Alterations in the hub structure of whole‐brain functional networks in patients with drug‐naïve schizophrenia: Insights from electroencephalography‐based research

Abstract Aim This study aimed to identify atypical hubs in the whole‐brain networks of patients with schizophrenia (SZ) and examine the effects of antipsychotic medications, using electroencephalography (EEG) data. Methods We estimated the functional connectivity across all electrodes by applying the phase lag index to the EEG signals of 21 drug‐naïve patients with SZ and 31 age‐matched healthy controls. Betweenness centrality (BC), a measure of hub status, was calculated for each electrode and frequency band. Data from 14 patients were re‐evaluated after initiating treatment with antipsychotic medications. Results BC values decreased significantly at the Fz site in the beta band, decreased significantly at Pz in the gamma band, and increased significantly at O1 in the gamma band among patients with SZ. These changes persisted after antipsychotic treatment and were unrelated to clinical symptoms. Conclusion The abnormal hub topology we observed, especially in the high‐frequency band, may reflect the pathophysiology of SZ, and this study highlights the utility of BC analysis of EEG data for detecting alterations in the whole‐brain networks of patients with SZ.


INTRODUCTION
Schizophrenia (SZ) manifests with hallucinations, delusions, disorganized speech and behavior, as well as negative symptoms involving withdrawal, decreased motivation, and cognitive deficits. 1,2Cognitive deficits are a major field of interest in studies on SZ and are attributed to intense or continued psychiatric symptoms. 3However, these deficits are understudied and are not included in the diagnostic criteria for SZ. 4 Nevertheless, diagnostic tools for neurocognitive evaluation and years of observational studies have revealed that cognitive deficits occur in the early phase of SZ and are directly associated with poor functional outcomes. 5,6Conventional treatments, including pharmacotherapy and psychosocial treatment, have not significantly improved cognitive deficits. 7,8Although aberrant neural networks reportedly contribute to cognitive deficits, the exact underlying mechanisms remain unclear.
The notion of brain dysfunction in SZ originated from the term "schizophrenia" (schizo: divided, phrenia: mental) coined by Bleuler.
The disconnection hypothesis was proposed as an expression of the pathophysiology of SZ, and it has been validated by several clinical studies. 9,10Recently, studies in SZ have attempted to elucidate the associated abnormalities in whole-brain functional networks, 11 in addition to conventional studies that have investigated the pathophysiology of SZ from a microscopic perspective by focusing on the dopamine hypothesis or glutamate hypothesis. 12,13Functional magnetic resonance imaging (fMRI) is a widely used neuroimaging modality for investigating the functional network.However, fMRI signals are an indirect measure of neural activity and have a limited temporal resolution.Contrarily, electroencephalography (EEG) is a convenient and noninvasive technique that directly detects the electrical fields that the cortex generates with excellent temporal resolution.Therefore, EEG has been effectively used to capture instantaneous functional connectivity and has long been used to detect aberrant neural networks generated in patients with SZ.EEG dynamics at different temporal scales and frequency bands (such as the alpha, beta, gamma, and theta bands) are associated with different memory, cognitive, and perceptual functions. 14Recently, graph theory-based approaches have been applied to EEG data analysis, revealing the hub structure of whole-brain functional networks that plays an essential role in cognitive functions.SZ is characterized by abnormalities in the interactions between regions of the whole brain, therefore changes in the topological characteristics of the whole brain are more important to investigate than changes in the connectivity of individual regions.From this perspective, Takahashi et al. have used the phase lag index (PLI) to clarify anomalies in the SZ brain network in each EEG band. 15man cognitive functions are not simply based on sensory receptor signals (bottom-up information) but are also modified by memory, experience, and emotional brain signals (top-down information).7][18] The gamma-band oscillation has been associated with bottom-up processes, whereas top-down processes are likely mediated by beta-band oscillation. 19,20It has long been noted that an imbalance between top-down and bottom-up information can cause hallucinations and illusions. 213][24][25] Changes in the whole-brain network that underlie various symptoms of SZ may comprehensively explain the pathology of the disease.A more detailed analysis of these differences from healthy controls-or from patients with other psychiatric disordersmay provide objective biomarkers and information that can be used to develop efficient methods of therapeutic intervention.Previously, we identified abnormal functional connectivity in patients with SZ using graph theoretical analysis of EEG data. 15Specifically, the PLI, which captures the true synchronization of paired EEG signals, was used to discover diminished functional connectivity in the beta band across frontal regions and in the gamma band throughout the brain among patients with SZ in comparison to control participants.However, PLI evaluates the strength of the association between the two electrodes and is limited in capturing whole-brain network features.Here, we identified atypical hubs of the neural network in patients with SZ and evaluated their responses to antipsychotic medications.Accordingly, betweenness centrality (BC; a measure of hub status), which is more suitable than the PLI for evaluating the importance of each node in terms of the entire network from the viewpoint of minimum paths, was used in post-PLI analysis.

MATERIAL AND METHODS
This research involves a secondary analysis of data acquired in a prior study, 15 in which we revealed beta and gamma band-specific reductions of PLI values in SZ in comparison with HC.To avoid duplication, the results presented in previous study, including the PLI figures, are not included here, therefore the reader is referred to the previous study for details of the PLI analysis.

Participants
The clinical group comprised 21 individuals diagnosed with SZ who were recruited from the outpatient department of Kanazawa University Hospital in Ishikawa, Japan.Diagnosis of SZ was made in accordance with the criteria outlined in the Diagnostic and Statistical Manual, 4th edition (DSM-IV).Patients with concurrent neuropsychiatric conditions or those under ongoing medication were excluded.Subsequently, seven patients from the SZ group were excluded during the second EEG session, 2-6 weeks following the initiation of antipsychotic treatment, primarily due to their refusal to continue participation or to worsening of psychotic symptoms.In contrast, the healthy control (HC) group consisted of 31 individuals with no reported history of neuropsychiatric disorders.These individuals were recruited from both the staff of Kanazawa University Hospital and their family members.While SZ and HC groups were matched based on age and gender, their educational backgrounds varied.The Brief Psychiatric Rating Scale (BPRS) was employed to assess the symptoms of each patient on the day of the EEG recording, and this assessment was performed by the same clinician both before and after treatment.The demographic features of each group are outlined in Table 1.All participants who agreed to participate were made aware of the research and provided written informed consent.The Kanazawa University Ethics Committee approved the study and it adhered to the Declaration of Helsinki.

EEG recordings
The EEG data were recorded using 16 electrodes based on the International 10-20 system, which included Fp1, Fp2, F3, F4, Fz, F7, F8, C3, C4, P3, P4, Pz, T5, T6, O1, and O2.The reference electrodes were situated on the connected earlobes, and an additional electrooculogram was employed to monitor eye movements during EEG recordings.The impedance of each electrode was carefully maintained below 5 kΩ.The EEG data were sampled at 200 Hz and bandpass filtered at 1.5-60 Hz.An 18-channel system (EEG-44189; Nihon Kohden) was utilized for storing the EEG data offline.
Participants were instructed to recline in a soundproof, lightcontrolled, and electrically shielded recording room with their eyes closed.EEG signals were meticulously inspected for the identification of artifacts such as muscle activity, blinks, and eye movements.
Continuous epochs of 600 s free from artifacts were extracted from each participant's data.Subsequently, bandpass filtering was carried out for each epoch to separate the conventional frequency bands, including delta (2-4 Hz), theta (4-8 Hz), alpha (8-13 Hz), beta (13-30 Hz), and gamma (30-60 Hz).To eliminate the line noise at 60 Hz introduced by the notch filter, a bandpass was applied.A 4000th-order bandpass finite impulse response (FIR) filter with a linear phase was designed, utilizing the Hamming window.For PLI analysis, the use of excessively long epochs hinders the identification of disease-specific changes since the values decrease with increasing epoch length. 26Conversely, adopting short epochs may fail to capture behaviors characterized by slow frequency components.To address this, each epoch was subdivided into 5-s intervals for the purpose of balance. 15,27The initial and final epochs were excluded from subsequent connectivity analysis to enhance the stability of the EEG signals.

PLI
Functional brain networks were formed utilizing the PLI-a metric that attenuates artifactual interactions-to estimate functional connectivity between electrodes. 28For each epoch, the instantaneous phase at each time point of the band-passed-filtered waveform was calculated with the Hilbert transform.Subsequently, the phase difference between electrodes, t Δφ( ) T, where T is the number of time points in an epoch), was calculated.
The PLI between the pair-wise electrodes was obtained with the following equation in each epoch: PLI is a measure of the concordance rate that represents the degree of phase difference, and its values range from 0 to 1, where 0 means no coupling or zero-lag coupling, and 1 means perfect phase

BC
For both groups of participants, we calculated BC at each electrode and frequency band.BC is an indicator that can be used to identify focal nodes in a network 29 and is superior to other indicators in that it is able to assess not only neighborhood connectivity relationships, but also broader network trends, highlighting already reported SZ network abnormalities. 15Mathematically, the definition of BC is the ratio of the number of shortest paths crossing a node to the total number of shortest paths within the network.The BC estimation of functional connectivity requires the path length between electrodes; a stronger degree of coupling synchronization corresponds to a shorter distance between nodes.In other words, a higher PLI value results in a shorter path length of functional connectivity.Here, the length of a path between a pair of electrodes was defined as the inverse of the PLI value.Nodes with high BC values play an important role as hubs and correspond to bridge nodes in the network.
The BC of node i (b i ) is defined as follows: where ρ hj represents the number of shortest paths from node h to node j, ρ i ( ) hj represents the number of shortest paths passing through node i, N is the set of all nodes in the network, and n is the number of nodes.The value of b i is normalized to 0 and 1 via division by n n ( − 1)( − 2).We used MATLAB's Brain Connectivity Toolbox 30 to calculate BC.2][33][34] However, the pruning process of MST removes weak connections, which can affect the global pathways, therefore here we derived the BC from whole-brain connections to comprehensively analyze functional connectivity.

Statistical analysis
We conducted a repeated measures analysis of variance (ANOVA) to compare the pretreatment BC values between the HC and SZ, with group (HC vs. SZ before treatment) as the between-participants factor and interaction of nodes (16 electrodes) as the withinparticipants factor at each frequency band.First, the Greenhouse-Geisser adjustment was applied to the degrees of freedom.A two-sided α of 0.05 was considered statistically significant to avoid type I errors.Then, we performed post hoc ttests controlled by the Benjamini-Hochberg false discovery rate (FDR) procedure 35 for the BC values of all electrodes (HC vs. SZ before treatment) to identify the BC values with significant main and interaction effects.Following this, t values adjusted to q < 0.05 against the frequency bands with significant effects were used for further analysis.We employed a repeated measures ANOVA with treatment condition (SZ before treatment vs. SZ after treatment) as a within-participant factor and interaction of nodes (16 electrodes) as a within-participant factor at each frequency band.
The Greenhouse-Geisser adjustment was applied to compare the pre-and posttreatment BC values in the SZ group to the degrees of freedom, and a two-sided α of 0.05 was considered statistically significant.Next, BC values with significant main (treatment) and interaction (treatment × node) effects were used for post hoc paired t-tests controlled by the FDR procedure. 35Following this, t values adjusted to q < 0.05 against the frequency bands with significant main (treatment) and interaction (treatment × node) effects were used for further analysis.Finally, we also evaluated the relationship between changes in BPRS and improved BC values by calculating Pearson's correlation coefficient (R) between these factors.A significant correlation was shown by P < 0.05.
We calculated Pearson's correlation coefficient (R) between these factors to evaluate the relationship between BC values and the severity of disease in SZ (as measured by BPRS scores), and between BC values and age.A significant correlation was indicated by P < 0.05.

RESULTS
The connectivity indices obtained before calculating BC (i.e., the PLI) were validated in our previous study. 15We evaluated the betweengroup differences in pretreatment BC values in the delta, theta, alpha, beta, and gamma bands (Table 2).We found a significant main effect of group in the gamma band and a significant interaction effect with node in the beta and gamma bands.To identify the BC values with significant main and interaction effects, we visualized mean pretreatment BC values in the HC and SZ groups in the beta and gamma T A B L E 2 Results of a repeated measures ANOVA for betweenness centrality differences between the healthy control and pretreatment schizophrenia groups in the delta, theta, alpha, beta, and gamma bands.bands (Figure 1a).We visualized the PLI values for each frequency band for each electrode pair with connectivity strength in the top 20%.The Fz and Pz electrodes showed high BC values in the beta and gamma bands in both groups (HC and pretreatment SZ).The results of the post hoc t-test are shown in Figure 1b.The findings demonstrated a notable decrease in the BC value of Fz in the beta band (q < 0.05), a decrease in the BC value of Pz in the gamma band (q < 0.05), and an increase in the BC value of O1 in the gamma band (q < 0.05).We generated a scatter plot of symptom severity (BPRS scores) versus BC and calculated the Pearson's correlation coefficient (R) (Figure 2), observing no significant correlations.We used the same process to evaluate the relationship between BC and age, and again found no significant correlations.In the area of interest listed above (Fz in the beta band, Pz in the gamma band, and O1 in the gamma band), weak correlations of BC and age were seen only in HC gamma.
Evaluating the relationship between these BC values and BPRS scores in the SZ group for O1 gave ρ = −0.45 and P = 0.11.
We also evaluated the differences in the SZ group's pre-and posttreatment BC values (Table 3).However, we found no significant main (treatment) or interaction (treatment × node) effect, nor any significant association between BPRS change and BC values.

DISCUSSION
We analyzed EEG data using PLI-based BC values (a measure of hub status) to examine the changes in resting-state functional connectivity in drug-naïve patients with SZ.Our results revealed frequency-and region-specific abnormalities in the hubs of neural networks, and the results were especially evident in the high-frequency bands.The clinical symptoms of SZ (as measured by BPRS) were not correlated with BC values, and we observed no significant differences in BC values before and after medication.
Functional connectivity patterns in patients with SZ are reportedly atypical, but findings have been inconsistent.This may be due to differences in the methods of analysis used or patient backgrounds (including the effects of medication and disease symptoms). 36In particular, EEG studies have reported a significant impact of medication on network analysis results. 37,38Additionally, Takahashi et al. showed that in patients with SZ, abnormal EEG signal complexity in the forebrain region (as confirmed by multiscale entropy analysis) was selectively normalized by antipsychotic treatment. 39Generally, these results show that functional connectivity is directly affected by medication, therefore analyzing data from drug-naïve patients and evaluating the effects of medication on whole-brain networks may provide useful insights into the pathophysiology of the disease.
Patients with SZ show significant changes in their brain network organization, as indicated by graph-analytical measurements of global short communication paths, local organization, and smallworldness. 40A growing number of studies, including both anatomical and neurophysiological approaches using diffusion tensor imaging (DTI), fMRI, and EEG, have focused on hub status and have identified atypical network patterns.The importance of abnormalities in brain hubs is established and such abnormalities comprise the background of many psychiatric disorders. 41Regarding anatomical alterations of white matter structure examined by DTI, connections between hub regions comprising the "rich club" were disproportionately affected in patients with SZ, a difference that was not ameliorated with medication. 42In an fMRI study, frontal hubs of patients with SZ showed a significant reduction in BC values. 43Cheng et al. also showed that changes in functional hubs are associated with SZ and that BC values can identify patients with SZ with a high level of accuracy. 44The decreased BC values in the frontal lobes observed in these studies are consistent with our results.Furthermore, Crossley et al. 41 performed a meta-analysis of 314 task-based functional neuroimaging studies.They reported that patients with SZ showed abnormal hub structure and decreased hub function in specific brain regions.They indicated that these abnormalities increased compensatory activity in other brain regions. 45The established anatomical locations of under-and overactivations in SZ are consistent in crucial respects with our results.
In contrast to fMRI studies, relatively few EEG studies examine the hub structure of whole-brain networks.Krukow et al. analyzed   the EEG signals of first-episode patients with SZ and demonstrated increased gamma band BC in posterior nodes and a correlation between these BC values and cognitive deficits. 46This is consistent with our findings of increased BC in O1.However, Krukow et al.
found no differences in other frequency bands or nodes between patients with SZ and controls.In a comparison of neural network topologies between patients with first-and multi-episode SZ, 47 significant between-group differences in maximal BC and tree hierarchy were observed in both the beta and the gamma bands.
On the basis of these results, the authors showed that the duration of illness significantly affects the topology of resting-state functional networks.Despite regional variations, their findings align with ours, indicating noteworthy group differences solely in the high-frequency bands.
Our findings of frequency-and region-specific abnormalities in neural network hubs may be interpreted as follows.7][18] An imbalance between these can cause hallucinations and illusions. 213][24][25] Recent findings show that bottom-up and top-down signaling use gamma and beta frequencies, respectively.Uhlhaas et al. reported that in the primate visual system, feedforward (bottom-up) influences are carried by thetaband and gamma-band synchronization, whereas feedback (top-down) influences are carried by beta-band synchronization. 14,20 for top-down and bottom-up at the hub of a functional network. 53Fifth, despite the evidence of a medication effect on functional connectivity, we did not find a significant difference in BC before and after treatment.
However, this result may be due to the limitations of neuroimaging modalities and analysis methods.Additionally, the small sample sizes of the SZ groups, notably of the posttreatment group, may have admitted selection bias and influenced the result.Sixth, due to the measurement environment, the EEG data in the very high-frequency band was not sufficient to support division of the gamma band into high-gamma and low-gamma bands for analysis.The effect of gamma subband distinctions on our main outcomes should therefore be investigated in further studies.
Finally, regarding technical issues, applying Laplace re-reference was appropriate here to weaken the common source problem.However, PLI analysis achieves relatively high spatial resolution without re-reference, 15 and re-reference is not always necessary, notably in an extended version of PLI called wPLI. 54However, the actual effect of the re-reference on the detection of the hub structure at the level of the global topology has not yet been evaluated, therefore detailed verification our main results with re-referencing will be necessary in future studies.

CONCLUSION
We report a significant alteration of hub structure in the neural network of drug-naïve patients with SZ, as evaluated using restingstate EEG rhythms in the high-frequency bands.Despite clinical T A B L E 3 Pre-versus posttreatment differences in betweenness centrality within the schizophrenia group: Results of a repeated measures ANOVA in the delta, theta, alpha, beta, and gamma bands.
coupling.In particular, with low PLI values, t Δφ( ) k distributes homogeneously in the phase space or at around zero and π [rad], while with high PLI values, t Δφ( ) k distributes at around a specific phase difference that is not zero or π [rad].
Mean betweenness centrality (BC) values in the healthy control (HC) and pretreatment schizophrenia (SZ) groups in the beta and gamma bands.Phase lag index (PLI) values with strength in the top 20% at each frequency band are shown for each electrode pair.The width of the circles corresponds to the strength of the PLI value.In both groups (HC and SZ), the Fz and Pz electrodes show high BC values in the beta and gamma bands.(b) the t values of comparisons between the pretreatment SZ and HC groups in the beta and gamma bands ARE SHOWN.High (low) t values correspond to high (low) BC values in the pretreatment SZ group compared with those in the HC group.The results show a significant decrease in the BC values of Fz in the beta band (q < 0.05, blue arrow), a decrease in the BC values of Pz in the gamma band (q < 0.05, blue arrow), and an increase in the BC values of O1 in the gamma band (q < 0.05, red arrow).WHOLE-BRAIN FUNCTIONAL NETWORK OF SCHIZOPHRENIA | 5 of 10 WHOLE-BRAIN FUNCTIONAL NETWORK OF SCHIZOPHRENIA | 7 of 10 evidence demonstrating the positive effects of antipsychotic medication, the observed alterations were not ameliorated by treatment with antipsychotics.Our findings are likely related to abnormalities in the top-down and bottom-up processes that may contribute to cognitive deficits in SZ.These findings must be validated by longitudinal studies evaluating specific clinical symptoms and cognitive deficits associated with SZ.This study highlights the potential of a BC index based on conventional EEG data, and we demonstrate its performance in elucidating the pathophysiological bases and therapeutic mechanisms of SZ.
Scatter plots of pretreatment betweenness centrality values versus BPRS scores in the schizophrenia group, with Pearson's correlation coefficients (R).No significant correlations are observed.BPRS, Brief Psychiatric Rating Scale.
BPRS scores).Thus, our finding of frequency-specific treatment-refractory abnormal hub topologies in SZ may reflect cognitive deficits in this disorder.Our results demonstrate abnormalities in whole-brain functional networks consistent with the disconnection hypothesis and indicate the locations of important anomalies associated with SZ.This study has some limitations that should be considered.First, EEG data have limited spatial resolution.Additionally, our EEG measurements were performed according to typical protocols followed in real-world In addition, given the small number of participants in this study, it is difficult to evaluate complex interactive factors.Thus, further studies that include a larger number of participants are warranted to permit the statistical analysis of different backgrounds, and longer-term longitudinal studies are required to interpret the present results accurately.Fourth, the notion of top-down and bottom-up processes requires careful discussion because BC does not reflect the directionality of neural signal propagation.It is therefore necessary to evaluate the directional functional connectivity with directed PLI to reveal the information flow