Introduction

Schizophrenia is a complex mental disorder, characterized by positive, negative, and cognitive symptoms1,2. It’s known as a brain network disorder involving dysfunction across distributed brain regions and exhibits degenerative changes with disease progression3,4. Hypotheses such as the gating theory and the disconnection hypothesis have been proposed to explain the multidimensional pathological mechanisms in patients with schizophrenia1,5.

Convergent observations showed abnormal interaction patterns of the frontal and occipital lobes, which was associated with the deficit of top-down information processing in schizophrenia and disease progression6,7,8,9. The disrupted connection patterns of the frontal and occipital lobes reflect the deficit of brain integration and segregation in schizophrenia described by functional magnetic resonance imaging (fMRI)2,10,11. Investigation of brain hierarchy alterations and evaluation of segregation and integration properties further advance our understanding of disorders12. Functional connectivity (FC) is a widely used approach to measure the coordinated activity between brain regions, primarily focusing on the strength of pairwise connections. Building upon this, functional connectivity gradients extract continuous patterns of variation in connectivity through dimensionality reduction13,14. This approach reveals the gradients and hierarchical relationships between brain regions, offering a more intuitive global network perspective that better reflects the overall organization and hierarchical structure of brain functional networks13,14. Changes in the gradient range are referred to match brain states change or task demands15,16. In view of the stability and validity of the functional gradient approach, we employed the gradient approach to try to quantify changes in the degree of hierarchical separation of the brain in patients with schizophrenia.

Although MRI provides us with rich spatial detail, due to the inherent limitations of MRI, it lacks sufficient temporal information. Therefore, we also considered Electroencephalography (EEG). EEG directly measures neural activity and provides a high temporal resolution examination of brain function compared to fMRI. A series of researches indicates that the resting-state spectral profile in schizophrenia is characterized by an increase in power at lower frequencies, specifically in the delta and theta ranges17,18. The results obtained by the connection method support these findings and point out a deficit in the precise synchronization between cortical areas17,18. Focusing attention onto normal brain function, Andrea et al. demonstrated the effectiveness of using traveling waves to explore information transfer between functional hierarchies in both resting-state and task-related data, providing a practical approach for this analysis19,20. To be specific, forward traveling waves spreading from the occipital to the frontal cortex reflect the bottom-up flow of information, and top-down propagation can be depicted by backward traveling waves21,22,23. This method offers feasibility for exploring the characteristics of fronto-occipital information exchange in schizophrenia. The single-modality information limits us to a passive, localized observation of characteristics in schizophrenia patients, and there are relatively few studies integrating EEG and fMRI. Our previous work has revealed the altered ability of EEG oscillations to influence the coordination between brain networks in schizophrenia, indicating the advantage that simultaneous EEG-fMRI can provide more comprehensive information24. Therefore, combining EEG with fMRI allows for a more comprehensive understanding of the information exchange and functional patterns between the frontal and occipital lobes in schizophrenia.

This work used simultaneous EEG-fMRI data to explore both top-down and bottom-up processes of the fronto-occipital lobes in patients with schizophrenia. The current work described brain hierarchies using eccentricity based on functional connectivity gradients and further quantifies information transfer between high-level and low-level regions using traveling waves. The association between traveling waves and eccentricity and their relationship with neurotransmitters and cognition was assessed by dominance analysis. Compared to examining regression coefficients, dominance analysis provides a more comprehensive assessment by accounting for variable influences across different model contexts25. We hypothesized that fronto-occipital dyscommunication is associated with network segregation properties of the connectivity hierarchy, which might be supported by neurotransmitter and cognition in patients with schizophrenia.

Methods

Participants

All individuals, including schizophrenia (n = 107) and age- and sex-matched HC (n = 62), were recruited from the Clinical Hospital of Chengdu Brain Science Institute from 2021 to 2023. The patients were diagnosed using the structured clinical interview for the DSM-V axis 1 disorders-clinical version (SCID-I-CV). No participants reported any neurological disorders, head injury, alcohol, or other substance abuse. Patients were on a stable dose of an antipsychotic medication. The demographic and clinical information of all eligible participants is shown in Table 1. The legal guardian of the patients and HC provided written informed consent and received monetary compensation. This study was approved by the ethical committee of the University of Electronic Science and Technology of China. All ethical regulations relevant to human research participants were followed.

Table 1 Demographic characteristics of patients with schizophrenia and controls
Full size table

Data acquisition and preprocessing

Simultaneous EEG and fMRI data. The simultaneous resting-state EEG and fMRI data were collected and used to explore the functional pattern in schizophrenia. Acquired data were preprocessed using the DPABI and EEGLAB pipeline24,26,27,28. For complete acquisition and preprocessing details about fMRI and EEG data, please refer to previous work24 and supplementary material.

PET data. The Neuromaps (https://github.com/netneurolab/neuromaps) covering various neurotransmitter maps helped assess associations of spatial patterns of alterations in patients (relative to HCs) with specific neurotransmitter systems. Volumetric PET images for 19 different neurotransmitter receptors, which were derived from specific numbers of HCs (Supplementary Table 1), were included in current work: (1) serotonin receptors: 5-HT1a, 5-HT1b, 5-HT2a, 5-HT4 and 5-HT6; (2) serotonin transporter: 5-HTT; (3) G-protein coupled receptors - endocannabinoid receptor type 1 (CB1); (4) α4β2 nAChR; (5) dopamine transmission: dopamine D1 and dopamine D2, dopamine transporter (DAT); (6) Gamma-aminobutyric acid type a (GABAa); (7) histamine H3 receptor; (8) muscarinic acetylcholine receptor subtype M1; (9) metabotropic glutamate receptor 5 (mGluR5); (10) μ-opioid receptor (MOR); (11) the norepinephrine transporter (NET); (12) N-methyl-D-aspartate (NMDA) receptor; (13) vesicular transporters for acetylcholine (VAChT).

Cognitive activation data. Cognitive weighted patterns associated between voxels with cognitive processes were obtained from Neurosynth (https://github.com/neurosynth/neurosynth), which is a meta-analytic tool that synthesizes results from a series of published functional MRI studies related to high-frequency cognitive terms. We used the results compiled by Justine et al. 25.

Cortical gradients and eccentricity analyses

Cortical gradients were computed using the BrainSpace toolbox in MATLAB29. This technique offers detailed hierarchical insights into the segregation and integration of functional networks by measuring the gradual transitions between distinct regions in their connectivity14. This resulted in a series of FC gradients, where a region’s position on the gradient space reflects higher similarity between FC patterns and those of other regions nearby on the gradient space, with lower similarity to regions farther away. Volumes were first downsampled into 4 × 4 × 4 mm3 in MNI space for computational efficiency. For each subject, a 16,493×16,493 connectivity matrix was then obtained by calculating Pearson’s correlation between all voxels that were bound to the cortical area30. The cerebellum and subcortical areas were excluded. Matrices were converted to z-values by using Fisher’s r-to-z transformation and thresholded row-wise at 90% sparsity to retain the strongest connections. Cosine similarity was computed to generate a similarity matrix that captures the similarity in whole-brain connectivity patterns for each matrix. Subsequently, Cosine similarity was submitted to diffusion map embedding, a nonlinear dimensionality reduction technique, to extract gradient components at the individual subject level. Specifically, diffusion map embedding is characterized by α = 0.5, which allows the influence of both global and local relationships between data points in the estimation of the embedding space23. Furthermore, the all-subjects group average gradient component was constructed as a template, and all individual gradient components were aligned to this template via Procrustes rotation.

To abstract brain region separation trends in the gradient space, the first three gradients, accounted for 56.55% of the variance, were chosen to calculate gradient eccentricity14.

Gradient eccentricity values were computed by the Euclidean distance between each voxel and the centroid of all voxels in each individual gradient space. We compared distributions of the gradient component and gradient eccentricity14, respectively, with age and sex as covariates. To achieve stable repeatability, a permutation test with a threshold-free cluster enhancement31,32 (TFCE, number of permutations = 5000, PFDR < 0.005) was performed.

Cortical traveling waves

Recent studies have shown that both top-down and bottom-up information flows can be quantified by traveling wave methods21,22. We briefly show the calculation flow of traveling wave power in Supplementary Fig. 1. To be specific, traveling wave analysis was performed on seven midline electrodes (i.e., OZ, POZ, PZ, CPZ, CZ, FCz, and FZ)21. The EEG signals were segmented into 1 s windows, sliding with a step of 500 ms. For each window, the 2D fast Fourier transform was computed on each 2D time-electrode map which was constructed by arranging the time series of seven midline electrodes. The maximum values in the upper and lower quadrants were extracted to quantify the power of forward waves (FW) and backward waves (BW), respectively21. Importantly, the spectrum was divided into the following frequency bands: delta (2 to 4 Hz) and theta (4 to 7 Hz). Therefore, the power of FW and BW for two frequency bands was obtained. To quantify the amount of significant waves, the null distribution was established by performing the same steps after having randomized the electrodes’ order 100 times. This generates a surrogate spectrum, which has the same temporal information but without spatial information. The amount of significant forward and backward waves was the log-ratio between the real values and the average of the 100 repetitions. It is worth noting that this value represents the amount of significant waves against the null distribution, that is against the hypothesis of having no FW or BW waves. Finally, this value quantifies the amount of forward and backward waves compared with the null hypothesis of having no cortical waves. Power changes were assessed by linear mixed-effects models using FW power (or BW power) as the response variable and group and frequency, and its interactive item as fixed variables (with the intercept and slope as random variables, grouped per subject). The age and sex were excluded as covariates. All linear mixed-effects models and multiple comparisons using Bonferroni correction were constructed using the lme4 package in R 4.3.133.

Combining EEG and fMRI

To gain a deeper understanding of the coordinated mechanisms underlying hierarchical brain organization and dynamic information flow, we quantified the relationship between hierarchical segregation and information transfer using Pearson correlation, reflecting their population covariation across subjects. Simply, the correlation between the gradient eccentricity based on the first three gradients and power of the traveling wave was calculated across subjects for each group in the voxel level with sex as a covariate. The cortical abnormality pattern described the difference between the two groups of the relationship. To ascertain whether there exists a statistical difference, a permutation test was employed to construct a null distribution (shuffling the original data and recalculating the difference; cluster-corrected P).

Dominance analysis with neurotransmitters and cognitive patterns

To identify the neurotransmitters and cognitive patterns most closely associated with brain activity, we employed dominance analysis. Dominance analysis was used to evaluate the relative importance of neurotransmitter maps and cognitive weighted maps in relation to association maps between the gradient eccentricity and the power of traveling wave in different EEG oscillations within the patient group. Each cortical map was divided into up to 100 regions using the Schaefer atlas30. Dominance analysis offered a comprehensive comparison of the relative contribution of each independent variable (Neurotransmitters or Cognitive patterns) to explain or predict the dependent variable (association maps) in all possible submodels (2k-1 submodels) in a given overall fit (the multiple linear regression model with k input variables)34. To express the relative importance of each independent variable more accurately and intuitively, it was quantified as a percentage of its total average contribution to the overall fit of the model25. Spatial autocorrelation-preserving permutation tests (5,000 repetitions) were used to assess the statistical significance of associations across brain regions and P values were corrected by FDR35.

Associations between dysfunction and clinic

To further investigate whether the coordination of the alteration of network segregation properties of the connectivity hierarchy and the intensity of information transfer is affected by disease, the PROCESS macro for the Statistical Package for the Social Sciences (SPSS)36 was used to perform the moderation analysis to test the moderation effect of illness (PANSS score or disease course) in the patient group, which also includes bootstrap-based confidence interval calculations. Only regions with significant differences in association maps between the gradient eccentricity and power of the traveling wave will perform the analysis. In addition, the interquartile range, which evaluated the dispersion of the distribution of the gradient eccentricity in the frontal and occipital lobes, was calculated to describe the characteristics of the brain. Using the duration of illness as a moderator, we explored the effects of disease on network segregation properties of the connectivity hierarchy at the global level of the brain. The sex was contained in the model as a covariate. To reduce the potential multicollinearity, both the independent variable and the moderator were mean-centered before calculating the interaction37. The interaction term network segregation properties * effect of illness (X *M) is added to the general linear model to examine the moderation effect:

$$Y={\beta }_{0}+{\beta }_{1}X+{\beta }_{2}M+{\beta }_{3}(X\times M)+\varepsilon$$

In this model, Y was the power of the traveling wave, and β3 represents the moderation effect.

Statistics and reproducibility

For the statistical analysis, several methods were applied. In the voxel-wise gradient eccentricity analysis, independent sample t-tests were conducted using the threshold-free cluster enhancement (TFCE) permutation test, a stringent multiple comparison correction strategy. For the comparison of traveling waves, a mixed-effects model was used, with post hoc p-values corrected using the Bonferroni method. To assess the correlation between traveling waves and gradient eccentricity, a permutation test was employed, with p-values cluster-corrected. Finally, for dominance analysis, spatial autocorrelation-preserving permutation tests were applied for post hoc evaluation, with control by FDR.

To test the impact of the number of gradient components on EEG-fMRI fusion analysis, we calculated the gradient eccentricity using the first two and the first four gradient components and examined its correlation with traveling waves. Additionally, considering the potential impact of the number of subjects on the association between neuroimaging features and neurotransmitters, we sampled participants without replacement, with sampling rate ranging from 60% to 100% in increments of 5%. We repeated the random sampling 1000 times and averaged the correlation obtained from all sampling to achieve robust outcomes. We also correlated the altered association between eccentricity and traveling waves in schizophrenia to neurotransmitters and cognitive terms.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Result

The alteration of the gradient eccentricity and traveling waves

To better evaluate the effects of schizophrenia on brain function, we examined the performance of gradients and found that the current work replicates previous findings very well (Supplementary Fig. 2)2. To quantify the brain functional separation trend based on three main gradients, a metric of gradient eccentricity was calculated. Increased degrees of gradient eccentricity (Fig. 1A) were distributed in the frontal and temporal lobes. Regions of reduced degrees involved the frontal and parietal lobes.

Fig. 1: The alteration of network segregation properties in frontal and occipital lobes.
figure 1

A Differences in the gradient eccentricity at the voxel-wise (P < 0.005, TFCE FDR correction). B The 3D gradient space. The lighter the color, the closer it is to the centroid. C Interquartile range of distribution of the gradient eccentricity. Cortex: the interquartile range was calculated using values of all voxels in the cortex. Within F-O: the interquartile range was calculated using voxel values where they were confined to the fronto-occipital lobe. Outside F-O: the interquartile range was calculated using values of voxels in the area outside the fronto-occipital lobe. *P = 0.006. D Traveling waves. **P < 0.05, Bonferroni-corrected. Abbreviation: IQR Interquartile range.

Full size image

We further checked the separation between the frontal and occipital lobes using the interquartile range (Fig. 1C) and found the degree of separation in patients was significantly reduced. However, the degree of separation of whole voxels in the cortex or the degree of separation in the area outside the fronto-occipital lobe showed no difference. The coefficient of variation was used instead of the interquartile range to measure the degree of separation as confirmatory analysis. The difference in separation between the frontal and occipital lobes has also emerged (Supplementary Fig. 3B).

The interaction between the group and frequency band had a significant effect on the power of the FW (p = 0.02). Post-hoc analysis showed that the SZ group has a higher theta power of FW than HC. In the HC group, theta power of FW was significantly decreased than delta power. Nevertheless, the power of BW did not demonstrate a significant interaction effect between the frequency band and group.

Combining EEG and fMRI

Exploiting simultaneous recording of EEG and fMRI data, traveling wave activity of delta or theta band in conjunction with gradient eccentricity was used to assess the functional hierarchical alteration. Supplementary Fig. 4 exhibited correlations across the subjects in each group. Compared with HC, the correlation (Fig. 2A) between gradient eccentricity and delta-FW was significantly enhanced in the Superior frontal gyrus, Inferior frontal gyrus, Anterior cingulate cortex, Rectus, Superior temporal gyrus, and Supramarginal gyrus but decreased in the Lingual gyrus, Cuneus, Superior occipital gyrus. The Superior frontal gyrus, Rectus, Superior temporal gyrus and Supramarginal gyrus showed stronger relationships between gradient eccentricity and theta-FW but in the Lingual gyrus, Superior occipital gyrus, Median cingulate cortex the correlation was reduced (Fig. 2B). Enhanced correlations of the SZ group between gradient eccentricity and delta-BW appeared in the Middle frontal gyrus and inferior temporal gyrus (Fig. 2C) and decreased correlations in the Middle temporal gyrus, Rolandic operculum and Insula. Different from the HC group, the SZ group showed higher correlations between gradient eccentricity and theta-BW in the Precentral gyrus, Middle frontal gyrus, Postcentral gyrus and Inferior occipital gyrus but lower correlations in the Inferior frontal gyrus, Middle temporal gyrus and Insula (Fig. 2D). As shown in Supplementary Figs. 5 and 6, we observed a similar abnormality pattern between eccentricity and traveling waves using either the first 2 or first 4 gradient components in the frontal lobe, occipital lobe, and insula.

Fig. 2: Threshold cortical abnormality pattern.
figure 2

The relationship difference between the gradient eccentricity and power of traveling waves. A The correlation between gradient eccentricity and delta-power of forward waves shows significant group differences. B Similar differences are observed for theta-power of forward waves. C, D For backward waves, significant group differences are found in the correlations with both delta and theta power. P < 0.05, cluster-corrected.

Full size image

Mapping receptors to disease vulnerability

We used the multiple linear regression which predicted association patterns between the gradient eccentricity and power of the traveling wave from neurotransmitter receptor and transporter densities, and the contribution of this receptors was showed in Fig. 3A and B. The spatial distribution of H3, 5-HTT, and DAT made a large contribution to the fit between receptors and the association pattern between gradient eccentricity and delta-power of backward waves. The MOR was contributed to the fit between receptors and the association pattern between gradient eccentricity and theta-power of backward waves. Also, the NET, MOR, and 5-HT1B have the large contribution to the fit between receptors and the association pattern between gradient eccentricity and delta-power of forward waves. The spatial distribution of MOR, CB1, and GABAa represents relatively greater contributions to the association pattern of the correlation between gradient eccentricity and theta-power of forward waves. Sampling analysis supports our current findings (Supplementary Fig. 7). MOR is also associated with the altered correlations between gradient eccentricity and theta-power of forward waves observed in patients (Supplementary Fig. 8B).

Fig. 3: Dominance analysis.
figure 3

A Cortical association pattern between gradient eccentricity and power of traveling waves. B The percent contribution of each input variable is defined as the variable’s dominance, which is normalized by the total fit of the model. C The evaluation of the model fitting for cognitive weighted patterns. D seven cognitive weighted patterns, which were significantly correlated with the association between the gradient eccentricity and theta-power of the forward waves, were evaluated by dominance analysis for their contribution to the association. The significance of each model is assessed using a spatial autocorrelation-preserving null model and is corrected for multiple comparisons (FDR). Asterisks denote significant models (FDR-corrected Pspin < 0.05). Abbreviation: FW Forward waves, BW Backward waves.

Full size image

Relationship between disease vulnerability and cognitive function

A particular model was retained (Adjusted R2 = 0.286, Pspin = 0.011) as shown in Fig. 3 C and D. To be specific, seven cognitive weighted patterns that were most relevant to the association patterns between the gradient eccentricity and theta-power of the forward wave in patients, namely Visual perception, Search, Face recognition, Recognition, Rhythm, Motor control and Encoding (Supplementary Fig. 9). Visual perception makes a large contribution to the fit (Fig. 3D) and the visual perception, search and face recognition conditional dominates others (Supplementary Fig. 10). We further examined the relationship between cognition and the altered association between eccentricity and traveling waves, finding that the Fixation, Navigation, Categorization, Visual attention and Visual perception contributed the most to the altered association (Supplementary Fig. 8C).

Moderation effects of the clinic

As shown in Fig. 4, the modulatory effects of higher positive symptom scores on the relationship between the power of forward waves and gradient eccentricity mainly appeared in the ventrolateral prefrontal lobe, while the modulatory effects of negative symptom scores on the relationship between the power of forward or backward waves and gradient eccentricity were concentrated in the insula and frontal and temporal opercula, the inferior temporal gyrus and middle temporal gyrus. General symptom scores mainly affect the ventrolateral prefrontal lobe on the relationship between the power of forward waves and gradient eccentricity. The details are shown in Supplementary Figs. 11, 12 and 13.

Fig. 4: Abstract of moderation effects in different symptom dimensions.
figure 4

A The correlation between gradient eccentricity and forward wave power in certain brain regions was moderated by positive symptoms. B Regions showing correlations between gradient eccentricity and either forward or backward wave power were moderated by negative symptoms. C The correlation between gradient eccentricity and forward wave power in some regions was moderated by general symptoms. D The interquartile range of gradient eccentricity distribution between the frontal and occipital lobes showed a correlation with forward wave power that was moderated by illness duration.

Full size image

The interaction effect (β=-0.046, P = 0.005) of duration and the interquartile range of gradient eccentricity values in the fronto-occipital lobe on the theta power of forward waves highlights the moderating role of duration. The relationship between the theta-power and the interquartile range was stronger when the duration was higher (slope  = −0.537, P = 0 .009) rather than low (slope  =  0.395, P = 0 .057). In addition, we calculated the interquartile range across the cerebral cortex and reproduced the moderating effect (β = −0.060, P = 0.014; Supplementary Fig. 14C) of the duration of the illness, but no such moderating effect (β = −0.010, P = 0.558; Supplementary Fig. 14D) was found when the interquartile range was calculated from the distribution of values in the parietal and temporal regions. In addition, unlike the theta power of forward waves, the moderating effect of the duration of illness was unexpressed in the relationship associated with the delta power of forward waves or delta and theta power of backward waves.

The potential effects of the treatment

We found no correlation between chlorpromazine equivalents and the gradient eccentricity distribution or the power of traveling waves (P > 0.05), nor did we find any interaction effects of chlorpromazine equivalents with the gradient eccentricity distribution or the power of traveling waves (P > 0.05) via moderator models. When the drug appeared as a covariate, the significant moderation effect of disease duration still existed (Pint = 0.016).

Discussion

The current research provides a comprehensive new perspective on frontoparietal interaction and its association with brain architecture, enabling us to make new inferences about top-down and bottom-up dyscommunication in schizophrenia. Patients showed increased and concentrated gradient eccentricity in the fronto-occipital lobes, which is associated with enhanced theta power of forward waves. The association between theta power of forward waves and eccentricity increased in the frontal and decreased in the occipital lobe. A reduced correlation between theta power of backward waves and gradient eccentricity was also found in the insula. The relationship between the interquartile range of the distribution of gradient eccentricity of the fronto-occipital lobes and the theta power of forward waves was moderated by the duration of the illness. Moreover, the association between gradient eccentricity and traveling waves in distinct regions was moderated in a symptom dimension-specific manner, which further supported that fronto-occipital dyscommunication can be influenced by the brain hierarchy under the disease condition. Furthermore, the opioid receptor (MOR) and psychological functions (e.g., visual perception, search, and face recognition) were related with the association between gradient eccentricity and traveling waves, which provides multimodal evidence to understand the functional hierarchical dyscommunication in schizophrenia.

The current study reveals the relationship between fronto-occipital bottom-up information transmission and brain hierarchy. Modulation of bottom-up processing was specific for perception38. The occipital cortex provides visual signals to the frontal cortex, and the frontal cortex guides attention by combining the bottom-up and top-down factors39. In healthy controls, we found that the theta power of forward waves is positively associated with gradient eccentricity in the occipital region. Additionally, this finding is in line with the unimodal network attribute of the occipital cortex40, indicating coincident network segregation properties in the occipital cortex with bottom-up information transmission. In contrast, the frontal region has the opposite association between the forward wave theta power and gradient eccentricity. In contrast to the occipital cortex, the frontal cortex is multimodal and primarily responsible for cognitive control and engages in extensive interactions across the entire brain41,42. The occipital lobe and frontal lobe are situated at opposite ends of the brain’s functional hierarchy gradient, contributing to distinct functional patterns in bottom-up information processing42,43. Our finding supports the perspective that bottom-up signaling is associated with the frontal segregation properties of the functional hierarchy44. However, in patients, we found that the stronger the theta power of forward waves, the greater the gradient eccentricity in the frontal region, indicating abnormal over-integration of the frontal in bottom-up information flow. Previous studies have shown that the excessive integration of the frontal and occipital lobes leads to a loss of the appropriate physiological basis for bottom-up information processing, which may further result in the emergence of visual-related impairments in patients45,46,47,48,49. Moreover, we also found that the relationship between forward waves and the fronto-occipital segregation properties was moderated by the duration of illness, which further indicated in the disease progression disruption of bottom-up information processing.

Biologically informed modeling suggested that instability in the occipital lobe may promote the emergence of hallucinations through self-organized patterns of neural excitation50. The current work indicates a reorganization of the functional hierarchy and significantly enhanced theta-power of forward waves in the frontal and occipital lobes in schizophrenia. The reversal correlation between the two properties further implies that activity of the occipital lobe is disturbed in schizophrenia. This might indicate the presence of hallucinations22,48,51. Theta waves are widely known to be involved in cognitive processing and information transmission52. Visual activity is accompanied by regular theta activity in the frontal and occipital lobes51,52. We further demonstrated through dominance analysis that the association between gradient eccentricity and the forward wave theta power is related to psychological functions (e.g., visual perception, search, and face recognition) related to vision. The current findings update the previous conclusions that the altered association between fronto-occipital bottom-up information processing and network segregation properties might be the pathomechanism of hallucinations underlying schizophrenia45,46,47,48,49. Furthermore, the current study also found the association between gradient eccentricity and the forward wave theta power was related to the MOR and CB1. The spatial distribution of MOR shapes lower-frequency (delta and theta) neural dynamics25 and abnormal expression of MOR may be accompanied by changes in frontal lobe activity53,54. Our findings indicated that the association between neural activities and hierarchy segregation might be shaped by MOR and consequently contribute to bottom-up dysfunction in schizophrenia.

In schizophrenia, insular abnormalities have been widely demonstrated55,56 and related to hallucinations, emotional dysregulation, and cognitive impairments57,58. In this work, the gradient eccentricity in the insula was decreased in patients relative to controls, implying insufficient hierarchy segregation, which might be related to decreased gray matter volume along with the disease progression55,56. The insula plays an important role in top-down and bottom-up information processing and is thought to be involved in integrating sensory and cognitive information in supporting subjective feeling states56,59. Interestingly, by combining EEG-fMRI, the current work further demonstrates the decreased correlation between backward wave activity and gradient eccentricity in the insula, suggesting that the insula might be involved in the disrupted top-down information processing in schizophrenia. These findings support previous findings that insular abnormalities may be a marker of top-down visual information-processing deficits in schizophrenia60,61.

Schizophrenia involves widespread brain functional impairments, and with the advancement of neuroscience, key circuits relevant to schizophrenia symptoms have been identified to involve the frontal, temporal, and mesostriatal regions6. We found that positive symptom of schizophrenia is accompanied by an association between the network hierarchy segregation properties of the ventrolateral prefrontal gyrus and forward wave power. Positive symptoms, such as hallucinations and cognitive impairments, are associated with pathological mechanisms of abnormal bottom-up and top-down information processing62,63. A disruption in the balance between information streams in the sensory cortex and higher brain cortex might be fundamental for the development of hallucinations in schizophrenia patients63. The current work further points out using a moderation model that the more severe the positive symptoms, the greater the impact on the association between the bottom-up information transmission and the hierarchical segregation properties of the ventrolateral prefrontal gyrus, providing multimodal evidence for understanding the mechanisms of positive symptoms in schizophrenia patients. Moreover, we found that the relationship between the network hierarchy segregation properties of the insula and frontal and temporal opercula and backward wave power can be weakened by the negative symptoms of schizophrenia. As known, negative symptoms appear to involve more intricate physiological mechanisms6,64,65. Our previous work has shown that the salience network exhibits the greater neuroimaging loadings during the course of the disease, dominated by negative symptoms9. Schizophrenia patients with negative symptoms often exhibit reduced volume in the temporal and frontal lobes64. Abnormal connectivity between the temporal lobe and insula, as well as between the temporal and frontal lobes, is associated with apathy and cognitive deficits65. We updated existing findings, suggesting that the reduction of the insula and temporal lobe’s involvement in top-down information processing may be an important factor in its dysfunction. In addition, we also observed the contribution of MOR to the association between the network hierarchy segregation properties and top-down information processing. MOR may mediate the anhedonia, amotivation, and asociality53. Ashok et al. mentioned significantly reduced MOR availability in frontal lobe and insula in schizophrenia, and MOR alterations underlie the development of negative symptoms53. However, they did not observe a directly significant association between MOR availability and negative symptoms. Our current work further indicated that MOR may cause changes between the network hierarchy segregation properties and top-down information processing, and that such changes were associated with negative symptoms.

Limitations

There are several limitations of the present study. Although we were able to investigate and control for the potential effect of some factors (age, sex), cross-sectional design cannot remove the potential influence of individual specificity. Furthermore, the number of male patients is relatively small. It is recommended to collect more male patients in subsequent studies to further investigate whether sex has a significant impact on the current research results.

Conclusion

The hierarchical organization of the fronto-occipital lobes significantly influences bottom-up information processing, while the insula plays a key role in top-down information processing in schizophrenia. The association between bottom-up information flow and the hierarchical segregation properties of the ventrolateral prefrontal cortex may be influenced by positive symptoms, while negative symptoms modulate the relationship between top-down information flow and the hierarchical segregation properties of the insula. Neurotransmitters, MOR and CB1, may underlie the alteration of links between segregation properties of connectivity hierarchy and the transmission of the top-down and bottom-up information flow. These alterations may be contributed to deficits of visual information processing in schizophrenia. These findings further facilitate our understanding of the insula and fronto-occipital dyscommunication in schizophrenia. Furthermore, exploration based on simultaneous EEG-fMRI is still insufficient. Our current work suggests that compared to unimodal analysis, this technique helps provide a more comprehensive understanding of pathological characteristics of the top-down and bottom-up processing in schizophrenia patients.