Aberrant Cortical Morphological Networks in First-Episode Schizophrenia

Abstract

Background and Hypothesis

Population-based morphological covariance networks are widely reported to be altered in schizophrenia. Individualized morphological brain network approaches have emerged recently. We hypothesize that individualized morphological brain networks are disrupted in schizophrenia.

Study Design

We constructed single-subject morphological brain networks for 203 patients with first-episode schizophrenia (FES) and 131 healthy controls separately based on regional cortical thickness (CT), fractal dimension (FD), gyrification index, and sulcal depth (SD) by dividing the cerebral cortex into 360 regions in terms of the Human Connectome Project Multi-Modal Parcellation atlas.

Results

Compared with the controls, the patients exhibited morphological similarity reductions in all types of networks while increases in FD- and SD-based networks. The altered morphological similarities were commonly involved in cingulo-opercular and default mode networks. Interestingly, the altered morphological similarities accounted for clinical symptoms and cognitive dysfunction in the patients and distinguished the patients from controls, with better performance than altered local morphology. In addition, graph-based analysis revealed that global organization was intact while nodal centrality was altered in the patients as characterized by decreased degree and efficiency in the left inferior parietal cortex and increased efficiency in left area superior temporal gyrus for the CT-based networks, decreased degree and efficiency in the left Posterior Insular Area 2 for the FD-based networks, and decreased betweenness in the left Area 52 for the SD-based networks.

Conclusions

These findings indicate that FES is accompanied by characteristic disruptions in single-subject cortical morphological networks, which provide novel insights into neurobiological mechanisms underlying schizophrenia.

Introduction

Schizophrenia is characterized by a lack of integration between thought, emotion, and behavior.¹ Besides positive and negative symptoms, cognitive deficits are a major contributor to poor functional outcomes in schizophrenia.^2,3 Evidence from neuroimaging studies has shown that cerebral cortical and subcortical abnormalities are related to clinical symptoms and cognitive deficits in individuals with schizophrenia.^4–8 Therefore, characterizing structural and functional brain alterations in schizophrenia is of great significance for understanding the neurobiological substrate of this disease and ultimately preventing cognitive impairment in patients.

Schizophrenia-related morphological alterations have been well documented in the literature.⁹ For example, widespread cortical thinning and smaller surface area are frequently reported in individuals with schizophrenia, particularly in the prefrontal and temporal cortices.⁹ Beyond local morphological alterations, several studies have further shown that schizophrenia is associated with disrupted morphological brain networks.^10,11 However, morphological brain networks in these studies are constructed by estimating interregional covariance across participants in a certain morphological index.^12,13 The population-based method provides only 1 network for a group of participants and thus is difficult to be used to identify biomarkers for schizophrenia. In recent years, such population-based morphological brain networks can be mapped at the single-subject level,¹⁴ largely extending the application scenarios of morphological brain networks. By individualized modeling, single-subject morphological brain networks are demonstrated to be of great value in helping clinical diagnosis and prognosis of brain disorders.¹⁵ Thus, several studies have applied the individualized modeling methods to schizophrenia and found that single-subject morphological brain networks facilitate delineating abnormal organization,¹⁶ identifying biological subtypes,¹⁷ finding clinical correlations,¹⁸ and predicting treatment outcomes.¹⁹ These findings bring considerable benefits to our understanding of the pathology and identification of biomarkers of schizophrenia. However, most of the studies focus on patients with chronic schizophrenia, which may bring bias in the results due to previous antipsychotic medication exposure. In contrast, patients with first-episode schizophrenia (FES) are ideal research samples to rule out possible confounding effects of long-term medication, allowing us to reveal brain network alterations that relatively rarely associated with the pathology of the disease. To date, there is only 1 study examining FES-related alterations in single-subject morphological brain networks, which are constructed based on regional gray matter volume.¹⁷ Since regional gray matter volume is thought to reflect a composite of multiple morphological indices,^20,21 the gray matter volume-based single-subject morphological brain networks may overlook specific alterations in FES. To more comprehensively and precisely characterize FES-related alterations, it is necessary to employ different nonredundant morphological indices to construct single-subject morphological brain networks.

This multicentric study aimed to systematically examine group-level alterations of single-subject morphological brain networks in FES. Specifically, we utilized different morphological indices to construct 4 types of single-subject morphological brain networks for 203 patients with FES and 131 matched healthy controls (HCs) from 3 sites by estimating interregional morphological similarity among 360 regions of interest (ROIs).²² For each type single-subject morphological brain networks, we examined FES-related alterations not only in interregional morphological similarity between each pair of ROIs but also in graph theoretical network measures derived from the entire single-subject morphological brain network. Graph theory provides a set of well-defined measures to quantify the system-level organization of a collection of interconnected regions as a whole.²³ Thus, the edge- and system-level comparisons were complementary to each other to allow characterizing FES-related alterations in single-subject morphological brain networks in a comprehensive manner. For significant alterations in the patients, we further explored their clinical relevance by investigating their relationships with clinical symptoms and neuropsychological tests of the patients and their abilities to differentiate the patients from controls. Finally, considering that interregional morphological similarity was estimated based on regional morphology, we examined whether the alterations in interregional morphological similarity performed better than alterations in local morphology in accounting for clinical symptoms and cognitive dysfunction of the patients and in distinguishing the patients from controls. All these analyses went hand in hand and were directly related to the aim of this study. Figure 1 illustrates the schematic representation of the main analytical process in this study.

Figure 1.

A Flowchart of the Analysis Process in This Study. Briefly, 4 Vertexwise Morphological Maps Were First Extracted From the Structural Image of Each Participant. Each Morphological Map Was Then Divided Into 360 Regions According to the HCP-MMP Atlas. For Each Region, the Mean Value Was Calculated for Each Morphological Map and the Resultant Values Were Used to Examine Between-Group Differences in Local Morphology. Meanwhile, All Values Within Each Region Were Extracted for Each Morphological Map, Which Was Used to Estimate Interregional Morphological Similarity in Terms of the Distribution of the Values. This Resulted in 4 Morphological Similarity Matrices for Each Participant, Which Were Used to Examine Between-Group Differences in Interregional Morphological Similarity. Finally, Each Morphological Similarity Matrix Was Thresholded Into a Series of Binary Networks or Graphs, for Which Multiple Grapy Theoretical Measures Were Calculated. The Grapy Theoretical Measures Were Used to Examine Between-Group Differences in the System-Level Organization of the Collection of Interconnected Regions as a Whole. Abbreviations: CT, Cortical Thickness; FD, Fractal Dimension; FES, First-Episode Schizophrenia; GI, Gyrification Index; HCP-MMP, Human Connectome Project Multi-Modal Parcellation; HCs, Healthy Controls; JSDs, Jensen–Shannon Divergence-Based Similarity; PD, Probability Distribution; SD, Sulcal Depth

Methods

Participants and Clinical Symptom Evaluation

This multicentric study included a total of 203 patients with FES (male/female, 95/108; average age = 24.7 ± 5.6 years old) and 131 HCs (male/female, 73/58; average age = 27.8 ± 6.0 years old) from 3 clinical sites, including Chongqing Three Gorges Central Hospital, Chongqing, China (Dataset 1: 95 patients [average age = 24.7 ± 6.1 years old] and 49 HCs [average age = 25.3 ± 5.4 years old]), Zhumadian Psychiatry Hospital, Henan Province, China (Dataset 2: 75 patients [average age = 24.4 ± 4.7 years old] and 39 HCs [average age = 31.9 ± 6.5 years old]), and Beijing Huilongguan Hospital, Beijing, China (Dataset 3: 35 patients [average age = 25.7 ± 5.9 years old] and 42 HCs [average age = 27.1 ± 4.1 years old]). The patients met the following criteria: (1) DSM-IV (American Psychiatric Association, 1994) diagnostic criteria for schizophrenia; (2) inpatients or outpatients: First outpatient treatment or first hospitalization of less than 2 weeks; (3) at least 6 years of education; (4) right-handed, confirmed by the short version of the Edinburgh Handedness Scale; and (5) aged 15 years old and above. The exclusion criteria for the patients and HCs included: (1) a history of head trauma; (2) concurrent or previous substance dependence, besides smoking, or alcoholism; (3) gross brain organic disease confirmed on T2 MRI; (4) symptoms of significant involuntary movement; and (5) learning disability or intellectual disability. The HCs had no family history of psychotic illnesses, according to the Family History Research Diagnostic Criteria. The patients had no previous chronic antipsychotic medication exposure, and were recruited immediately after stabilization of psychosis symptoms after a sufficient but minimal duration of antipsychotic medications. All participants gave written informed consent and this study was approved by the Ethics Committee of Beijing Huilongguan Hospital, Beijing, China.

Patients in each site were evaluated using the Positive and Negative Syndrome Scale (PANSS)^24,25 by an attending psychiatrist, with inter-rater reliability measured with intraclass correlation coefficient values of 0.80 or above before the trial. Medication dosages (chlorpromazine equivalents) were calculated for each patient.^26,27 Cognitive function was measured using the validated Chinese versions of MATRICS Consensus Cognitive Battery (MCCB),²⁸ and composite T-scores were computed.

Imaging Protocol

All participants underwent multimodal MRI on a 3T scanner in their own site. In this study, only T1-weighted structural images were used. Other modalities have been published previously.^8,29

Dataset 1: Participants in the Dataset 1 were scanned on a Siemens 3T Trio MRI scanner with a sagittal 3D-magnetization prepared rapid acquisition gradient echo (MPRAGE) sequence: echo time (TE) = 2.98 ms, inversion time (TI) = 900 ms, repetition time (TR) = 2300 ms, flip angle (FA) = 9°, the field of view (FOV) = 240 × 256 mm², matrix size = 256 × 240, and thickness/gap = 1/0 mm.

Dataset 2: Participants in Dataset 2 were scanned on a GE 3T MRI scanner with a sagittal 3D-MPRAGE sequence: TE = 2.49 ms, TI = 1100 ms, TR = 6.77 ms, FA = 7°, FOV = 256 × 256 mm², matrix size = 256 × 256, and thickness/gap = 1/0 mm.

Dataset 3: Participants in Dataset 3 were scanned on a Siemens 3T Prisma MRI scanner with a sagittal 3D-MPRAGE sequence: TE = 2.98 ms, TI = 1100 ms, TR = 2530 ms, FA = 7°, FOV = 256 × 224 mm², matrix size = 256 × 224, and thickness/gap = 1/0 mm.

Structural MRI Data Processing

All structural MRI images underwent a standard preprocessing pipeline using the CAT12 toolbox (http://www.neuro.uni-jena.de/cat/) based on the SPM12 software (https://www.fil.ion.ucl.ac.uk/spm/software/spm12). Briefly, each structural MRI image was first segmented into gray matter, white matter, and cerebrospinal fluid. Then, 4 morphological indices were computed, including cortical thickness (CT), fractal dimension (FD), gyrification index (GI), and sulcal depth (SD). These 4 morphological indices offer nonredundant descriptions of cortical characteristics, are widely used to explore cortical architecture in brain development, aging, and disease, and are computationally fast in the CAT12 toolbox. More importantly, our previous studies have demonstrated that single-subject morphological brain networks constructed with these 4 indices exhibit distinct connectivity patterns and topological organizations,³⁰ display different sensitivities in detecting disease- and age-related alterations,^31–33 and have different neurobiological substrates.³⁴ Specifically, CT was estimated using a projection-based thickness method,³⁵ and the central cortical surface was reconstructed. After repairing the topological defects in the central surface using a spherical harmonics-based method,³⁶ FD, GI, and SD were calculated as the slope of the linear part between the area and the maximum l-value,³⁷ the absolute mean curvature,³⁸ and the Euclidean distance between the central surface and its convex hull, respectively. Finally, the resultant maps of CT, FD, GI, and SD were resampled into a common fsaverage template, and smoothed using a Gaussian kernel. According to the recommendations of the CAT12 manual, individual CT maps were smoothed using a Gaussian kernel with a 12-mm full width at half maximum, while individual FD, GI, and SD maps were smoothed using a Gaussian kernel with a 25-mm full width at half maximum. The usage of larger smoothing kernel sizes for the FD, GI, and SD maps is due to the underlying nature of these folding measures that reflect contributions from both sulci and gyri. Therefore, the filter size should exceed the distance between a gyral crown and a sulcal fundus.

Construction of Single-Subject Morphological Brain Networks

Currently, there are numerous methods to constructed single-subject morphological brain networks.¹⁴ Among these methods, the divergence-based one is gaining popularity because of its flexibility in the definition of network nodes and edges and its ability to make use of voxel- or vertex-level data with brain regions. Furthermore, single-subject morphological brain networks constructed with the divergence-based method are found to account for interindividual variation in behavior and cognition, to be related to genetic, cytoarchitectonic, and chemoarchitectonic factors, and to align with axonal connectivity,^34,39–41 indicative of their biological meanings. Thus, we employed the divergence-based method to construct single-subject morphological brain networks in this study. The key idea of the divergence-based method is to estimate a probability distribution from a morphological feature within a brain region and then quantify the similarity in the probability distributions between regions with a divergence measure. To date, only the Kullback-Leibler divergence and Jensen-Shannon divergence have been used to estimate interregional morphological similarity in previous studies.^{30,34,41–43} Compared with the Kullback-Leibler divergence, the Jensen-Shannon divergence has several mathematical advantages, such as symmetry and boundedness. Moreover, our previous studies have demonstrated that single-subject morphological brain networks constructed with the Jensen-Shannon divergence have significantly higher test-retest reliability than those constructed based on the Kullback-Leibler divergence.^30,44 Therefore, we utilized the Jensen-Shannon divergence to estimate interregional morphological similarity in this study.

Specifically, using the Jensen-Shannon divergence-based method, we constructed 4 types of single-subject morphological brain networks (ie, CT-based networks [CTNs]; FD-based networks [FDNs]; GI-based networks [GINs]; and SD-based networks [SDNs]) for each participant in this study. First, the cerebral cortex was parceled into 360 ROIs in terms of the Human Connectome Project Multi-Modal Parcellation (HCP-MMP) atlas.²² These ROIs can be categorized into 12 modules according to their resting-state functional connectivity patterns⁴⁵: Primary visual, secondary visual, somatomotor, cingulo-opercular, dorsal attention, language, frontoparietal, auditory, default mode, posterior multimodal, ventral multimodal, and orbito-affective. Then, for each type of morphological map (eg, CT maps), all values in each ROI were extracted for each participant and used to obtain a probability density function by a kernel density estimation method (MATLAB function, ksdensity). After converting the resultant probability density estimates to probability distributions, the morphological similarity between each pair of ROIs was estimated by calculating the Jensen-Shannon divergence-based similarity (JSDs):

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{J}\mathrm{S}\mathrm{D}\mathrm{s}(P,Q)=1-\sqrt{\mathrm{J}\mathrm{S}\mathrm{D}(P||Q)}}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle M=\frac{1}{2}(P+Q)}\end{array}}$$

$${\displaystyle \begin{array}{l}{\displaystyle \mathrm{J}\mathrm{S}\mathrm{D}(P||Q)=\frac{1}{2}{\displaystyle \sum _{i=1}^n}P(i)\mathrm{l}\mathrm{o}\mathrm{g}\frac{P(i)}{M(i)}+\frac{1}{2}{\displaystyle \sum _{i=1}^n}Q(i)\mathrm{l}\mathrm{o}\mathrm{g}\frac{Q(i)}{M(i)}}\end{array}}$$

where P and Q denote regional probability distributions, and n denotes the number of sampling points (2⁸ in this study).⁴³ The JSD value indicates the extent to which 2 probability distributions are similar (0 indicating completely different and 1 exactly the same).

Removal of Site Effects

In this study, we utilized a harmonization approach called ComBat to moderate site effects on interregional morphological similarities. ComBat harmonization is demonstrated to successfully remove inter-site technical variability while preserving inter-site biological variability in image-based measurements.⁴⁶ Specifically, the ComBat model can be written as:

$${\displaystyle \begin{array}{l}{\displaystyle y_{ijv}={\alpha }_v+X_{ij}{\beta }_v+{\gamma }_{iv}+{\delta }_{iv}{\epsilon }_{ijv}}\end{array}}$$

where y_ijv represents the connectivity strength of edge v for subject j in site i, α_v is the average connectivity strength for edge v, X is a design matrix for the covariate (ie, group indicator variable), β_v is a vector of regression coefficients corresponding to covariates in X and ε_ijv is the residual term that is assumed to follow a normal distribution with zero means. The terms γ_iv and δ_iv represent the additive and multiplicative site effects of the site i on edge v, respectively, and are estimated by conditional posterior means as described in previous studies.⁴⁷ The final ComBat-harmonized connectivity strength for edge v is calculated as:

$${\displaystyle \begin{array}{l}{\displaystyle y_{ijv}^{\mathrm{C}\mathrm{o}\mathrm{m}\mathrm{B}\mathrm{a}\mathrm{t}}=\frac{y_{ijv}-{{\displaystyle \widehat{\alpha }}}_v-X_{ij}{{\displaystyle \widehat{\beta }}}_v-{\gamma }_{iv}^{\ast }}{{\delta }_{iv}^{\ast }}+{{\displaystyle \widehat{\alpha }}}_v+X_{ij}{{\displaystyle \widehat{\beta }}}_v}\end{array}}$$

where ${\gamma }_{iv}^{\ast }$ and ${\delta }_{iv}^{\ast }$ are the empirical Bayes estimates of γ_iv and δ_iv, respectively.

Network Analysis

All network analyses, including threshold selection and network parameter calculation, were conducted with the open-source GRETNA toolbox.⁴⁸

Threshold Selection

In this study, a sparsity-based thresholding procedure was used to exclude connections with low morphological similarities in each morphological similarly matrix. Sparsity is defined as the ratio of the number of actual edges to the maximum possible number of edges in a network. The sparsity-based thresholding procedure thus ensures the same number of edges across participants by applying subject-specific JSDs thresholds. Given the absence of a definitive method for selecting a single sparsity, a set of binary morphological brain networks were obtained from each morphological similarity matrix with sparsities ranging from 0.02 to 0.3 (interval = 0.02). This sparsity range was determined to ensure that the resultant binary networks were estimable for small-worldness and had sparse properties.^49–51 The lower limit of the sparsity range is determined to ensure that the resultant binary networks are estimable for the small-world parameters.⁵¹ That is, for each binary network, the average degree over all nodes should be larger than 2 × log(N), with N denoting the number of nodes in the network (ie, 360 in this study). This condition is important because only when it is satisfied, the graph-based network parameters used in this study are computationally feasible. As for the upper limit of the sparsity range, it is empirically chosen to guarantee that the resultant binary networks have sparse properties, consistent with our previous studies.^40,41

Network Parameter Calculation

For each binary morphological brain network derived above, we calculated global small-world parameters (clustering coefficient, C_p and characteristic path length, Lp) and local nodal centrality metrics (degree, k_i; efficiency, e_i; and betweenness, b_i) with the GRETNA toolbox.⁵² Details on the formula and interpretation of the network parameters can be found elsewhere.^23,53 The small-world parameters were further normalized by the mean of corresponding parameters derived from 100 random networks, which were generated via the most widely used topological rewiring procedure to preserve the same degree distributions as the real networks.⁵⁴ Given that all network parameters were calculated as functions of sparsity, we finally calculated the area under the curve (AUC; ie, the integral over the entire sparsity range) for each network parameter to provide summarized scalars for subsequent statistical analysis.

Statistical Analysis

In this study, all statistical analyses related to MRI-derived measures were conducted with nonparametric permutation testing because they were typically not normally distributed.

Between-Group Differences in Demographic, Clinical, and Neuropsychological Variables

For the discrete sex data, a χ² test was used to examine their between-group differences. For continuous variables, 2-sample t-tests or Wilcoxon rank sum tests were used to examine their between-group differences depending on whether they followed normal distributions (Lilliefors tests).

Between-Group Differences in Interregional Morphological Similarities

To examine between-group differences in interregional morphological similarities, a network-based statistic (NBS) method was used.⁵⁵ Briefly, for each type of morphological similarity matrices, a t-statistic matrix was first derived via edgewise between-group comparisons (2-sample t-tests) with age, sex, and education as covariates. A primary significance threshold was then applied to the t-statistic matrix to select suprathreshold connections, among which all connected components were identified with their sizes (ie, number of links) recorded. Here, P < .001 was used because it was recommended as a primary threshold for traditional cluster extent-based thresholding to improve spatial specificity and control for false positives.⁵⁶ To estimate the significance of each identified component, a null distribution of the connected component size was empirically derived using a nonparametric permutation approach (10 000 permutations). For each permutation, all participants were randomly rearranged into 2 groups, and the same primary significance threshold (ie, P < .001) was used to filter suprathreshold connections from the edgewise comparisons between the 2 randomized groups. The size of the maximal connected component among the suprathreshold connections was recorded to form the null distribution. Finally, for any connected component of size M derived from the comparison of the real groups, a corrected P value was determined by calculating the proportion of the 10 000 permutations for which the maximal connected component was larger than M.

Between-Group Differences in Inter-subnetwork Morphological Similarities

Given that the ROIs included in the HCP-MMP belonged to 12 subnetworks, we further examined FES-related alterations in morphological similarities at the subnetwork level. First, each single-subject morphological similarity matrix was converted into a 12 × 12 matrix, in which the elements represented the mean morphological similarity across edges within and between the subnetworks. Then, between-group differences in the mean morphological similarities were statistically inferred using a nonparametric permutation test (10 000 permutations) based on the t-statistic derived from a 2-sample t-test. Age, sex, and education were treated as covariates in the t-tests. Multiple comparisons were corrected with the false discovery rate (FDR) procedure at the level of q < 0.05.

Between-Group Differences in Graph-Based Network Parameters

For each graph-based network parameter, between-group differences were statistically inferred using a nonparametric permutation test (10 000 permutations) based on the t-statistics derived from a 2-sample t-test. Age, sex, and education were treated as covariates in the t-tests. For each type of single-subject morphological brain network, multiple comparisons were corrected separately for global small-world parameters and each local nodal centrality metric with an FDR procedure at the level of q < 0.05.

Between-Group Differences in Regional Morphology

We examined between-group differences in regional morphology using a nonparametric permutation test (10 000 permutations) based on the t-statistics derived from a 2-sample t-test. Age, sex, and education were treated as covariates in the t-tests. For each type of the 4 indices (CT, FD, GI, and SD), multiple comparisons were corrected separately with an FDR procedure at the level of q < 0.05.

Relationships Between MRI-Based Measures and Clinical and Neuropsychological Variables

For each graph-based network parameter, and intra-/inter-subnetwork connection showing significant alterations in the patients, Spearman partial correlation was used to examine its relationship with the PANSS and MCCB scores in the patients. Effects of sex, age, and education were controlled. Multiple comparisons were corrected with the FDR procedure at the level of q < 0.05. For each NBS component showing significantly altered morphological similarities in the patients, partial least-squares (PLS) regression was used to examine the relationship of edges in the component with the PANSS and MCCB scores in the patients. The PLS regression is a well-known dimensionality reduction technique by projecting the input data (predictor variables; edges in each NBS component here) and output data (dependent variable; PANSS or MCCB scores here) to latent variable space. In this study, we only examined the first component of the PLS (PLS1), which was the linear combination of the predictor variables that exhibited the strongest correlation with the response variable. Multiple comparisons were corrected with the FDR procedure at the level of q < 0.05.

Classification Between Patients and Controls

For each graph-based network parameter, intra-/inter-subnetwork connection, and NBS component showing significant alterations in the patients, the receiver operating characteristic curve was plotted to examine whether they might serve as potential biomarkers for differentiating the patients from HCs. Specifically, for a given measure, many different thresholds were used to classify each participant into either patient or control group. For each threshold, the fraction of correctly identified patients (ie, sensitivity or true positive rate) and the fraction of correctly identified controls (ie, specificity or true negative rate) were calculated. Finally, a cutoff point that simultaneously optimized the sensitivity and specificity was determined and the classification accuracy at the cutoff point was calculated as the fraction of correctly identified participants. This procedure was performed using the public MATLAB codes (https://ww2.mathworks.cn/matlabcentral/fileexchange/19950-roc-curve).

Results

Demographic, Clinical, and Cognitive Data

The demographic, clinical, and cognitive characteristics of all participants are shown in Table 1. Age and education significantly differed between the FES patients and HCs (P < .001). Compared with the HCs, the FES patients exhibited significantly lower MCCB scores (P < .001).

Table 1.

Demographic and Clinical Characteristics of Patients of All Participants

	FES (n = 203)	HCs (n = 131)	χ 2/t	P
Demographic information
Sex (M/F)	95/108	73/58	2.4	.12
Age (y)	24.75 ± 5.58	27.82 ± 6.02	4.77	3.00 × 10−6
Education (y)	10.61 ± 3.17	13.24 ± 3.64	6.96	1.82 × 10−11
Illness duration (y)	0.96 ± 1.06	N/A
Age at onset (y)	23.83 ± 5.63	N/A
Duration of untherapy (y)	0.70 ± 1.01	N/A
Clinical symptoms
PANSS positive	21.47 ± 6.51	N/A
PANSS negative	17.56 ± 7.01	N/A
PANSS general psychosis	37.94 ± 8.40	N/A
PANSS total	76.96 ± 17.03	N/A
Cognitive function
MCCB	36.77 ± 10.27	53.65 ± 7.21	15.70	2.91 × 10−41

Abbreviations: FES, first-episode schizophrenia; HCs, healthy controls; MCCB, MATRICS Consensus Cognitive Battery; PANSS, Positive and Negative Syndrome Scale.

Disruption of Interregional Morphological Similarities in FES

Compared with the HCs, the patients with FES exhibited abnormal morphological similarities in all types of single-subject morphological brain networks (Figure 2 and Supplementary Figure S1). Specifically, an NBS component was identified for the CTNs that showed significantly decreased morphological similarities in the patients with FES (P = .014, NBS corrected). This component contained 88 nodes and 125 connections with the connections mainly involving the secondary visual (58.0%), cingulo-opercular (53.6%), frontoparietal (23.2%), and default mode (27.7%) modules (Figure 3). For the FDNs, 2 NBS components were identified to show significantly decreased (P < .001, NBS corrected) and increased (P = .014, NBS corrected) morphological similarities in the patients with FES, respectively. The component showing decreased morphological similarities included 114 nodes and 166 connections with the connections mainly in the default mode (62.0%), frontoparietal (41.7%), somatomotor (39.6%), and cingulo-opercular (25.1%) modules and the component showing increased morphological similarities contained 179 nodes and 284 connections with the connections mainly in the cingulo-opercular (61.5%), default mode (47.9%), and frontoparietal (30.2%) modules (Figure 3). With respect to the GINs, an NBS component was identified to show significantly decreased morphological similarities in the patients with FES (P = .033, NBS corrected), which included 79 nodes and 114 connections with the connections mainly linking the cingulo-opercular (66.7%), secondary visual (38.5%), and default mode (34.7%) modules (Figure 3). Finally, 2 NBS components were identified for the SDNs that exhibited significantly decreased (P = .031, NBS corrected) and increased (P = .035, NBS corrected) morphological similarities in the patients with FES, respectively. The component showing decreased morphological similarities contained 85 nodes and 170 connections with the connections mainly linking the default mode (50.3%), cingulo-opercular (34.4%), and somatomotor (26.1%) modules and the component showing increased morphological similarities contained 125 nodes and 178 connections with the connections mainly in the cingulo-opercular (56.3%), default mode (44.7%), language (24.5%), and somatomotor (22.0%) modules (Figure 3).

Figure 2.

Altered Interregional Morphological Similarities in FES. Compared With the HCs, the FES Patients Showed Significantly Decreased Morphological Similarities in Each Type of Single-Subject Morphological Brain Network Increased Morphological Similarities in the FDNs and SDNs. Abbreviations: CTNs, Cortical Thickness-Based Networks; FDNs, Fractal Dimension-Based Networks; FES, First-Episode Schizophrenia; GINs, Gyrification Index-Based Networks; HCs, Healthy Controls; SDNs, Sulcal Depth-Based Networks

Figure 3.

Network Distribution of Altered Interregional Morphological Similarities in FES. The Decreased Morphological Similarities in FES Were Consistently Found to Be Involved in the Cingulo-opercular and Default Mode Networks Regardless of the Type of Single-Subject Morphological Brain Networks. For the Increased Morphological Similarities in FES, They Were Mainly Involved in the Default Mode, Frontoparietal, and Somatomotor Networks for the FDNs and the Cingulo-opercular and Default Mode Networks for the SDNs. Abbreviations: CTNs, Cortical Thickness-Based Networks; FDNs, Fractal Dimension-Based Networks; FES, First-Episode Schizophrenia; GINs, Gyrification Index-Based Networks; HCs, Healthy Controls; SDNs, Sulcal Depth-Based Networks

Disruption of Inter-subnetwork Morphological Similarities in FES

Compared with the HCs, the patients with FES showed significantly increased morphological similarity within the posterior multimodal module and between the posterior multimodal module and the secondary visual, cingulo-opercular, and ventral multimodal modules (P < .05, FDR corrected; Figure 4).

Figure 4.

Increased Intra- and Inter-module Morphological Similarities in FES. (A) Regions in the HCP-MMP Atlas Can Be Categorized Into 12 Modules According to Their Resting-State Functional Connectivity Patterns. (B) Between-Group Differences in the Morphological Similarities Within Each and Between Each Pair of Modules. Significantly Increased Morphological Similarities Were Observed in the FES Patients Within the Posterior Multimodal Module, Between the Posterior Multimodal Module and Secondary Visual Module, Between the Posterior Multimodal Module and Cingulo-opercular Module, and Between the Posterior Multimodal Module and Ventral Multimodal Module. (C) Violin Plots Showing the Significantly Increased Morphological Similarities in the FES Patients. Abbreviations: FES, First-Episode Schizophrenia; HCs, Healthy Controls; JSDs, Jensen–Shannon Divergence-Based Similarity

Alterations of Topological Properties in FES

No significant between-group differences were found in any small-world parameters for any type of single-subject morphological brain networks. For local nodal centrality measures, significant alterations were observed in the FES patients, characterized by lower nodal degree and efficiency in the left Area PGs Area V6A in inferior parietal cortex and higher nodal efficiency in left area superior temporal gyrus (STGa) for the CTNs, lower nodal degree and efficiency in the left Posterior Insular Area 2 for the FDNs, and lower nodal betweenness in the left Area 52 for the SDNs (Figure 5). No significant alterations were observed in any regions for the GINs regardless of the nodal centrality metrics.

Figure 5.

Altered Nodal Centralities in FES. Compared With the HCs, the FES Patients Showed Significantly Decreased Nodal Degree and Efficiency in the Left Area PGs Area V6A and Increased Nodal Efficiency in the Left Area Superior Temporal Gyrus for the CTNs, Decreased Nodal Degree and Efficiency in the Left Posterior Insular Area 2 for the FDNs, and Decreased Nodal Betweenness in the left Area 52 for the SDNs. Abbreviations: CTNs, Cortical Thickness-Based Networks; FDNs, Fractal Dimension-Based Networks; FES, First-Episode Schizophrenia; HCs, Healthy Controls; SDNs, Sulcal Depth-Based Networks; STGa, Area Superior Temporal Gyrus

Brain-Clinical/Cognitive Relationships in the FES Patients

For the regions showing nodal centrality alterations and connections showing intra-/inter-subnetwork morphological similarity alterations in the patients with FES, no significant correlations were observed with the PANSS or MCCB scores in the patients with FES (P > .05). However, all NBS components showing FES-related morphological similarity alterations were significantly correlated with the PANSS and MCCB scores in the patients with FES (P < .05, FDR corrected; Table 2). Specifically, the connections showing decreased morphological similarities for the CTNs account for 25.2% interindividual variance in the PANSS scores (r = 0.502, P < .001) and 17.0% interindividual variance in the MCCB scores (r = 0.412, P < .001), the connections showing decreased morphological similarities for the FDNs account for 7.5% interindividual variance in the PANSS scores (r = 0.274, P < .001) and 2.9% interindividual variance in the MCCB scores (r = 0.171, P = .017), the connections showing increased morphological similarities for the FDNs account for 14.5% interindividual variance in the PANSS scores (r = 0.380, P < .001) and 6.1% interindividual variance in the MCCB scores (r = 0.246, P = .001), the connections showing decreased morphological similarities for the GINs account for 8.3% interindividual variance in the PANSS scores (r = 0.288, P < .001) and 5.6% interindividual variance in the MCCB scores (r = 0.236, P = .001), the connections showing decreased morphological similarities for the SDNs account for 8.2% interindividual variance in the PANSS scores (r = 0.286, P < .001) and 13.7% interindividual variance in the MCCB scores (r = 0.371, P < .001), and the connections showing increased morphological similarities for the SDNs account for 8.9% interindividual variance in the PANSS scores (r = 0.298, P < .001) and 6.0% interindividual variance in the MCCB scores (r = 0.244, P = .001).

Table 2.

The Ability of Altered Regional Morphology and Interregional Morphological Similarity in Accounting for Symptom Severity and Cognitive Dysfunction in the FES Patients

Feature	$R_{\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{S}\mathrm{S}}^2$	$R_{\mathrm{M}\mathrm{C}\mathrm{C}\mathrm{B}}^2$	Feature	$R_{\mathrm{P}\mathrm{A}\mathrm{N}\mathrm{S}\mathrm{S}}^2$	$R_{\mathrm{M}\mathrm{C}\mathrm{C}\mathrm{B}}^2$
↑ CT	0.038*	0.023	↓ morphological similarity in the CTNs	0.252*	0.170*
↓ FD	0.019	0.021	↓ morphological similarity in the FDNs	0.075*	0.029*
↑ FD	0.020	0.001	↑ morphological similarity in the FDNs	0.145*	0.061*
↓ GI	0.038*	0.033*	↓ morphological similarity in the GINs	0.083*	0.056*
↓ SD	0.031*	0.037*	↓ morphological similarity in the SDNs	0.082*	0.137*
↑ SD	0.009	0.001	↑ morphological similarity in the SDNs	0.089*	0.060*

The numbers represent R²-values derived from partial least-squares regression analyses. Abbreviations: ↑, significant increases in the patients with FES; ↓, significant decreases in the patients with FES; CT, cortical thickness; CTNs, cortical thickness-based networks; FD, fractal dimension; FDNs, fractal dimension-based networks; FES, first-episode schizophrenia; GI, gyrification index; GINs, gyrification index-based networks; MCCB, MATRICS Consensus Cognitive Battery; PANSS, Positive and Negative Syndrome Scale; SD, sulcal depth; SDNs, sulcal depth-based networks; *, P < 0.05, FDR corrected.

Sensitivity and Specificity of Morphological Metrics in Classification of FES

Poor discriminant performance was observed for the regions showing nodal centrality alterations and connections showing intra-/inter-subnetwork morphological similarity alterations in the patients with FES (AUC <0.7). In contrast, the connections showing altered morphological similarities in the patients exhibited better discriminant ability (AUC: 0.713-0.874; Table 3). In particular, the mean strength of connections showing decreased morphological similarities in the CTNs exhibited the best performance for distinguishing the FES patients from HCs (AUC = 0.874). Specifically, using the cutoff value (0.754) that maximized cost-efficiency, 164 out of the 203 patients with FES and 105 out of the 131 HCs were classified correctly, resulting in a sensitivity of 80.8% and specificity of 80.2% (Figure 6).

Table 3.

Discriminant Performance of Altered Regional Morphology and Interregional Morphological Similarity in Distinguishing the FES Patients From Controls

Feature	AUC	Feature	AUC
↑ CT	0.742	↓ morphological similarity in the CTNs	0.874
↓ FD	0.681	↓ morphological similarity in the FDNs	0.713
↑ FD	0.645	↑ morphological similarity in the FDNs	0.784
↓ GI	0.696	↓ morphological similarity in the GINs	0.787
↓ SD	0.648	↓ morphological similarity in the SDNs	0.801
↑ SD	0.604	↑ morphological similarity in the SDNs	0.804

The numbers represent values of area under the curve derived from the receiver operating characteristic curve analyses. Abbreviations: ↑, significant increases in the patients with FES; ↓, significant decreases in the patients with FES; AUC, area under curve; CT, cortical thickness; CTNs, cortical thickness-based networks; FD, fractal dimension; FES, first-episode schizophrenia; FDNs, fractal dimension-based networks; GI, gyrification index; GINs, gyrification index-based networks; ROC, receiver operating characteristic; SD, sulcal depth; SDNs, sulcal depth-based networks.

Figure 6.

FES-Control Classification. The Mean Strength of Decreased Morphological Similarities in the CTNs Distinguished the FES Patients From HCs With Good Performance. Abbreviations: AUC, Area Under Curve; CTNs, Cortical Thickness-Based Networks; FES, First-Episode Schizophrenia; ROC, Receiver Operating Characteristic

Alterations of Regional Morphology in FES

In addition to the FES-related alterations in single-subject morphological brain networks, we examined between-group differences in regional morphology (permutation test; age, sex, and education as covariates). Significant alterations were observed in regional CT, FD, GI, and SD in the patients with FES in comparison with the HCs (P < .05, corrected; Supplementary Figure S2). Only some of the alterations significantly explained interindividual variance in the PANSS or MCCB scores among the patients (PLS regression; P < .05, FDR corrected; Table 2). All the alterations exhibited poor performance in distinguishing the patients from controls (AUC: 0.604-0.742; Table 3). We further found that: (1) out of the 125 edges showing decreased morphological similarities in the CTNs, 4 (3.2%) were linked to the regions showing altered CT; (2) out of the 450 edges showing altered morphological similarities in the FDNs, 13 (2.9%) were linked to the regions showing altered FD; (3) out of the 114 edges showing decreased morphological similarities in the GINs, 61 (53.5%) were linked to the regions showing altered GI; and (4) out of the 348 edges showing altered morphological similarities in the SDNs, 41 (11.8%) were linked to the regions showing altered SD.

Discussion

In the current study, we employed 4 surface mesh-based indices to separately construct single-subject morphological brain networks and investigate morphological alterations in FES. We found altered numerous alterations in interregional morphological similarity in the FES patients in all types of single-subject morphological brain networks. At the subnetwork level, the FES patients showed increased morphological similarities within the posterior multimodal module and between the posterior multimodal module and secondary visual, cingulo-opercular, and ventral multimodal modules. Finally, we showed that global small-world organization was intact while the nodal centrality of several regions was altered in the FES patient. Interestingly, we found that altered interregional morphological similarities accounted for clinical symptoms and cognitive dysfunction in the FES patients and distinguished the patients from controls with better performance than altered local morphology. These findings provide new insights into the neurobiological mechanisms of schizophrenia.

We identified 6 connected components that exhibited altered morphological similarities in the FES patients. Of the 6 components, 4 exhibited decreased morphological similarities in CTNs, FDNs, GINs, and SDNs, and 2 exhibited increased morphological similarities in the FDNs and SDNs in the FES patients. Presumably, the decreases may reflect pathological alterations in the brain induced by FES while the increases may reflect a compensatory mechanism to balance metabolic cost and communication efficiency of the brain networks.⁵⁷ Notably, the decreased morphological similarities were mainly involved in the cingulo-opercular and default mode networks regardless of the type of single-subject morphological brain networks, suggesting that the coordination of local morphological features was heavily disrupted for regions in these 2 networks in FES. For regions in the cingulo-opercular and default mode networks, there are numerous studies showing structural and functional impairments in FES.^58–64 Moreover, the impairments are found to correlate with positive and/or negative symptoms in patients with FES.^65,66 Thus, our findings are consistent with previous studies and provide new evidence for the involvement of the cingulo-opercular and default mode networks in FES. Interestingly, the cingulo-opercular and default mode networks were also observed to show increased morphological similarities in the FDNs and SDNs in the FES patients. This implies a fractionation of these 2 networks in revealing FES-related morphological similarity alterations. Previous studies have shown that both the cingulo-opercular and default mode networks are composed of distinct subsystems.^67,68 We thus speculate that the fractionation may imply differential roles of the subsystems in these 2 networks in FES, which should be examined in future studies.

Furthermore, we found that the PLS1 scores derived from these morphological similarity alterations were significantly correlated with the PANSS and MCCB scores among the FES patients. These findings indicate that the altered morphological similarities derived from group-level between-group comparisons account for, to some extent, interindividual variance in clinical symptoms and cognitive dysfunction of the FES patients. It should be noted that since the PLS1 scores were the linear combinations of the edges showing altered morphological similarities in the FES patients, the directions of the correlations did not make sense here. Interestingly, we found that the altered morphological similarities exhibited better performance than altered local morphology not only in explaining interindividual variance in clinical symptoms and cognitive dysfunction in the patients but also in distinguishing the patients from controls. These findings underscore the significant clinical value of single-subject morphological brain networks in FES.

At the subnetwork level, we found increased morphological similarities within the posterior multimodal module and between the posterior multimodal module and secondary visual, cingulo-opercular, and ventral multimodal modules in patients with FES. All the increased morphological similarities were related to the posterior multimodal module, indicating the importance of this module in FES. Thus, we focused on discussing the posterior multimodal module. The posterior multimodal module consists of bilateral dorsomedial parietal lobe, bilateral temporal-parietal-occipital junction, and right dorsocaudal temporal lobe. Functionally, these regions are related to spatial navigation.⁶⁹ A previous study found that individuals with FES showed impaired spatial navigation during a virtual reality navigation task.⁷⁰ We thus speculate that the increased morphological similarities of the posterior multimodal module observed in this study might underlie the spatial navigation deficits in FES. This speculation can be examined in future studies by exploring the relationship between morphological similarity of the posterior multimodal module and the performance of FES patients in spatial navigation tasks in future studies.

Using the graph theory-based approaches, we found no significant between-group differences in the small-world parameters. The small-world organization reflects an optimal balance between local specialization and global integration. Thus, our findings indicate that the optimal balance is preserved in the FES patients. This is in contrast to previous studies of structural⁷¹ and functional^72–74 brain network studies in FES, which reported disrupted small-world organization in FES patients. This discrepancy may be due to differences in demographic and cultural factors and/or network analytical methods between studies. Insights into this speculation could benefit from future studies by collecting and analyzing multimodal neuroimaging data from the same cohort of FES patients.

For between-group comparisons of nodal centrality measures, the FES patients showed lower nodal centralities in the left PGs area V6A (located in the inferior parietal cortex), Posterior Insular Area 2, and Area 52. The first 2 regions are frequently reported to show structural and functional impairments in FES, such as reduced gray matter volume and CT,^75,76 lower activation during neutral action of emotion tasks,⁷⁷ disrupted resting-state functional connectivity,⁷⁸ and lower levels of glutamate.⁷⁹ As for Area 52, it is identified based on its similarity in location to Brodmann’s Area 52 and is considered as a transitional auditory area.²² Thus, the decreased nodal centrality in this region may be related to auditory hallucination, a major positive symptom in FES. In addition to decreased centralities, we found higher nodal efficiency in the left STGa in the FES patients. Given that this region is previously found to show reduced gray matter volume and CT in FES,^76,80,81 the increased nodal centrality may reflect the reorganization of this region’s connectivity profile to compensate for its morphological deterioration. For a better understanding of the altered nodal centralities, future studies are warranted to examine their relationships with cognitive dysfunctions in patients with FES.

The current study has several limitations to be considered and addressed in future work. First, the patients and HCs were not matched in age and education. Although these 2 variables were treated as covariates in our statistical analyses, our results may still be confounded to some extent. Second, this study focused exclusively on patients with FES to avoid the potential confounding effects of long-term medication. Nevertheless, the patients included in this study underwent short-term antipsychotic medication. We thus examined the relationship between the medication dosages (chlorpromazine equivalents) and those imaging measures that exhibited significant alterations in the FES patients. No significant correlations were observed, indicating limited effects of short-term medication on our results. In addition, it is interesting to explore how the observed alterations in single-subject morphological brain networks change with the disease progresses or differ between different stages of the disease. Third, there are currently numerous methods to construct single-subject morphological brain networks.¹⁴ Thus, future studies are warranted to examine whether our findings are reproducible when other methods are used. Finally, future studies are required to examine the similarities and differences in schizophrenia-related alterations between brain networks constructed with different imaging modalities.

Conclusion

The current study demonstrated characteristic disruptions in morphological similarity and topological organization of single-subject morphological brain networks in FES. Specifically, decreased morphological similarities were found in all types of morphological brain networks, while increased morphological similarities were observed only in the FDNs and SDNs in the FES patients. Of the alterations, reduced morphological similarities in the CTNs exhibited the best performance in accounting for the clinical symptoms of the patients and in distinguishing the patients from controls. These findings provide novel insights into neurobiological mechanisms underlying schizophrenia and offer potential new biomarkers for the early diagnosis of the disease.

Author Contributions

Conceived and designed the study: J.W. and Y.T.; collected the data: Y.T., S.T., Z.W., F.Y., H.F., H.X., H.G., and X.H.; analyzed the data: S.J., F.F., Y.L., Z.L., J.R., and Q.Z.; wrote the manuscript: F.F. and S.J.; revised the manuscript: Y.L., J.W., and Y.T.

Funding

This work was supported by the National Natural Science Foundation of China (Nos. 82472092 and 82171507), a grant from the Research Center for Brain Cognition and Human Development, Guangdong, China (No. 2024B0303390003), STI 2030—Major Projects (No. 2021ZD0200500), and Striving for the First-Class, Improving Weak Links and Highlighting Features (SIH) Key Discipline for Psychology in South China Normal University, and Capital’s Funds for Health Improvement and Research (No. 2022-1-2131), and grant from Beijing High-Level Public Health Technical Talent Development Project (Leading talents-03-04, Academic Backbones-03-39).

Conflicts of Interest

All authors have declared no conflicting interests.

Data Availability

Fengmei Fan has access to all of the data collected in the study and takes responsibility for the integrity of the data and the accuracy of data analysis. In this study, the authors used structural MRI data of patients with first-episode schizophrenia and healthy controls.