Brain structural damage networks at different stages of schizophrenia

Abstract

Background

Neuroimaging studies have documented brain structural changes in schizophrenia at different stages of the illness, including clinical high-risk (cHR), genetic high-risk (gHR), first-episode schizophrenia (FES), and chronic schizophrenia (ChS). There is growing awareness that neuropathological processes associated with a disease fail to map to a specific brain region but do map to a specific brain network. We sought to investigate brain structural damage networks across different stages of schizophrenia.

Methods

We initially identified gray matter alterations in 523 cHR, 855 gHR, 2162 FES, and 2640 ChS individuals relative to 6963 healthy controls. By applying novel functional connectivity network mapping to large-scale discovery and validation resting-state functional magnetic resonance imaging datasets, we mapped these affected brain locations to four specific networks.

Results

Brain structural damage networks of cHR and gHR had limited and non-overlapping spatial distributions, with the former mainly involving the frontoparietal network and the latter principally implicating the subcortical network, indicative of distinct neuropathological mechanisms underlying cHR and gHR. By contrast, brain structural damage networks of FES and ChS manifested as similar patterns of widespread brain areas predominantly involving the somatomotor, ventral attention, and subcortical networks, suggesting an emergence of more prominent brain structural abnormalities with illness onset that have trait-like stability over time.

Conclusions

Our findings may not only provide a refined picture of schizophrenia neuropathology from a network perspective, but also potentially contribute to more targeted and effective intervention strategies for individuals at different schizophrenia stages.

Introduction

Schizophrenia is a severe psychiatric disorder with a typical onset in late adolescence or early adulthood, progressive deterioration, and a poor long-term outcome (Kahn et al., 2015; Marder, Ropper, & Cannon, 2019). Schizophrenia could be staged from a high-risk phase including individuals with an at-risk mental state (clinical risk) and unaffected relatives of schizophrenia patients (genetic risk), to full-blown psychosis and eventually a chronic condition (Davis, Moylan, Harvey, Maes, & Berk, 2014; Mcgorry, Hickie, Yung, Pantelis, & Jackson, 2016; Ortiz et al., 2017; Wood, Yung, Mcgorry, & Pantelis, 2011). The clinical staging model is paralleled by prominent pathophysiological paradigms that have conceptualized schizophrenia as a neurodevelopmental disease with progressive brain abnormalities (Andreasen, 2010; Buoli, Serati, Caldiroli, Cremaschi, & Altamura, 2017; Velakoulis, Wood, Mcgorry, & Pantelis, 2000). From a clinical perspective, identification and characterization of the neurobiological markers across different stages of schizophrenia may allow for more targeted and effective intervention strategies.

As an emerging discipline at the intersection of psychiatry and radiology, psychoradiology applies neuroimaging technologies to psychiatry, which not only yields insight into brain abnormalities in patients with psychiatric disorders but also has potential clinical utility (Luo, You, Delbello, Gong, & Li, 2022). A number of neuroimaging studies have documented brain structural changes in schizophrenia at different stages of the illness, including clinical high-risk (cHR) (Collins et al., 2022; Jalbrzikowski et al., 2021; Luna et al., 2022; Niznikiewicz, 2019), genetic high-risk (gHR) (Cattarinussi, Kubera, Hirjak, Wolf, & Sambataro, 2022; De Zwarte et al., 2019, 2020; Luna et al., 2022; Zhang et al., 2020), first-episode schizophrenia (FES) (Brugger & Howes, 2017; Hoang et al., 2022; Moon et al., 2023; Remiszewski et al., 2022; Shah et al., 2016), and chronic schizophrenia (ChS) (Brandl et al., 2019; Cui et al., 2022; Kuo & Pogue-Geile, 2019; Picó-Pérez et al., 2022; Van Erp et al., 2018). Moreover, a selective review of longitudinal magnetic resonance imaging (MRI) studies suggests that schizophrenia is associated with progressive gray matter abnormalities from cHR to FES to ChS (Dietsche, Kircher, & Falkenberg, 2017). A recent neuroimaging meta-analysis of cortical thickness changes at different schizophrenia stages reveals no significant differences between cHR and FES as well as greater cortical thinning in ChS compared to FES, highlighting progressive brain structural alterations following illness onset (Zhao et al., 2022). A comprehensive meta-analysis of voxel-based morphometry (VBM) studies demonstrates a cortical–subcortical, left-to-right homotopic progression of gray matter loss from cHR to FES to ChS, but no significant gray matter variations in gHR relative to healthy controls (HC) (Liloia et al., 2021). These prior findings jointly work to support the neuroprogressive models of schizophrenia from the viewpoint of regional morphology.

A coordinate-based meta-analysis can summarize the extensive body of neuroimaging literature in schizophrenia and identify the convergent neuroanatomical alterations linked to this illness across studies (Brugger & Howes, 2017; Eickhoff et al., 2009). Nonetheless, there is growing awareness that neuropathological processes associated with a disease fail to map to a specific brain region but do map to a specific brain network (Fornito, Zalesky, & Breakspear, 2015; Fox, 2018; Taylor, Siddiqi, & Fox, 2021). Accordingly, localization of a disease has recently shifted from a conventional region-based method to an updated network-based approach. Functional connectivity network mapping (FCNM) is a novel and well-validated technique that can map a disease, symptom, or psychological process to a common brain network, which is achieved through a combined analysis of brain locations of interest (e.g. lesion, structural damage, functional abnormality, and neural activation) and large-scale functional connectome data (Darby, Joutsa, & Fox, 2019; Joutsa et al., 2022; Peng, Xu, Jiang, & Gong, 2022; Taylor et al., 2023). This connectome localization approach has been widely applied to neuropsychiatric disorders and has enjoyed considerable success in enhancing our understanding of disease neuropathology from a network perspective (Cheng et al., 2024; Cotovio et al., 2020; Cotovio, Faro Viana, Fox, & Oliveira-Maia, 2022; Darby et al., 2019; Jones et al., 2023; Mo et al., 2024; Padmanabhan et al., 2019; Taylor et al., 2021, 2023; Tetreault et al., 2020; Trapp et al., 2023; Zhang, Xu, Ma, Qian, & Zhu, 2024; Zhukovsky et al., 2021). Despite these promising findings, investigations of brain structural damage networks of different stages of schizophrenia are still lacking, leaving a research gap in this field.

The objective of this exploratory study was to examine brain structural damage networks at different stages of schizophrenia. To achieve this goal, we initially synthesized published literature to identify gray matter alterations in cHR, gHR, FES, and ChS individuals relative to HC. By combining these affected brain locations with large-scale discovery and validation resting-state functional MRI (fMRI) datasets, we then used the FCNM approach to construct four brain structural damage networks corresponding to different stages of schizophrenia. Schematic representation of the study design and analytical procedure is shown in Fig. 1. Additionally, we assessed the spatial similarity of network patterns between different schizophrenia stages to investigate differences and commonalities in neuropathology across disease progression.

Figure 1. Study design and analytical procedure. We initially synthesized the published literature to identify gray matter alterations in cHR, gHR, FES, and ChS individuals relative to HC. By combining these affected brain locations with large-scale discovery (AMUD) and validation (SALD) resting-state fMRI datasets, we then used the FCNM approach to construct four brain structural damage networks corresponding to different stages of schizophrenia in the following way. Specifically, spheres centered at each coordinate of a contrast were first created and merged together to generate a contrast-specific combined seed mask. Second, based on the resting-state fMRI data, we computed a contrast seed-to-whole brain rsFC map for each subject. Third, the subject-level rsFC maps were entered into a voxel-wise one-sample t test to identify brain regions functionally connected to each contrast seed. Fourth, the resulting group-level t maps were thresholded and binarized at p < 0.05 corrected for multiple testing using a voxel-level FDR method. Finally, the binarized maps were overlaid to produce four network probability maps, which were thresholded at 60% to yield brain structural damage networks of cHR, gHR, FES, and ChS respectively. AMUD, Anhui Medical University Dataset; cHR, clinical high-risk; ChS, chronic schizophrenia; FDR: false-discovery rate; FES, first-episode schizophrenia; fMRI, functional magnetic resonance imaging; gHR, genetic high-risk; HC, health controls; rsFC, resting-state functional connectivity; SALD, Southwest University Adult Lifespan Dataset.

Materials and methods

Study search and selection

We conducted a systematic literature search on the PubMed, Brainmap, and Web of Science databases to identify studies examining gray matter alterations in cHR, gHR, FES, or ChS individuals relative to HC, published before March 1, 2023. We used the following terms to search in title/abstracts: (‘voxel-based morphometry’ OR ‘VBM’ OR ‘voxel-wise’) AND (‘schizophrenia’ OR ‘chronic schizophrenia’ OR ‘SZ’ OR ‘first episode schizophrenia’ OR ‘first episode psychosis’ OR ‘high risk schizophrenia’ OR ‘siblings schizophrenia’ OR ‘first degree relatives’ OR ‘genetic risk schizophrenia’ OR ‘at risk of mental state’ OR ‘ARMS’ OR ‘ultra-high risk’). We also screened the reference lists of the identified studies, review articles, and meta-analyses to search for additional qualified studies. Note that cHR individuals met the criteria of either the Personal Assessment and Crisis Evaluation (Yung et al., 1998), the Comprehensive Assessment of At-Risk Mental States (Yung et al., 2005), or the Structured Interview for Prodromal Symptoms (Miller et al., 2003). The cHR factors mainly include early clinical symptoms (e.g. cognitive deficits) and detrimental environmental factors (e.g. adverse life experiences) (Catalan et al., 2021; Jauhar, Johnstone, & Mckenna, 2022; Radua et al., 2018). The gHR factors primarily include monozygotic twins, siblings, and first-/second-degree relatives of schizophrenia patients. FES referred to an illness duration of less than 2 years, and ChS referred to an illness duration of more than 2 years.

Since there are some cases where a single study contains multiple neuroimaging contrasts (i.e. cHR, gHR, FES, or ChS individuals v. HC), we focused our analysis on contrasts rather than studies. Inclusion criteria for neuroimaging contrasts were: (a) to be included in a research article published in a peer-review journal; (b) to use a specified whole-brain VBM analysis; (c) to report gray matter alterations in cHR, gHR, FES, or ChS individuals by means of a between-group comparison with HC; (d) to report results in the form of stereotactic space (i.e. x, y, z coordinates in Talairach or Montreal Neurological Institute [MNI] space); (e) experimental groups without other medical comorbidities; and (f) diagnosis of schizophrenia based on fulfilling International Classification of Diseases or Diagnostic and Statistical Manual of Mental Disorders criteria. Exclusion criteria were: (a) experimental groups with a sample size <10 subjects; (b) use of region of interest analysis or small volume correction; and (c) mixed experimental sample (e.g. FES and ChS). Study search and selection were independently performed by two investigators. A flow diagram of the study selection process is shown in Fig. S1 in the online Supplementary materials. This protocol was registered on PROSPERO (https://www.crd.york.ac.uk/PROSPERO/, registration number: CRD42023481658). Coordinates of peak voxels of significant clusters reported in each contrast were extracted, with coordinates in Talairach space converted to MNI space. It is notable that subsequent analysis was based on the peak coordinates extracted from the published studies rather than MRI images from our patient and control groups.

Discovery and validation datasets

Our study used Anhui Medical University Dataset (AMUD) as a discovery dataset and Southwest University Adult Lifespan Dataset (SALD) (Wei et al., 2018) as a cross-scanner validation dataset. AMUD included 656 healthy adults of Chinese Han and right handedness (396 female, mean 26.57 ± 8.57 years), who were enrolled from local universities and communities through poster advertisements. Participants with neuropsychiatric or severe somatic disorders, a history of head injury with consciousness loss, MRI contraindications, or a family history of psychiatric diseases among first-degree relatives were excluded. SALD included 329 healthy adults (207 female, mean 37.81 ± 13.79 years) and full details regarding the sample have been described in the data descriptor publication (Wei et al., 2018). For the SALD, the exclusion criteria included MRI contraindications, current psychiatric or neurological disorders, use of psychiatric drugs within 3 months, pregnancy, or a history of head trauma. It is noteworthy that all included participants were restricted to an age range of 18–60 years to exclude the potential influences of neurodevelopment and neurodegeneration. Demographic information of the discovery and validation datasets is provided in Table S1 in the online Supplementary materials.

The authors assert that all procedures contributing to this work comply with the ethical standards of the relevant national and institutional committees on human experimentation and with the Helsinki Declaration of 1975, as revised in 2008. All procedures involving human subjects in the AMUD were conducted following approval by the ethics committee of The First Affiliated Hospital of Anhui Medical University (approval number 20200094), and all participants provided written informed consent after being given a complete description of the study. As for the SALD, it is a publicly available resource and detailed ethical information can be found at http://fcon_1000.projects.nitrc.org/indi/retro/sald.html.

fMRI data acquisition and preprocessing

Resting-state fMRI data of AMUD were collected on a 3.0-Tesla General Electric Discovery MR750w scanner, and those of SALD on a 3.0-Tesla Siemens Trio scanner. The fMRI parameters of the two datasets are provided in Table S2 in the online Supplementary materials. Participants with poor image quality (e.g. visible artifacts) and incomplete brain coverage were excluded.

Resting-state fMRI data were preprocessed using Statistical Parametric Mapping software (SPM12, http://www.fil.ion.ucl.ac.uk/spm) and Data Processing & Analysis for Brain Imaging (DPABI, http://rfmri.org/dpabi) (Yan, Wang, Zuo, & Zang, 2016). The first 10 volumes for each participant were discarded to allow the signal to reach equilibrium and the participants to adapt to the scanning noise. The remaining volumes were corrected for the acquisition time delay between slices. Then, realignment was performed to correct the motion between time points. Head motion parameters were computed by estimating the translation in each direction and the angular rotation on each axis for each volume. All participants' data were within the defined motion thresholds (i.e. maximal translational or rotational motion parameters less than 2 mm or 2°). We also calculated frame-wise displacement (FD), which indexes the volume-to-volume changes in head position. Several nuisance covariates (the linear drift, the estimated motion parameters based on the Friston-24 model, the spike volumes with FD > 0.5 mm, the global signal, the white matter signal, and the cerebrospinal fluid signal) were regressed out from the data. Since global signal regression can enhance the detection of system-specific correlations and improve the correspondence to anatomical connectivity, we included this step in the preprocessing of resting-state fMRI data (Murphy & Fox, 2017). Then, the datasets were band-pass filtered using a frequency range of 0.01–0.1 Hz. In the normalization step, individual structural images were first co-registered with the mean functional images; the transformed structural images were then segmented and normalized to MNI space using a high-level nonlinear warping algorithm, i.e. the diffeomorphic anatomical registration through exponentiated Lie algebra technique (Ashburner, 2007). Next, each filtered functional volume was spatially normalized to MNI space using the deformation parameters estimated during the above step and resampled into 3 mm isotropic voxel. Finally, all data were spatially smoothed with a Gaussian kernel of 6 × 6 × 6 mm³ full-width at half maximum.

Functional connectivity network mapping

We adopted the FCNM approach to construct brain structural damage networks of different schizophrenia stages based on the extracted coordinates of gray matter alterations in cHR, gHR, FES, and ChS individuals relative to HC (Fig. 1). First, 4 mm radius spheres centered at each coordinate of a contrast were created and merged together to generate a contrast-specific combined seed mask (henceforth referred to as the contrast seed). Second, based on the preprocessed resting-state fMRI data of AMUD, we computed a contrast seed-to-whole brain functional connectivity (FC) map for each subject, by calculating Pearson's correlation coefficients between time courses of the contrast seed and each voxel within the whole brain, followed by Fisher's Z transformation to improve normality. Third, the 656 subject-level FC maps were entered into a voxel-wise one-sample t test to identify brain regions functionally connected to each contrast seed. It is noteworthy that we only considered positive FC as the biological meaning of negative FC is still a matter of debate (Murphy & Fox, 2017; Murphy, Birn, Handwerker, Jones, & Bandettini, 2009). Fourth, the resulting group-level t maps were thresholded and binarized at p < 0.05 corrected for multiple testing using a voxel-level false-discovery rate method. Finally, the binarized maps were overlaid to produce four network probability maps, which were thresholded at 60% to yield brain structural damage networks of cHR, gHR, FES, and ChS, respectively.

Association with canonical brain networks

To ease interpretability, we tested the spatial associations between four schizophrenia brain structural damage networks and eight well-established canonical brain networks. The seven cortical networks were the visual, somatomotor, dorsal attention, ventral attention, limbic, frontoparietal, and default networks from the study by Yeo et al. (2011). The subcortical network was composed of the amygdala, hippocampus, basal ganglia, and thalamus from the Human Brainnetome Atlas (Fan et al., 2016). The proportion of overlapping voxels between each brain structural damage network and a canonical network to all voxels within the corresponding canonical network was calculated to estimate their spatial association.

Validation analyses

We carried out several validation analyses to examine the robustness of our results. First, we repeated our analyses by use of an independent validation SALD to assess the influence of dataset selection. Second, we repeated the FCNM procedure using 1 and 7 mm radius spheres to test the effect of seed size. Finally, we adopted 60% to threshold the network probability maps in our main analysis. To examine the impact of threshold selection, we repeated the FCNM procedure using 55% and 65% thresholds.

Results

Included studies

A total of 136 studies with 156 neuroimaging contrasts were included in our analyses. Specifically, 14 studies contained 14 contrasts from 523 cHR and 524 HC, 16 studies contained 17 contrasts from 855 gHR and 888 HC, 55 studies contained 58 contrasts from 2162 FES and 2488 HC, and 61 studies contained 67 contrasts from 2640 ChS and 3063 HC. Table S3 in the online Supplementary materials summarizes sample characteristics of the included studies.

Brain structural damage network of cHR

Brain structural damage network of cHR was primarily composed of the bilateral orbitofrontal cortex, lateral prefrontal cortex, posterior parietal cortex, and lateral temporal cortex (Fig. 2). With regard to canonical brain networks, brain structural damage network of cHR mainly involved the frontoparietal network (overlapping proportion: 24.5%) (Fig. 3a).

Figure 2. Brain structural damage networks of different schizophrenia stages. Brain structural damage networks of cHR, gHR, FES, and ChS are shown as network probability maps thresholded at 60%, showing brain regions functionally connected to more than 60% of the contrast seeds. cHR, clinical high-risk; ChS, chronic schizophrenia; FES, first-episode schizophrenia; gHR, genetic high-risk; L, left; R, right.

Figure 3. Associations of brain structural damage networks of cHR (a), gHR (b), FES (c), and ChS (d) with canonical brain networks. Polar plots display the proportion of overlapping voxels between each brain structural damage network and a canonical network to all voxels within the corresponding canonical network. cHR, clinical high-risk; ChS, chronic schizophrenia; FES, first-episode schizophrenia; gHR, genetic high-risk.

Brain structural damage network of gHR

Brain structural damage network of gHR chiefly consisted of the bilateral striatum, hippocampus, amygdala, thalamus, medial prefrontal cortex, posterior cingulate cortex/precuneus, and temporal pole (Fig. 2). As to canonical networks, brain structural damage network of gHR principally implicated the subcortical network (23.6%) (Fig. 3b).

Brain structural damage network of FES

Brain structural damage network of FES comprised of widespread brain areas mainly including the bilateral auditory cortex, middle cingulate cortex, ventrolateral prefrontal cortex, insula and operculum, temporal pole, medial prefrontal cortex, striatum, hippocampus, amygdala, and thalamus (Fig. 2). With respect to canonical networks, brain structural damage network of FES predominantly involved the somatomotor (42.1%), ventral attention (40.0%), and subcortical (39.5%) networks (Fig. 3c).

Brain structural damage network of ChS

Brain structural damage network of ChS was similar to but less extensive than structural damage network of FES, mainly comprising the bilateral auditory cortex, middle cingulate cortex, ventrolateral prefrontal cortex, insula and operculum, temporal pole, medial prefrontal cortex, striatum, amygdala, and thalamus (Fig. 2). Regarding canonical networks, brain structural damage network of ChS also primarily implicated the ventral attention (35.0%), subcortical (30.4%), and somatomotor (25.1%) networks (Fig. 3d).

Network similarity between schizophrenia stages

To quantify the similarity of network patterns between different schizophrenia stages, we estimated the spatial overlap between the networks of two schizophrenia stages by calculating a Dice coefficient, defined as 2 × (overlapping voxels)/(network #1 voxels) + (network #2 voxels). A higher Dice coefficient indicates more similar networks. As shown in Fig. 4, we observed a quite low spatial overlap between brain structural damage networks of cHR and gHR (Dice coefficient = 0.04), partial overlaps between cHR and FES (Dice coefficient = 0.19) and between gHR and FES (Dice coefficient = 0.31), and a high overlap between FES and ChS (Dice coefficient = 0.70).

Figure 4. Network similarity between schizophrenia stages. To quantify the similarity of network patterns between different schizophrenia stages, we estimated the spatial overlap between the networks of two schizophrenia stages by calculating a Dice coefficient, defined as 2 × (overlapping voxels)/(network #1 voxels) + (network #2 voxels). A higher Dice coefficient indicates more similar networks. cHR, clinical high-risk; ChS, chronic schizophrenia; FES, first-episode schizophrenia; gHR, genetic high-risk; L, left; R, right.

Validation analyses

First, brain structural damage networks of different schizophrenia stages derived from the validation SALD were similar to those from the discovery AMUD, with subtle differences largely attributed to variation in sample sizes (656 v. 329) (Fig. S2 in the online Supplementary materials). Second, when repeating the FCNM procedure using 1 and 7 mm radius spheres, we observed that the resulting brain structural damage networks of cHR, gHR, FES, and ChS were nearly identical to those using the 4 mm radius sphere (Figs S3 and S4 in the online Supplementary materials). Finally, using the 55% and 65% thresholds yielded brain structural damage networks that had larger and smaller spatial extents than those based on the 60% threshold (Figs S5 and S6 in the online Supplementary materials). However, the core areas of brain structural damage networks were still present across these thresholds. These findings verified the robustness of our results to distinct datasets and methodological differences.

Discussion

By a combination of the novel FCNM approach and large-scale brain functional connectome data, the present study interrogated brain structural damage networks of different schizophrenia stages based on gray matter alterations in cHR, gHR, FES, and ChS reported in previous literature. Results showed that brain structural damage networks of cHR and gHR had limited and non-overlapping spatial distributions, with the former mainly involving the frontoparietal network and the latter principally implicating the subcortical network, indicative of distinct neuropathological mechanisms underlying cHR and gHR. By contrast, brain structural damage networks of FES and ChS manifested as similar patterns of widespread brain areas predominantly involving the somatomotor, ventral attention, and subcortical networks, suggesting an emergence of more prominent brain structural abnormalities with illness onset that have trait-like stability over time. Our findings may not only provide a refined picture of schizophrenia neuropathology from a network perspective, but also potentially contribute to more targeted and effective intervention strategies for individuals at different schizophrenia stages.

Although neuroimaging has strengthened its position as the most widely used tool in the study of brain abnormalities in psychiatric illnesses, results are often irreproducible in this area (Elliott, Knodt, & Hariri, 2021; Etkin, 2019; Westlin et al., 2023) such that neuroimaging findings are considered of limited clinical and translational value (Poldrack et al., 2017). This lack of reproducibility may be due to a series of factors, such as limited sample sizes, clinical heterogeneity of patient populations, varying experimental designs, and different analytic techniques (Etkin, 2019; Poldrack et al., 2017). One potent approach to mitigating these concerns is to perform a coordinate-based meta-analysis, which is able to synthesize the multitude of results from different neuroimaging studies and identify more consistently affected anatomical regions in psychiatric disorders (Eickhoff et al., 2009; Wager, Lindquist, & Kaplan, 2007). Nonetheless, researchers recently caution against mapping diseases to single brain regions and instead emphasize network localization, as it is largely accepted that neuropathological processes by no means act in isolation, but rather are interconnected via distributed brain networks (Fornito et al., 2015; Van Den Heuvel & Sporns, 2019). In this context, localization of a disease has increasingly shifted from a traditional region-based method to an updated network-based approach. The conceptual and methodological advances have been facilitated by the development of the FCNM approach that can map a disease, symptom, or psychological process to a common network through a combined analysis of brain locations of interest (e.g. lesion, structural damage, functional abnormality, and neural activation) and large-scale brain functional connectome data (Darby et al., 2019; Joutsa et al., 2022; Peng et al., 2022; Taylor et al., 2023). By leveraging FCNM, investigators have mapped many neuropsychiatric diseases, some of which have eluded traditional region-based localization, to specific brain networks (Cheng et al., 2024; Cotovio et al., 2020, 2022; Darby et al., 2019; Jones et al., 2023; Mo et al., 2024; Padmanabhan et al., 2019; Taylor et al., 2021, 2023; Tetreault et al., 2020; Trapp et al., 2023; Zhang et al., 2024; Zhukovsky et al., 2021). With respect to brain structural alterations across different stages of schizophrenia, previous coordinate-based neuroimaging meta-analyses and reviews have demonstrated progressive brain abnormalities in schizophrenia from the viewpoint of regional morphology (Dietsche et al., 2017; Liloia et al., 2021; Zhao et al., 2022). Complementing and expanding these pilot efforts, we used the FCNM approach to characterize the neuropathological progression in schizophrenia from a network perspective. The present work, in concert with other FCNM investigations, suggests that the network-based framework holds substantial potential for refining our understanding of disease neuropathology.

We found that brain structural damage networks of cHR and gHR had distinct spatial patterns, possibly reflective of underlying neuropathological differences. The current observation of neuropathological differences between cHR and gHR is consistent with previous findings (Luna et al., 2022; Smieskova et al., 2013). A plausible explanation lies in the fact that the former may be a consequence of complex interactions between genetic architecture and environmental factors, whereas the latter may reflect a genetic effect. Brain structural damage network of cHR mainly involved the frontoparietal network, which is engaged in high-order cognitive processes (Lorenz et al., 2018; Uddin, 2021; Zuo, Yang, Liu, Li, & Jiang, 2018). Behavioral evidence has shown that cHR state for psychosis is associated with significant and widespread impairments in neurocognitive functioning and social cognition (Carrión et al., 2011; Catalan et al., 2021; Pedruzo et al., 2023), highlighting the importance of frontoparietal network dysfunction in the pathophysiology of prodromal psychosis. Brain structural damage network of gHR principally implicated the subcortical network, which was also found to be strongly affected in FES and ChS, suggesting that subcortical network abnormalities may be at the root of pathogenesis of schizophrenia. Indeed, gray matter alterations in subcortical structures including the striatum, hippocampus, amygdala, and thalamus have been frequently reported in gHR, FES, and ChS individuals (Brosch et al., 2022; Cattarinussi et al., 2022; Cooper, Barker, Radua, Fusar-Poli, & Lawrie, 2014; Cui et al., 2022; Ganzola, Maziade, & Duchesne, 2014; Nenadic et al., 2015; Van Erp et al., 2015). The dopamine hypothesis is the longest standing pathoetiologic theory of schizophrenia (Howes et al., 2012; Mccutcheon, Abi-Dargham, & Howes, 2019; Mccutcheon, Krystal, & Howes, 2020; Purves-Tyson et al., 2017). Altered striatal dopamine synthesis and release are present in schizophrenia, with the changes to a lesser degree in individuals at risk of schizophrenia and more prominent in patients with established schizophrenia (Howes, Mccutcheon, Owen, & Murray, 2017; Kesby, Eyles, Mcgrath, & Scott, 2018). Earlier work indicates that hippocampal pathology in schizophrenia may be due to genetic factors, aberrant neurodevelopment, and/or abnormal neural plasticity; hippocampal pathology is likely to be associated with the neuropsychological impairments of schizophrenia rather than with its psychotic symptoms (Harrison, 2004; Szeszko et al., 2002). Evidence from functional neuroimaging suggests that abnormal amygdala activation in response to emotional stimuli might contribute to emotional dysfunction, a critical clinical feature of schizophrenia (Anticevic et al., 2012; Guimond, Mothi, Makowski, Chakravarty, & Keshavan, 2022; Mukherjee et al., 2016). The thalamus is a key hub of cortical–subcortical circuitry and plays a vital role in the coordination of information within the brain; given that thalamic abnormalities have been observed early in the progression of schizophrenia, one may postulate that a disruption of the thalamocortical network might give rise to some cardinal symptoms of schizophrenia (Benoit, Canetta, & Kellendonk, 2022; Cronenwett & Csernansky, 2010; Giraldo-Chica & Woodward, 2017).

Brain structural damage networks of FES and ChS manifested as similar patterns of widespread brain areas, suggesting an emergence of more prominent brain structural abnormalities with illness onset that have trait-like stability over time. Strikingly, structural damage network of FES was slightly more extensive than that of ChS, which might seem counter-intuitive at first sight. Treatment-induced or experience-dependent neuroplasticity during the chronic condition may provide a potentially mechanistic account for this finding. An alternative methodological explanation is that variation in the number of the included studies on FES and ChS (55 v. 61) may lead to the subtle difference. Aside from the subcortical network, brain structural damage networks of FES and ChS prominently involved the somatomotor and ventral attention networks. The somatomotor network is implicated in auditory, visual, tactile, and direct pain perception (Keysers, Kaas, & Gazzola, 2010; Lee et al., 2021). Empirical findings have shown a strong link between robotically controlled alterations in somatomotor processing and auditory misattribution in psychosis, providing evidence for the role of somatomotor processes in altered self-monitoring in psychosis (Salomon et al., 2020). In addition, recent finding of abnormal cortical hierarchy organization in schizophrenia suggests that cascading impairments from the disruption of the somatosensory-motor system and inefficient integration of bottom-up sensory information with attentional demands and executive control processes partially account for high-level cognitive deficit characteristics of schizophrenia (Dong et al., 2021; Kaufmann et al., 2015). With respect to the ventral attention network, prior data have demonstrated its prominent engagement in reorienting attention to salient stimuli (Corbetta, Patel, & Shulman, 2008; Decety & Lamm, 2007). Intense research has reported structural, functional, and connectivity changes in the ventral attention network in schizophrenia (Dong, Wang, Chang, Luo, & Yao, 2018; Jimenez et al., 2016; Shafiei et al., 2020). This invites the speculation that faulty assignment of salience to internally generated mental events forms the basis of positive symptoms, such as delusions and hallucinations, in patients suffering from schizophrenia. Echoing our findings, there is also evidence from functional neuroimaging that patients with schizophrenia exhibit functional abnormalities in the frontoparietal, somatomotor, ventral attention, and subcortical networks (Li et al., 2023; Luo et al., 2023; You et al., 2022).

Several caveats should be considered when interpreting the current findings. First, it is important to point out that our analyses cannot determine causality of illness progression due to the lack of longitudinal design. However, robust cross-sectional analyses are also essential to clarify this issue and build a more integrative view. Second, we cannot rule out the influence of antipsychotic medications on our results, particularly those of ChS individuals who were treated using different types of antipsychotics. Third, we utilized resting-state fMRI data from healthy subjects to construct brain structural damage networks of schizophrenia across different stages of the illness. It appears more reasonable to utilize fMRI data from individuals at different stages of schizophrenia. However, there is evidence showing that sample selection has a negligible effect on network localization results (Boes et al., 2015; Fox et al., 2014; Horn et al., 2017). Fourth, the present study did not perform any correlation analysis due to the study design. Exploring the associations between brain MRI characteristics and clinical profiles has been a focus of psychiatric research (Li et al., 2024; Vieira et al., 2024). Investigation into the relationship between the identified brain structural damage networks and clinical characteristics in schizophrenia will be part of our future work. Likewise, subsequent research is needed to determine whether the brain structural damage networks found in clinical and gHR individuals could constitute effective biomarkers to predict the onset or the treatment outcome and prognosis in the long-term course of schizophrenia (Jauhar et al., 2022; Long et al., 2024). Fifth, we did not consider each study's sample size and effect size in our FCNM analysis, since there has been no consensus yet on how to account for these factors. Further investigation of their influences, in concert with analytical advances in the future, will help address this issue. Finally, we adopted an overlapping threshold of 60% to identify brain regions functionally connected to 60% of the contrast seeds. This less stringent threshold was selected due to the fact that many sources of variance may impede the identification of a common brain network at a more rigorous threshold.

In summary, this study combined the novel FCNM approach and large-scale brain functional connectome data to localize gray matter alterations in cHR, gHR, FES, and ChS to four brain structural damage networks corresponding to different stages of schizophrenia. Our findings may provide a refined picture of schizophrenia neuropathology from a network perspective. More broadly, the current identification of these brain structural damage networks may expose more effective prevention and treatment targets for individuals at different schizophrenia stages, laying the foundation for precision medicine in psychiatry.