYRATS-AH, RR_PSYRATS-AH, and P3 scores at 2- and 4-weeks posttreatment were significantly lower than baseline, suggesting sustained symptom relief. Further analysis showed that, in the schizophrenia group with p-AVH, nine patients had an RR_PSYRATS-AH exceeding 25% at the end of the 2-week treatment, and six patients exceeded 25% at 2 weeks after treatment completion. However, after 2- to 4-week follow-up, the treatment effect seemed to be gradually attenuated. As shown in Figure 2a and Supplementary Table S1.
Clinical characteristics of p-AVH group after cTBS treatment at different follow-up time points. Note: p-AVH, ‘persistent auditory verbal hallucinations’; cTBS, ‘continuous theta burst stimulation’; PSYRATS-AH, ‘the psychotic symptom rating scales’; RR, ‘reduction rate’; PANSS, ‘the positive and negative syndrome scale’; RBANS, ‘the repeatable battery for the assessment of neuropsychological status’.
Figure 2. Long description
The figure contains eight line graphs with error bars.
Panel a contains four graphs.
1. Top-left: Y-axis P S Y R A T S-A H scores from 22 to 30. The line shows a sharp decrease from week 0 to week 2, followed by a gradual recovery through week 10. Asterisks indicate significant differences between week 0 and weeks 2 and 4.
2. Top-right: Y-axis R R underscore P S Y R A T S-A H from 0.00 to 0.25. The line peaks at week 2 and then declines. Asterisks mark significance at weeks 2 and 4 compared to baseline.
3. Bottom-left: Y-axis P 3 scores from 4.5 to 6.5. The trend mirrors the first graph with a trough at week 2 and significant differences at weeks 2 and 4.
4. Bottom-right: Y-axis R R underscore P S Y R A T S-A H from 0.00 to 1.25. This graph shows individual subject lines with a horizontal dotted line at 0.25.
Panel b contains four graphs focused on R B A N S metrics.
1. Top-left: Y-axis Attention scores from 56 to 64. Scores peak at week 2 with a significant difference from baseline.
2. Top-right: Y-axis Delayed memory scores from 24 to 30. Scores peak at week 2 with significant differences at weeks 2 and 6.
3. Bottom-left: Y-axis R B A N S total scores from 145 to 165. Scores peak at week 2 with significant differences at weeks 2, 6, and 10.
4. Bottom-right: Y-axis Language scores from 23 to 27. Scores peak at week 2 with significant differences at weeks 2 and 10.
Changes in cognitive function
Cognitive assessments demonstrated significant improvements in RBANS total score (F = 27.030, p < 0.001), verbal function (F = 15.010, p < 0.001), attention (F = 12.854, p < 0.001), and delayed memory (F = 14.013, p < 0.001) after cTBS treatment, whereas immediate memory (F = 0.693, p = 0.564) and visuospatial skills (F = 1.089, p = 0.371) remained unchanged. As shown in Figure 2b and Supplementary Table S1.
Adverse effects
During cTBS treatment, mild side effects such as scalp discomfort, headache (7/30), or dizziness (3/30) were reported, which resolved spontaneously within a few days. No severe adverse events, including seizures, occurred throughout the study. As shown in Supplementary Table S2.
CPM analysis
Changes in whole-brain FC between baseline and post-cTBS were used as neuroimaging features, and RR_PSYRATS-AH scores at 2 weeks posttreatment served as the clinical variable. The CPM approach was applied with age and sex as covariates. The CPM model significantly predicted clinical improvement (r = −0.269, p = 0.047, MSE = 0.064, and n = 22). The results were obtained using a primary feature selection threshold of p = 0.005, and were further confirmed to be robust across a range of thresholds. As shown in Figure 3a. The positive prediction network comprised 30 edges, and the negative network contained 39 edges. As shown in Figure 3b. The sensitivity of CPM was reported in the Supplementary Figures S1 and S2. Permutation-based null distribution of CPM prediction performance was reported in the Supplementary Figure S3.
The CPM model predicted the RR_PSYRATS-AH of the schizophrenia patients with p-AVH. (a) The CPM model significantly predicted RR_PSYRATS-AH (r = −0.269, p = 0.047, MSE = 0.064, n = 22). (b) The positive prediction network comprised 30 edges, and the negative network contained 39 edges of the predictive model. The positive features (edges) are coded red, and the negative features (edges) are coded blue. (c) FC Analysis based on high-degree nodes. Both positive and negative predictive networks highlighted bilateral cerebellum as key hubs. Note: CPM, ‘connectome-based predictive modeling’; P-AVH, ‘persistent auditory verbal hallucinations’; PSYRATS-AH, ‘the psychotic symptom rating scales’, RR, ‘reduction rate’, FC, ‘functional connectivity’.
Figure 3. Long description
Panel a is a scatter plot with a green linear regression line and a light green shaded confidence interval. The x-axis is R R _ P S Y R A T S - A H predictive scores and the y-axis is R R _ P S Y R A T S - A H actual scores. Data points show a negative correlation with r = -0.269 and p = 0.047.
Panel b contains three circular connectome diagrams labeled R and L. The first circle shows combined red and blue edges for nodes >= 1. The second circle shows only red positive edges for nodes >= 3. The third circle shows only blue negative edges for nodes >= 3. A color-coded legend on the right identifies brain regions: Prefrontal (red), Motor Strip (orange), Insula (yellow), Parietal (green), Temporal (blue), Occipital (pink), Limbic (cyan), Cerebellum (brown), Subcortical (dark orange), and Brainstem (purple).
Panel c displays anatomical brain maps in two rows. The top row shows Positive F C in high-degree nodes with red connections and nodes concentrated in the bilateral cerebellum. The bottom row shows Negative F C in high-degree nodes with blue connections and nodes also concentrated in the bilateral cerebellum. Each row features four views: superior, inferior, left lateral, and right lateral. Color bars indicate a scale from 0 to 5 for positive and 0 to 7 for negative connectivity.
FC Analysis based on high-degree nodes
Nodes with three or more connections were defined as high-degree nodes. Both positive and negative predictive networks highlighted bilateral cerebellum as key hubs, with more extensive connections in the left cerebellum. FC changes between the cerebellum and prefrontal (dorsolateral, orbitofrontal, and anterior regions), temporal (middle/superior temporal gyri and temporal pole), parietal, occipital, limbic (hippocampus and posterior cingulate cortex), and subcortical (basal ganglia) regions were significantly associated with RR_PSYRATS-AH. As shown in Figure 3c and Table 2, these results suggest that cerebellar–cortical FC alterations play a crucial role in mediating the therapeutic response to cTBS in patients with refractory auditory hallucinations.
CPM predictive networks having high-degree nodes associated with RR_PSYRATS-AH in patients with p-AVH
Table 2. Long description
The table is organized into columns for Symptom, Network, Node, and Connections. The Connections column is further subdivided into Prefrontal, Parietal, Temporal, Occipital, Limbic, Cerebellum, Subcortical, and Motor.
Row 1: Symptom is R R underscore P S Y R A T S dash A H. Network is Positive network. Node is R Cerebellum . Connections include:
- Prefrontal: R dlPFC, and L O F C .
- Limbic: L Hippocampus.
Row 2: Symptom is R R underscore P S Y R A T S dash A H. Network is Positive network. Node is L Cerebellum. Connections include:
- Prefrontal: L dlPFC, R O F C, R Broca’s B A 44, and R Broca’s B A 45.- Parietal: L PrimSensory.
Row 3: Symptom is R R underscore P S Y R A T S dash A H. Network is Negative network. Node is R Cerebellum. Connections include:
- Prefrontal: R AntPFC, R dlPFC (2), and R O F C.
- Temporal: L M T G .
- Occipital: L Visual Assoc.
- Cerebellum: L Cerebellum (2) .
- Subcortical: R B G.
- Motor: R B A 6.Row 4: Symptom is R R underscore P S Y R A T S dash A H. Network is Negative network. Node is L Cerebellum. Connections include:- Parietal: L Vis Motor.- Temporal: R Temporalpole, and R S T G.- Limbic: R Ventral P C C.- Cerebellum: R Cerebellum (2).- Subcortical: R B G.
Note: CPM, connectome-based predictive modeling; p-AVH, persistent auditory verbal hallucinations; PSYRATS-AH, the psychotic symptom rating scales; RR, reduction rate; PFC, prefrontal cortex; dlPFC, dorsolateral prefrontal cortex; OFC, orbitofrontal area; PrimSensory, primary somatosensory cortex; BA44(Broca’s area), opercular part of the inferior frontal gyrus; BA45(Broca’s area), triangular part of the inferior frontal gyrus; AntPFC, anterior prefrontal cortex; Visual Assoc, visual Association cortex; BG, basal ganglia; BA6, pre-motor and supplementary motor cortex; Temporalpole, temporopolar area; MTG, middle temporal gyrus; Vis Motor, visual motor coordination; STG, superior temporal gyrus; Ventral PCC, ventral posterior cingulate cortex.
Discussion
This study demonstrated that cTBS targeting the left cerebellum significantly reduced PSYRATS-AH and PANSS-P3 scores in schizophrenia patients with p-AVH, while also being associated with the improvement in cognitive performance. CPM analysis revealed that posttreatment changes in FC relative to baseline predicted symptom improvement, with bilateral cerebellum serving as key nodes in the predictive network, suggesting a potential central role of the cerebellum in modulating auditory hallucinations.
Role of the cerebellum in auditory hallucination circuits
cTBS may alleviate hallucinations by modulating FC within the cortical–cerebellar–subcortical–cortical circuit. Anatomically, Purkinje cells – the sole output neurons of the cerebellar cortex – use Gamma-aminobutyric acid (GABA) as an inhibitory neurotransmitter, forming feedforward circuits that contribute to internal speech generation (Michie, Reference Michie2001). Based on the forward model hypothesis, altered cerebellar–frontal connectivity is a characteristic feature of AVH (Chang et al., Reference Chang, Xi, Cui, Wang, Sun, Zhu, Huang, Collin, Liu, Xi, Qi, Tan, Miao and Yin2015). Structural or functional disruptions in the cerebellum may impair self-monitoring of voice ownership, resulting in hallucinations. During sensory–motor processing, the cerebellum continuously updates predictions through ‘efference copy’ and ‘prediction error’ mechanisms; dysfunction in this feedback system can generate misattributed self-speech (Pinheiro et al., Reference Pinheiro, Schwartze and Kotz2020). Consistent with this model, posttreatment FC alterations between the cerebellum and the prefrontal cortex (including Broca’s area), auditory cortex, basal ganglia, and within the cerebellum itself were associated with hallucination improvement.
Primate studies revealed bidirectional projections between BA4/BA46 and cerebellar Crus I/II (Kelly & Strick, Reference Kelly and Strick2003), supporting the choice of this stimulation target. Compared with monkeys, the human cerebellum exhibits denser connections with the prefrontal cortex, implying its role in processing abstract information such as language and cognition beyond motor control. This aligns with our finding that cerebellum–cortical FC patterns correlate with positive symptom improvement. Previous studies have reported reduced thalamus–cerebellar FC but increased effective connectivity from the left thalamus to the right inferior frontal gyrus in first-episode AVH patients (Wei et al., Reference Wei, Xue, Yang, Wang, Chen, Han, Wang, Li, Zhang, Song and Cheng2022). Our results showed that FC alterations between the cerebellum, language-related, and subcortical regions were directly linked to symptom improvement, further supporting the regulatory role of this loop. Similarly, low-frequency rTMS over the left TPJ altered FC of the cerebellar dentate nucleus with temporal and frontal regions, correlating with clinical response (Xie et al., Reference Xie, He, Guan, Zhou, Wang, Ma, Wang and Yin2022), suggesting bidirectional plasticity within the cerebellum–cerebellar circuitry.
Cortical–cerebellar–subcortical–cortical circuit
These studies provide anatomical and functional evidence for the cortical–cerebellar–subcortical–cortical circuit in AVH. In first-episode schizophrenia with AVH, reduced FC between the bilateral thalami and cerebellum and enhanced effective connectivity from the left thalamus to the right inferior frontal gyrus have been reported, suggesting the involvement of the cerebellum–thalamus–cortical circuit (Wei et al., Reference Wei, Xue, Yang, Wang, Chen, Han, Wang, Li, Zhang, Song and Cheng2022). A randomized double-blind, sham-controlled trial of AVH treatment targeting the left TPJ with identical cTBS parameters reported that the e-field strength within the AVH network was independently associated with treatment outcomes. Although the e-field strength in the cerebellum might be very low because of its distance from the TPJ stimulation site, the AVH networks in this trial were defined and computed using the cerebellum as the seed hub. These findings in this trial highlighted the importance of the cerebellum-related AVH network modulation and further supported a role for the cerebellum in the treatment of AVH (Hua et al., Reference Hua, Wang, He, Sun, Xu, Zhang, Tian, Wang and Ji2024). However, the e-field strength was not measured in our study. By focusing on the FC changes pre- and posttreatment, our study observed that improvement in auditory hallucinations was directly associated with FC changes between the cerebellum and the motor language area, auditory cortex, and basal ganglia. Previous work has also reported similar FC alterations in patients with p-AVH (Rehme, Eickhoff, & Grefkes, Reference Rehme, Eickhoff and Grefkes2013), as well as baseline abnormalities in the superior temporal gyrus, posterior cingulate, and hippocampus (Mallikarjun et al., Reference Mallikarjun, Lalousis, Dunne, Heinze, Reniers, Broome, Farmah, Oyebode, Wood and Upthegrove2018). However, widespread FC abnormalities in AVH patients make it difficult to attribute symptom improvement to specific connections alone.
We propose that cTBS may reshape FC patterns within this loop while enhancing effective connectivity between key regions, which could directly influence AVH occurrence. In typical AVH patients, ratios of resting-state cerebral blood flow to FC in cerebellar regions (vermis, Crus I/II) are decreased, whereas this ratio in superior/middle temporal gyri is increased and correlates with hallucination severity (Chen et al., Reference Chen, Xue, Yang, Wang, Xu, Wen, Cheng, Han and Wei2022). Consistently, our results showed that posttreatment FC changes between the cerebellum and superior/middle temporal gyri were negatively correlated with AVH severity, suggesting that cTBS may modulate effective FC within the cortico–cerebellar–subcortical–cortical circuit to improve symptoms.
Moreover, studies using rTMS at cortical sites also reported altered cerebellar–cortical FC correlating with symptom improvement (Xie et al., Reference Xie, He, Guan, Zhou, Wang, Ma, Wang and Yin2022), confirming the bidirectional nature of cerebellar–cortical interactions. Anatomical evidence of reciprocal projections further supports this two-way connectivity.
Effects on the language network
Moreover, evidence from another study further underscores the importance of the cortical–cerebellar–subcortical–cortical circuit in the pathophysiology of auditory hallucinations in schizophrenia. Specifically, compared with schizophrenia patients without AVH, those experiencing AVH showed significantly stronger FC between the left inferior frontal gyrus (BA45/46; Broca’s area) and Wernicke’s area. Enhanced FC was also observed between Wernicke’s area and multiple subcortical structures, including the thalamus, midbrain, and putamen (Hoffman, Fernandez, Pittman, & Hampson, Reference Hoffman, Fernandez, Pittman and Hampson2011).
Wernicke’s area, traditionally mapped to BA22, is situated in the posterior portions of the superior and middle temporal gyri and the supramarginal gyrus of the dominant hemisphere. Together with Broca’s area, it constitutes the core language network, with the two regions connected via the arcuate fasciculus linking the frontal and temporal lobes (Andoh & Martinot, Reference Andoh and Martinot2008). Coordinated activity within this network is considered essential for determining whether neural signals are consciously experienced as percepts (Cosmelli et al., Reference Cosmelli, David, Lachaux, Martinerie, Garnero, Renault and Varela2004; Melloni et al., Reference Melloni, Molina, Pena, Torres, Singer and Rodriguez2007; Sergent & Dehaene, Reference Sergent and Dehaene2004). Thus, excessive FC between the frontal and bilateral temporal cortices may underlie the emergence of AVH, whereas normative FC within this pathway is crucial for processing speech content and the acoustic features of external speech (Benson et al., Reference Benson, Whalen, Richardson, Swainson, Clark, Lai and Liberman2001; Lattner, Meyer, & Friederici, Reference Lattner, Meyer and Friederici2005). Consequently, AVH may manifest as internally generated but externally perceived voices – unintended verbal content that intrudes into conscious awareness despite its unconscious origin. Patients with persistent AVH may therefore exhibit profound impairments in perceptual integration.
Importantly, our findings demonstrate that cTBS applied to the left cerebellum increases FC between the cerebellum and Broca’s area (BA44/45), while simultaneously reducing FC with the auditory cortex (superior and middle temporal gyri) and subcortical regions such as the basal ganglia. Notably, both patterns of FC change were positively associated with the degree of symptom improvement in patients with p-AVH. These results suggest that the observed improvement of AVH following cTBS may be associated with FC changes within the cortical–cerebellar–subcortical–cortical circuit, characterized by enhancing cerebellar coupling with frontal language regions and downregulating its connectivity with temporal auditory and subcortical structures – thereby restoring a more adaptive network configuration.
Mechanisms of cognitive performance
This study found that, in addition to improvement of p-AVH symptoms, cognitive performance in patients with schizophrenia improved following cTBS treatment, particularly in the domains of verbal function, attention, and delayed memory. Two possible explanations may account for these findings. The first is that cTBS may improve cognitive performance indirectly through the amelioration of auditory hallucinations. The second is that cTBS directly may enhance cognitive functioning in patients with schizophrenia.
Cognitive impairment is one of the core features of schizophrenia, with deficits being especially prominent in processing speed, verbal memory, and working memory (Gebreegziabhere et al., Reference Gebreegziabhere, Habatmu, Mihretu, Cella and Alem2022). AVHs are closely associated with a range of cognitive deficits involving memory, inhibition, and language processing. Previous studies have shown that AVH in schizophrenia is primarily related to impairments in perceptual decision-making, source monitoring, working memory, and language, and that additional deficits can be observed in recognition and delayed memory, executive function (attention), emotional processing, and language comprehension (Bell et al., Reference Bell, Toh, Allen, Cella, Jardri, Larøi, Moseley and Rossell2024). Thus, it is plausible that cTBS may improve cognitive symptoms in schizophrenia by reducing the severity of hallucinations.
Prior navigation-guided TBS studies have demonstrated that stimulation of the cerebellar vermis can enhance cognitive performance in patients (Demirtas-Tatlidede et al., Reference Demirtas-Tatlidede, Freitas, Cromer, Safar, Ongur, Stone, Seidman, Schmahmann and Pascual-Leone2010). Case reports have also suggested that cerebellar stimulation may influence the broad cognitive impairments observed in schizophrenia, including deficits in verbal episodic memory, short-term memory, and working memory (França et al., Reference França, de Andrade, Teixeira, Galhardoni, Silva, Barbosa and Cury2018; Laidi et al., Reference Laidi, Levenes, Suarez-Perez, Février, Durand, Bouaziz and Januel2020). Key regions implicated in cognition – such as the language motor area, hippocampus, and prefrontal cortex – are known to be impaired in schizophrenia. There is substantial evidence of altered dopamine (DA) and GABA neurotransmission in individuals with schizophrenia (Qi, Miller, & Voit, Reference Qi, Miller and Voit2008), and GABAergic dysfunction has been linked to cognitive deficits. The cerebellum may play a compensatory role by modulating cortical inhibition and thereby improving cognitive impairments in schizophrenia (Ji, Yu, & Ma, Reference Ji, Yu and Ma2025). However, as this is an open-label study without sham control, the cognitive performance following repeated assessment may be influenced by several factors, including practice effects, test–retest effects, nonspecific clinical improvement, or expectancy effects in this study, even though previous studies have provided related evidence. Futural double-blind study would further help clarify it.
Strengths and limitations
Major strengths include the recruitment of clinically stable participants with consistent medication regimens and the use of whole-brain FC analysis rather than seed-based methods, ensuring unbiased identification of cerebellar network effects. The former, to some extent, ensures that the observed changes in whole-brain fMRI before and after treatment are attributable to cTBS intervention, ruling out confounding effects from the natural course of the illness or medication changes. Limitations include small sample size, the absence of a sham control group and blind set, and non-navigated target localization, which has been acknowledged to offer better precision for cortical TMS targeting. Our results suggest an association between cTBS treatment and whole-brain FC changes; however, due to the lack of a sham control group and blind set, we cannot completely rule out the effects of the natural disease course or medication changes in this open-label study. Future larger cohort studies with randomized sham-controlled and double-blinded designs are warranted to validate the specific therapeutic effects of cTBS. Besides, the sustained benefit seemed to be attenuated after 2–4 weeks in this study; this might be associated with the treatment session setting and might suggest that the cerebellar single-target treatment has limited treatment effect. More frequently accelerated treatment could be explored in the future and a dual-target setting with other sites in the cortical–cerebellar–subcortical–cortical network (like TPJ) may help increase a profound effect.
Accumulating evidence suggested that white matter signals and connectivity may provide complementary information about brain network organization, neuropsychiatric disorders, and treatment-related plasticity (Ji et al., Reference Ji, Liao, Chen, Zhang and Wang2017, Reference Ji, Ren, Li, Sun, Liu, Gao, Xue, Shen, Cheng, Zhu, Tian, Hu, Chen and Wang2019, Reference Ji, Sun, Hua, Zhang, Zhang, Bai, Wei, Wang, Qiu, Wang, Sun, Liao, Yu, Zhu, Tian, He and Wang2023, Reference Ji, Cui, D’Arcy, Liao, Biswal, Zhang, Luo, Zang, Ding, Zuo, Gore and Wang2025). Although this study exclusively focused on gray matter regions, future multimodal studies integrating white matter assessment would help provide a more comprehensive understanding of the neural mechanisms underlying the cTBS response in schizophrenia with p-AVH.
Conclusion
Targeting the left cerebellum with cTBS as an adjunctive treatment may help alleviate auditory hallucinations and improved cognitive performance in patients with schizophrenia and p-AVH. Its mechanism of action may involve modulating the FC patterns of the cortical–cerebellar–subcortical–cortical circuit. This study highlights the critical role of this circuit in the neuropathological mechanisms underlying p-AVH.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S0033291726106497.
Data availability statement
All data may be shared upon request. Please reach out to the corresponding author for any data inquiries.
Acknowledgments
We thank all participants for their voluntary participation in the study.
Author contributions
W.G. and Y.O. designed and supervised the study. Y.O. analyzed data and drafted the original manuscript. Y.Z., Y.D., M.Y., and H.L. assessed the subjects and collected imaging data. All authors contributed to and approved the final manuscript and were responsible for the decision to submit the manuscript.
Funding statement
This work was supported by the National Natural Science Foundation of China (82171508), the Natural Science Foundation of Hunan (Grant No. 2025JJ60706), and the China Postdoctoral Science Foundation (2025M782320).
Competing interests
The authors declare that there are no competing interests.
Open access
