Introduction
Thought disorder, clinically expressed as disorganized speech, is among the most disabling features of schizophrenia and predicts long-term functional decline1,2 and treatment resistance.3,4 Contemporary accounts link thought disorder to impaired internal models that maintain and flexibly update internal representations in predictive processing, disrupting the coherence of self-generated thought and language (5 see also 6,7).
The brain generates internal models through the interplay of top-down predictions from anterior cortex and bottom-up sensory evidence from posterior regions,8 while resting-state activity helps maintain these representational hierarchies in between task demands.9 The default mode network (DMN) is a relevant substrate for this process, particularly for self-generated and internally oriented cognition.10,11 However, the DMN is not unitary: posterior and anterior components differ in their coupling to perceptible events and self-generated content.12,13 Across this posterior-to-anterior axis, internal representation becomes progressively more abstract, from sensory-motor through lexical-semantic to self-generated narrative. DMN dynamics are consistently reported to be disrupted in schizophrenia, with connectivity and activation abnormalities linked to symptom burden in affected individuals.3,14–16
Beta-band (15–30 Hz) neural activity is a candidate mechanism for implementing internal models. Beta activity has been proposed as a cortical signature of top-down, endogenous representation,17 mediating feedback from deep cortical layers to suppress conflicting input and coordinate predictions hierarchically.18 Rather than being sustained, beta-band activity is often organized into transient high-amplitude events (“bursts”),19–21 providing discrete windows during which cortical representations may be stabilized or reinstated (i.e., “distinct computational primitives”).22 If beta bursts implement internal models, the temporal coordination of burst events across distributed regions is crucial for hierarchical integration of representations. Time-averaged spectral power can obscure this event structure,23 whereas burst rate, duration, peak power, and inter-regional burst coincidence (synchrony) may capture distinct aspects of beta organization.24,25
Task-based evidence indicates that beta-burst dynamics are disrupted in schizophrenia. During working memory task performance, patients show reduced burst rate alongside greater burst-related BOLD activation, with more severe disorganization associated with more extensive spatial distribution of beta bursts.26 Resting-state studies, however, have largely focused on spectral power and temporal correlations, commonly reporting elevated frontal and reduced posterior beta power without characterizing temporal coordination of burst dynamics (i.e., duration and inter-regional synchrony; Table S1). Importantly, prior work indicates that thought and speech disorganization are reflected in intrinsic brain organization measured at rest.
Resting-state fMRI studies have related speech disorganization to network-level functional organization27,28 and effective-connectivity.29–31 Earlier resting M/EEG studies have also reported beta-band abnormalities associated with formal thought disorder.32,33 Together, these findings motivate testing whether intrinsic resting-state beta-burst organization, in the absence of immediate task demands, covaries with speech structure. Hierarchical predictive accounts further propose that disorganized speech reflects impaired use of higher-order context to constrain and update internal representations,5,34 while beta has been proposed to support the maintenance of top-down endogenous representations.17 Because beta bursts may stabilize and coordinate internal models across cortical hierarchies, resting-state burst organization may directly index the integrity of predictive architecture during task-free states. Within the DMN, the antero-posterior PFC-IPL axis is the specific substrate for this: PFC generates higher level predictions; the IPL integrates contextual evidence in support of semantic concepts.35–37 Impaired coordination on this axis, either through excess synchrony (locking of top-down predictions) or temporal instability (failure to sustain representations long enough to integrate them), may produce the conditions for predictive processing failure.
Because language is a quintessential function of hierarchical predictive processing,5,38 a bias towards context-insensitive predictions, may provide the necessary conditions for thought disorder manifested in speech to emerge.34,39–42
We examined beta-burst dynamics using resting-state MEG from source-localized cortical and DMN time series, with burst rate, duration, power, and synchrony quantified using established methods.43 We first tested for cortex-wide spatial organization of beta-burst abnormalities, then focused on DMN subregions. This two-stage approach separates a broad cortical anterior–posterior (A-P) gradient from its expression within a network directly relevant to the dynamics of spontaneous thought. Based on prior evidence of elevated frontal and reduced posterior beta power (Table S1), and impaired flexible maintenance of internal representation in schizophrenia,44–47 we predicted enhanced burst synchrony in anterior DMN regions and reduced temporal persistence of bursts in the IPL where semantic integration is implemented.
We further predicted that these burst abnormalities would relate to objective measures of thought disorder manifested as speech disorganization.1,48,49
Methods and materials
Participants
This case-control study recruited 25 community-living patients with clinically stable schizophrenia and 25 age- and sex-matched healthy controls. The study was approved by the Research Ethics Board of Douglas Mental Health University Institute (McGill University), and all participants provided written informed consent. Inclusion criteria were age 18–65 and fluency in English or French.
Patients were recruited from McGill-affiliated psychiatric services and met DSM-5 criteria for schizophrenia or schizoaffective disorder. Participants completed clinical, speech, MEG, and MRI assessments. Two patients were excluded from MEG analyses due to artifacts, resulting in a final analysis sample of 48 participants (Table 1).
Clinical measures included SOFAS, PANSS-8, and CGI-S. Medication exposure was characterized according to antipsychotic class, clozapine use, antipsychotic polypharmacy, and World Health Organization defined daily dose (DDD), with DDD-derived chlorpromazine equivalents calculated using established method.50 Associations of medication exposure with the three burst principal components were examined within patients using rank-based subgroup comparisons and Spearman correlations for continuous antipsychotic dose, with FDR correction applied within each set of medication analyses. Full medication characteristics and sensitivity analyses are reported in Supplementary 3.5.
Natural speech recording and analysis
Speech samples were collected using the DISCOURSE protocol,51 comprising structured and free speech tasks (~20 min). Recordings were transcribed using WhisperX and manually corrected, diarized for retaining only participant speech. Total word count did not differ significantly between groups (t(45.98) = 0.56, p = 0.58).
Six semantic and syntactic speech features were extracted using NLP and LLM-based methods (Supplementary 1; Table S2). See Table S3 for speech feature interpretation.
MEG resting-state and structural MRI acquisition
Resting-state MEG data (3 min, eyes open) were acquired using a 275-channel CTF system at 1200 Hz. Head position was continuously monitored using fiducial coils, and EOG/ECG were recorded for artifact detection. Structural MRI was acquired on a 3T Siemens Prisma scanner for source reconstruction.
Head shape was digitized using FASTRAK (Polhemus, USA; ~100 scalp points including fiducials), and MEG-MRI co-registration was performed in Brainstorm.
MEG data preprocessing
Data were processed in Brainstorm using automated MATLAB pipelines. Signals were filtered at 0.1–100 Hz with a 60 Hz notch, and cardiac and ocular artifacts were removed using signal-space projection. Source reconstruction used an LCMV beamformer with unconstrained dipole orientation and overlapping spheres head models.
Whole-brain parcellation used the Schaefer-200 atlas.52 Regional time series were extracted by reducing the three dipole orientations at each vertex to a single time course using principal component analysis (PCA), followed by sign-flipped averaging within each ROI. Analyses were restricted to the first 180 s to standardize data length across participants.
Beta oscillatory power analysis
Power spectral density was computed using Welch's method with 4 s windows and 50% overlap. Beta power was defined as mean power in the 15–30 Hz range for each region.
Beta-burst analysis
Beta bursts were identified following Shin and colleagues.21 Using their threshold-selection criterion, 4× the median power maximized the correspondence between the proportion of suprathreshold TFR pixels and mean beta (15–30 Hz) power in the present data. Data were downsampled to 600 Hz, and time–frequency representations (8–35 Hz, 1 Hz resolution) were computed using Morlet wavelets (width = 10). Local maxima were identified using the Python equivalent of MATLAB's “imregionalmax”.
Beta bursts were computed separately for each subject and region. Burst onset and offset were defined using the full width at half maximum around the peak. Burst duration, normalized peak power, and burst rate were then computed.
Inter-regional synchrony and connectivity
Inter-regional burst synchrony within the DMN was quantified as the probability of coincident bursting between region pairs. Time series were segmented into 150 ms epochs centered on burst onset in region A; burst probability in region B was estimated in 50 ms sliding windows advanced in 10 ms steps, generating a peri-burst probability curve. Synchrony was defined as the probability at time zero.
To assess the sensitivity of burst-synchrony estimates to recording duration, we evaluated their convergence across progressively longer data segments and independently assessed split-half reliability, confirming the stability of the estimates at the 180 s duration used in the primary analyses (Supplementary 2).
Burst synchrony captures temporal co-occurrence of bursts not limited to zero-lag overlap. To validate and interpret the burst synchrony findings, beta-band amplitude envelope correlation (AEC) was computed for DMN regions (MNE-Python; bandpass-filtered at 15–30 Hz using a zero-phase finite impulse response filter with a Hamming window; amplitude envelopes extracted using the Hilbert transform). Two forms of AEC were computed to distinguish between instantaneous and delayed coupling.
(1) Non-orthogonalized AEC, as a continuous-signal comparator, captures the full amplitude envelope correlation between regions, including zero-lag co-fluctuations that may reflect genuine instantaneous interactions or spurious connectivity from source leakage. (2) Symmetric pairwise orthogonalized AEC removes all zero-phase-lag components from each pair of signals before computing correlations, isolating time-delayed coupling while suppressing true instantaneous interactions as well as volume conduction.53 Comparing results across both measures therefore identifies whether a group effect is driven primarily by instantaneous co-occurrence (including potential leakage contributions) or by sustained delayed coupling that is robust to volume conduction confounds.
Regional AEC comparisons were restricted to region pairs where burst dynamics (synchrony or duration) showed FDR-corrected group differences. To characterize overall spatial correspondence between burst synchrony group differences and each AEC type across all DMN region pairs, Mantel tests were conducted. Sliding-window AEC was computed using window lengths of 0.5–180 s, to test whether connectivity differences depended on the temporal scale of observation.
Pairwise Euclidean distance between regions was computed to test whether burst synchrony differences scaled with spatial proximity, as would be expected under a signal leakage account.
Statistical analysis
Group differences
Beta power and burst characteristics were compared between groups for each regional time series using independent-samples t-tests. FDR correction was applied separately for each beta characteristic across the 200 regions. A-P trends were assessed using linear mixed-effects models: burst_feature ~ position*group + (position|subject), where position was defined as the mean y-coordinate of each Schaefer region.
DMN analyses grouped regions into four functionally and anatomically defined sets: prefrontal cortex, temporal lobe, inferior parietal lobe, and posterior cingulate cortex. Burst features were averaged within groups and compared using FDR-corrected t-tests, with correction applied separately for each burst feature across the four regional comparisons. DMN burst synchrony matrices were compared between groups, and regional synchrony was summarized within and between hemispheres, with FDR correction applied across the regional synchrony comparisons.
FDR correction was therefore performed within each beta-feature family rather than jointly across burst rate, power, duration, and synchrony. Thus, the overall multiplicity burden across beta analyses should be considered when interpreting the findings. Mantel tests assessed correspondence between synchrony, connectivity, and distance matrices, with analyses repeated for AEC metrics.
Results with p < 0.05 that did not survive FDR correction are reported as trending group differences. Associations with DDD were tested to assess medication effects.
Dimension reduction and regression
As pooling diagnostically distinct groups can induce spurious brain–behaviour associations, burst–speech relationships were modelled within the patient group alone (n = 23). The six burst features showing significant group differences were z-scored and summarized by their first three principal components (cumulative 85% of variance: PC1 36%, PC2 32%, PC3 16%), and the six semantic/syntactic speech features by their first principal component to provide a single speech organization factor (50% of variance). Speech organization was regressed on the three burst PCs in a single multiple linear regression; standardized coefficients (β), per-predictor significance, and the omnibus model fit are reported, with leave-one-out sensitivity to influential cases.
As a sensitivity analysis, the patient-only regression was repeated with years of education included as an additional covariate (Supplementary 3.2). To test symptom specificity to disorganized speech, the clinical severity component (first PC of 10 clinical scores: PANSS items, CGI-S and SOFAS; 58% of illness-burden variance; see Table S4) was regressed on the same three burst PCs. As a complementary multivariate analysis, a partial least squares (PLS) canonical correlation of the full burst and speech feature sets separately within patient and control groups, and covariate-adjusted (group, age, sex) whole sample models with 5000-permutation are reported in the Supplement (Supplementary 3; Figs.
S7–9).
Results
Spatial gradient of burst alterations
Patients and controls did not differ in regional beta-band PSD after correction for multiple comparisons (p > 0.05). For beta-burst dynamics, all three burst features showed a significant flattening of the A-P gradient between patients and controls, with relatively higher frontal and lower posterior values than controls. In healthy controls, burst power (coef = −4.63, p = 2.38e-5) and duration (coef = −0.0592, p = 3.56e-5) were greatest in posterior regions and declined toward the frontal cortex, while burst rate (coef = 0.275, p = 0.0143) was lowest in posterior regions and increased toward the frontal cortex.
In patients, this pattern was inverted: burst power and duration were elevated frontally and reduced posteriorly, while burst rate was disproportionately increased anteriorly (Fig. 1). This group-by-position interaction was significant for burst rate (p = 0.018, coef = 0.374 ± 0.152), burst power (p = 0.0014, coef = 4.51 ± 1.32), and burst duration (p < 0.001, coef = 0.063 ± 0.016); the crossing trajectories are illustrated in Figs.
1D–1F. Region-wise comparisons are provided in Fig. S1.
DMN burst alterations
The cortex-wide A-P gradient was also reflected in DMN subregions. Within the DMN, patients showed shorter burst duration in the IPL, a posterior DMN node (t(46.0) = −2.74, p = 0.00858, FDR-adjusted p = 0.0343, d = −0.790) (Fig. 2).
There was a significantly increased zero-lag burst coincidence within left PFC (t(31.0) = 2.81, p = 0.00858, FDR-adjusted p = 0.0343, d = 0.833) and between left and right PFCs (t(29.8) = 2.73, p = 0.0106, FDR-adjusted p = 0.0423, d = 0.811) (Fig. 3).
Increased PFC synchrony and reduced IPL burst duration were weakly negatively correlated (IPL duration with within-PFC synchrony: r = −0.15; with between-PFC synchrony: r = −0.076; Fig. S3A), indicating that these two patterns may not occur in the same individuals.
No significant differences in burst characteristics were observed in PFC or temporal regions of the DMN; no synchrony differences were observed within or between IPL or PCC regions (Fig. S4).
Some reduction was also noted in burst power at the IPL (t(37.7) = −2.25, p = 0.0304, d = −0.635) and PCC (t(46.0) = −2.41, p = 0.0200, d = −0.694), with some increase in synchrony within the left temporal regions (t(36.9) = 2.19, p = 0.0347, d = 0.645), but none of these three effects reached significance after FDR correction (FDR-adjusted p > 0.06).
Burst synchrony and amplitude envelope connectivity
Group differences in burst synchrony across all DMN region pairs closely matched group differences in non-orthogonalized AEC (Mantel r = 0.857, p < 0.0001), supporting that the burst synchrony pattern reflects a genuine property of amplitude covariation rather than an artefact of the burst-detection approach. Burst synchrony group differences showed no spatial correspondence with orthogonalized AEC differences (Mantel r = 0.102, p = 0.076), indicating that the synchrony effect is driven primarily by zero-lag co-occurrence rather than sustained delayed coupling.
Between left and right PFC and within left temporal regions where burst synchrony showed group differences/trends, non-orthogonalized AEC was also elevated in patients: between left and right PFC (t(38.8) = 2.07, p = 0.045, d = 0.61) and within left temporal regions (t(44.8) = 2.59, p = 0.0128, d = 0.75). Orthogonalized AEC showed no corresponding PFC increase, supporting that the prefrontal effect reflects instantaneous burst co-occurrence rather than sustained delayed coupling. The distance analysis argued against a leakage account: burst synchrony differences were positively correlated with inter-regional Euclidean distance (r = 0.234, p < 0.0001; Fig.
S5), the opposite of what proximity-driven leakage predicts.
In contrast, at the IPL region pairs where burst duration showed an FDR-corrected group difference, orthogonalized AEC was reduced in patients (within right IPL: t(33.4) = −2.22, p = 0.034, d = −0.622; between left and right IPLs: t(40.8) = −2.18, p = 0.035, d = −0.619). No corresponding IPL effect was observed for non-orthogonalized AEC.
Sliding-window AEC showed that this orthogonalized AEC reduction emerged only at window lengths of 50 s or longer, whereas non-orthogonalized AEC increases in PFC were consistent across all timescales (0.5–180 s; Fig. 4). Reduced IPL coupling was associated with shorter IPL burst duration (r = 0.41, p = 0.004) and lower burst power (r = 0.87, p < 0.001).
Together, the AEC findings indicate that the increased between PFC and left temporal burst synchrony is a zero-lag phenomenon: present in non-orthogonalized AEC, absent after orthogonalization, consistent across all observation timescales, and not attributable to spatial proximity. Reduced IPL coupling is a delayed, slower timescale phenomenon: detectable only in orthogonalized AEC, emerging only at window lengths of 50 s or longer, and linked to shorter burst duration.
Burst-speech and symptom associations
Within patients (n = 23), the six group-discriminating burst features reduced to three principal components (PC1 36%, PC2 32%, PC3 16% of variance; cumulative 85%) (see Fig. 5A). PC1 loaded on the A-P burst gradient; PC2 loaded on between- and within-prefrontal burst synchrony; and PC3 loaded predominantly on inferior parietal (IPL) burst duration (loading +0.81) with the A-P burst power gradient (+0.52).
The speech organization component derived from syntactic and semantic complexity features (50% of variance) indexed lower clause count, syntax depth, and perplexity and greater consecutive similarity (see Fig. 5B).
A multiple regression of speech organization on the three burst PCs was significant (R2 = 0.47, adjusted R2 = 0.39; F(3,19) = 5.66, p = 0.006). The association was carried specifically by PC3 reflecting IPL burst duration (β = –0.55, p = 0.0036), with a weak contribution of prefrontal synchrony (PC2: β = 0.34, p = 0.054) but not the A-P burst gradient component (PC1: β = 0.22, p = 0.20) (see Figs. 5C and 5D).
Unpacking this at the single feature level, higher prefrontal synchrony was associated with shallow syntax structure (r = −0.53 to −0.61, p = 0.002–0.009 for within and between PFC synchrony with clause count and syntax depth), while shorter IPL burst duration related to more predictable speech (r = −0.46, p = 0.028 for consecutive similarity). Results were robust to outliers (Supplement 3.3, Table S5). Neither antipsychotic dose nor medication subgroup was significantly associated with any of the three burst PCs after FDR correction (Supplementary 3.5).
The largest medication-subgroup effect was observed for Burst PC3, with lower PC3 scores in clozapine-treated patients than in those receiving other antipsychotics (rank-biserial r = −0.542, permutation p = 0.0469, FDR-adjusted p = 0.141); this direction corresponds primarily to shorter IPL burst duration, although the effect did not survive FDR correction.
The same three burst PCs were not significantly associated with clinical severity (clinical principal component; R2 = 0.20, F(3,19) = 1.57, p = 0.23) or specific symptom domains, including delusions (PANSS P1), hallucinations (P3), or negative symptoms (mean N1/N4/N6) (Supplementary 3.4). A trend-level contribution was observed for clinical severity and global A-P burst component (PC1: β = –0.37, p = 0.092). Burst dynamics therefore showed partially dissociable correlates: IPL burst duration and to some extent prefrontal synchrony tracked speech organization, but not the overall symptom burden.
See PLS analysis of the full burst and speech feature sets with concordant burst-speech association in the Supplement (Supplementary 3.1–3.2, Figs. S6–9).
Discussion
Beta abnormalities in schizophrenia are expressed through spatiotemporal changes in transient beta events rather than uniform changes in average beta power. The distinction between beta power and beta-burst organization is important for interpreting these findings. Time-averaged beta power pools periods of high-amplitude burst activity and therefore cannot distinguish whether a spectral effect reflects how often, how strongly, or how long beta states occur, or how their timing is coordinated across regions.23,25 Event-level burst measures separate these properties and provide access to the transient organization of beta activity and its contribution to functional coupling.24,25,43 This burst-based perspective provides a temporal bridge between prior reports of spectral beta abnormalities, altered connectivity, and reduced long-range temporal organization in schizophrenia (Table S1).
Patients showed a cortex-wide attenuation of the normal posterior-to-anterior gradient of burst rate, power, and duration, with two dissociable abnormalities within the DMN. Prefrontal regions showed excessive instantaneous co-occurrence of burst events. The inferior parietal lobule showed reduced burst duration and impaired slow-timescale coupling.
Together, these features were specifically associated with semantic and syntactic disorganization in spontaneous speech rather than the overall illness burden. The findings are consistent with spatially and mechanistically distinct failures of predictive representations at anterior and posterior nodes of the cortical hierarchy in schizophrenia.
Increased burst synchrony within and between prefrontal DMN regions was mirrored by non-orthogonalized AEC and was present consistently across timescales. Synchrony differences increased with inter-regional Euclidean distance. This pattern suggests the prefrontal effect as an increase in instantaneous, zero-lag burst co-occurrence rather than an artifact of volume conduction or event detection.
It is consistent with evidence that electrophysiological power-correlation networks reflect co-occurrence of transient high-amplitude events rather than sustained coupling,24 and with the view that beta bursts function as discrete computational events that stabilize and reinstate internal representations.17,54 Excessive instantaneous co-occurrence of prefrontal burst events may reflect overstabilization of internally generated representational states, providing a candidate burst-level account (i.e., increased coincidence) for prior reports of elevated frontal beta power and connectivity in schizophrenia.55,56 The contribution of prefrontal synchrony to speech syntax is consistent with the role of syntactic control on lexico-semantic choices and its impoverishment in schizophrenia.57 PFC oversynchrony may generate context-insensitive predictions that impoverish the representational input to the IPL.
Reduced burst duration in the IPL was accompanied by reductions in orthogonalized AEC at window lengths of 50 s or longer, which directly correlated with burst duration, with no corresponding reduction in non-orthogonalized AEC. This profile identifies the posterior effect as a deficit in sustained, time-delayed coupling that is associated with burst brevity, providing a candidate account of previously reported reductions in long-range temporal correlations.58–60 Shorter bursts narrow the temporal window during which neural representations can be stabilized and broadcast across distributed regions.20,21 In the IPL, a hub for contextual grounding and semantic integration, sufficient burst duration may enable the maintenance of contextual representations across the timescales required for coherent language production; when this is impaired, word choices may be driven by higher consecutive similarity, i.e., local rather than global context.
Complementary task-based evidence from the same cohort shows reduced event-related beta modulation in frontal regions during perception and motor action tasks.61 The combination of excessive anterior (PFC) resting synchrony, reduced posterior (IPL) burst duration, and impaired distributed task-evoked modulation suggests context-insensitive top-down signalling: prefrontal predictions overstabilized at rest and insufficiently reconfigured with contextual demands.
The cortex-wide group-by-position interaction for burst rate, power, and duration indicates that the observed DMN findings are the focal expression of a broader reorganization of beta-burst architecture across the entire cortex. Within the DMN, the gradient resolved to temporally distinct anterior and posterior effects: the prefrontal abnormality is a zero-lag, instantaneous phenomenon present at all timescales; the IPL abnormality is a delayed, slow-timescale phenomenon absent until zero-phase-lag components are removed. Within a predictive processing framework, these correspond to complementary failure modes: top-down predictions excessively instantiated at prefrontal sites while contextual evidence fails to be maintained long enough for posterior integration.
Two interpretive caveats apply. First, the predictive processing account is a plausible mechanistic framework, not a demonstrated functional assignment; attentional suppression, arousal regulation, and network excitability fluctuations remain viable alternative accounts. Second, at the individual level, prefrontal synchrony and IPL burst duration were essentially uncorrelated (also see distinct loading on PCs in burst-speech regression).
The two effects may reflect partially independent neurobiological processes affecting different subsets of patients rather than co-occurring consistently within individuals. Thus, the A-P burst imbalance should be treated as a group-level description rather than a demonstration of a mechanistic dissociation at an individual level.
Limitations
Several limitations apply. The modest sample restricts generalizability, and the burst-speech and burst-clinical associations are exploratory pending replication in larger cohorts. In our sample, neither antipsychotic dose nor clozapine/polypharmacy status was significantly associated with the burst PCs or constituent burst features after FDR correction (Supplementary 3.5).
However, the largest medication-subgroup effect involved lower Burst PC3 scores in clozapine-treated patients, corresponding primarily to shorter IPL burst duration; although this effect did not survive FDR correction, a clozapine-related contribution to the PC3 finding cannot be excluded. Antipsychotic treatment may influence beta dynamics through dopaminergic and serotonergic mechanisms, and medication-related modulation of beta activity has been reported in schizophrenia.62 Medication effects cannot be excluded given the small subgroup sizes and near-universal exposure to second-generation antipsychotics. Sex was considered during recruitment by age- and sex-matching patients and controls, and sex distribution is reported in Table 1.
Gender identity and socio-cultural gender-related variables were not systematically assessed; therefore, sex- or gender-specific effects could not be evaluated. MEG does not resolve laminar-specific signalling, limiting inference about whether prefrontal synchrony arises from superficial or deep cortical layers, as the predictive processing account would predict. Zero-phase-lag AEC retains some sensitivity to volume conduction; phase-based or laminar-resolved methods are the more stringent test.
Beta-burst metrics can be estimated reliably with 2–3 min recordings.63 Synchrony estimates showed strong convergence across recording durations and good overall split-half reliability, although between-PFC split-half agreement was more moderate in controls (Supplementary 2). The eyes-open resting paradigm may suppress posterior beta via visual drive and requires replication with eyes-closed recordings. The cross-sectional design prevents causal inference, although associations with speech and clinical measures support functional relevance.
Conclusions
In summary, patients with schizophrenia exhibited a cortex-wide reversal of the A-P gradient of beta-burst dynamics, with two temporally distinct abnormalities in the DMN: excessive instantaneous co-occurrence of prefrontal burst events, and reduced burst duration and slow-timescale coupling in the inferior parietal lobule. These findings suggest a hierarchical imbalance in which prefrontal predictions are overstabilized, while contextual representations in the parietal cortex are temporally too brief to sustain integration, both associated with disorganization in spontaneous speech. By demonstrating that transient burst organization rather than spectral power captures these group differences, we report disrupted beta-burst spatiotemporal architecture as a neurophysiological signature linking intrinsic network dysfunction to thought disorder in schizophrenia.