fMRI fluctuations within the language network are correlated with severity of hallucinatory symptoms in schizophrenia

Schizophrenia (SZ) presents heterogeneous symptoms and lacks reliable biological markers, complicating diagnosis. Auditory verbal hallucinations (AVH) are among the most prevalent and distressing symptoms, affecting 60–90% of patients. Building on Crow’s hypothesis that reduced left-hemisphere dominance for language contributes to SZ, the study focuses on the language network (LN) as a candidate system underlying self–other source distinctions implicated in AVH. Prior work suggests altered hemispheric communication and LN dysfunction in SZ at rest and during tasks. The study aims to investigate whether vulnerability to hallucinate in SZ is associated with altered LN connectivity and spectral properties of resting-state BOLD fluctuations, and how these relate to symptom severity (PANSS P1 delusions; P3 hallucinations). The central hypothesis predicts reduced left-hemisphere dominance with compensatory right-hemisphere recruitment in AVH-prone patients (AVH/D+), and specific associations between LN spectral content (fALFF bands) and positive symptom severity.

The paper situates its rationale in Crow’s model of schizophrenia as a failure of hemispheric language dominance linking AVH to decreased left-hemisphere lateralization, particularly in temporal speech-perception regions. Multiple studies report disrupted hemispheric communication and LN dysfunction in SZ across tasks and resting state, and AVH vulnerability has been tied to resting-state network alterations. Resting-state networks such as the default mode network overlap with task-related networks including the LN. Schizophrenia is commonly framed as a disconnection syndrome, with language posited as a high-level integrative function especially susceptible to such dysconnectivity. This background supports targeting the LN using data-driven connectivity and spectral analyses (fALFF) to capture potential frequency-specific dysfunctions linked to symptoms.

Design: Resting-state fMRI study combining independent component analysis (ICA) to extract the language network (LN) with spectral analysis of network and ROI time series using fractional amplitude of low-frequency fluctuations (fALFF). Symptom associations were tested with PANSS P1 (delusions) and P3 (hallucinatory behavior). Participants: De-identified data from Norwegian SZ patients divided into two mutually exclusive subgroups based on PANSS: AVH/D+ (predominance of positive symptoms; P3 ≥ 4; if P3=4 then P1≥4) and AVH/D− (predominance of negative symptoms; P3 ≤ 4; if P3=4 then P1≤3). Each subgroup n=17. All on second-generation antipsychotics; defined daily dose (DDD) recorded. Age-matched healthy controls (HC) n=17. Groups were similar in age, education, gender, and handedness. Ethics: Data transfer approved by REK-Vest (#04052020-6822). MRI acquisition: 3T GE Discovery 750. Resting-state fMRI: eyes-closed, 5.33 min, 160 volumes, 30 slices, 0.5 mm gap; voxel 1.72×1.72×3 mm³; TR=2000 ms, TE=30 ms, FA=90°, FOV=220 mm. Structural T1: SPGR, 7.42 min; TR=7.78 ms, TE=2.94 ms, FA=14°, FOV=256 mm, 1 mm³ isotropic. Preprocessing: SPM12 pipeline: structural alignment, motion correction, bias correction, spatial smoothing (6 mm FWHM), and standard space co-registration. LN extraction: Spatial ICA performed separately per subject to derive single-subject LN spatial map and time course (data-driven). Group-level LN maps obtained via one-sample t-tests with Benjamini–Hochberg FDR correction (p<0.05). Between-group comparisons: Two-sample t-tests on individual LN maps for contrasts HC vs AVH/D−, HC vs AVH/D+, and AVH/D+ vs AVH/D−. DDD included as covariate for analyses involving patient groups. BH-FDR corrected at p<0.05. Network-level fALFF: FFT of the LN time series; fALFF computed across slow-5 (0.01–0.027 Hz), slow-4 (0.027–0.073 Hz), slow-3 (0.073–0.198 Hz), slow-2 (0.198–0.25 Hz). Values normalized to canonical fALFF to limit individual confounds. Symptom correlations: Spearman correlations between normalized fALFF band values and PANSS P1 and P3 (patients only). ROI post hoc fALFF: Defined spherical ROIs (6 mm radius) at right BA45 (MNI 46,30,3; right frontal operculum/Broca’s homologue including insula) and contralateral left BA45 (MNI −46,30,3). For each ROI, extracted a representative time series via PCA (first principal component from all ROI voxels), computed fALFF, and normalized by the subject’s whole-brain mean fALFF across the detectable frequency range. Morphometry: Voxel-based morphometry (VBM) using CAT12/SPM12 to estimate gray matter volume in ROIs and test for structural differences between groups.

Participants: n=17 per group (HC, AVH/D−, AVH/D+). Demographics did not differ significantly. AVH/D− had higher antipsychotic DDD than AVH/D+ (p=0.01), included as covariate in between-patient analyses. Language network connectivity: All groups showed typical left-lateralized LN regions (left IFG/Broca’s BA44–45, insula BA13, premotor/SMA BA6, angular gyrus BA39, superior/middle temporal gyri BA21–22). AVH/D+ uniquely recruited additional right-hemisphere nodes: right frontal operculum/Broca’s homologue (BA44/45) and anterior insula (BA13). Between-group LN differences: - AVH/D+ vs HC: Increased connectivity in right IFG opercular part (BA47; MNI 47,32,5) and decreased connectivity in left IFG triangular part (BA45; MNI −51,24,17); FDR-corrected p<0.05. - AVH/D+ vs AVH/D−: Greater connectivity in right BA47 (MNI 43,16,10), right insula BA13 (MNI 43,16,−10), left insula BA13 (MNI 42,6,−12), and right premotor BA6 (MNI 57,5,15). Lower connectivity in left angular gyrus BA39 (MNI −54,−58,17) and left DLPFC BA46 (MNI −51,28,19). All FDR-corrected p<0.05, DDD covaried. Network-level fALFF differences (LN time series): - AVH/D+ vs HC: Lower amplitudes at 0.012–0.018 Hz (all t<−2.16, p<0.04) and higher at 0.03–0.036 Hz (all t>3.6, p<0.01). - AVH/D− vs HC: Significant differences across 0.06–0.08 Hz (mix of lower and higher AVH/D− values; all |t|>2.2, p<0.03). - AVH/D+ vs AVH/D−: Differences around ~0.06 Hz (all t>2.03, p<0.04). Symptom associations: - LN-level: In AVH/D− only, higher Slow-4 amplitude correlated positively with P1 (delusions) severity. - ROI-level (right BA45/insula region): In AVH/D+ only, Slow-4 amplitude correlated negatively with P1 (delusions) (p<0.04), and Slow-5 amplitude correlated positively with P3 (hallucinatory behavior) (p<0.04). ROI lateralization patterns: - HC: Similar left vs right ROI fALFF except slightly higher right at ~0.1 Hz (t>2.3, p<0.03). - AVH/D−: Left>Right at ~0.015 Hz (t>2.2, p<0.04); Right>Left at ~0.065 Hz (t>2.3, p<0.03). - AVH/D+: Right>Left at ~0.01 and ~0.1 Hz (t>2.17, p<0.02). Morphometry: No significant gray matter volume differences in left or right ROIs between HC and SZ groups, nor between AVH/D− and AVH/D+ (all t<1.47, p>0.15; between-patient comparison all t<0.57, p>0.57).

Findings support the hypothesis that AVH vulnerability in SZ is associated with altered hemispheric balance within the language network. AVH/D+ patients exhibited reduced left-hemisphere connectivity (left IFG, BA45) coupled with compensatory recruitment of homologous right-hemisphere regions (right frontal operculum/BA44–45) and right anterior insula, consistent with Crow’s proposal of diminished left language dominance in hallucinated psychosis. The right insula and frontal operculum are hubs interfacing language, salience, and interoceptive/self-awareness processes; dysfunction here could bias self–other source monitoring, contributing to AVH phenomenology. Spectral analyses revealed frequency-specific alterations in LN dynamics: AVH/D+ showed shifts in low-frequency power relative to HC and AVH/D−, and these alterations related to positive symptom severity. At the network level, greater Slow-4 power associated with delusional severity in AVH/D−, suggesting that LN dynamics within typical left-lateralized regions may underlie delusional thought in non-hallucinating SZ. Critically, ROI analyses highlighted that in AVH/D+, right IFG/insula spectral power tracked symptoms in opposite directions (lower Slow-4 with worse delusions; higher Slow-5 with worse hallucinations), indicating distinct frequency bands may index different positive-symptom dimensions. Absence of structural differences in ROIs underscores that functional alterations, rather than morphology, explain the observed symptom-linked LN changes. Overall, the data indicate that combining connectivity with fALFF provides complementary insights into LN dysfunctions that stratify SZ patients by AVH vulnerability and symptom severity.

This study demonstrates that SZ patients with high vulnerability to AVH show aberrant language network organization, with reduced left IFG connectivity and atypical recruitment of right-hemisphere homologues and insula. Frequency-specific alterations in LN and right IFG/insula fALFF correlate with positive symptom severity, distinguishing hallucinations from delusions. These LN connectivity and spectral features may serve as functional markers of AVH vulnerability in SZ, independent of structural abnormalities. Future work should include larger samples, more fine-grained characterization of AVH features, and state-dependent imaging during acute hallucinations to validate and extend these markers and clarify their mechanistic roles.

- Modest sample size (n=17 per group) limits power and generalizability; authors note need for larger datasets. - Cross-sectional, resting-state design assesses trait-like vulnerability; does not capture state-dependent dynamics during active hallucinations. - Potential confounding by antipsychotic dosage differences between AVH/D− and AVH/D+ (though DDD was included as covariate). - Eyes-closed resting scans were not continuously monitored; hallucinatory episodes during scanning cannot be fully excluded. - Focus on LN may overlook contributions from other networks involved in AVH; replication across modalities and tasks is warranted.