Background
Schizophrenia (SCZ) is a severely progressive psychiatric disease, characterized by both positive and negative symptoms. It is known to affect approximately 1% to 3% of the population worldwide [
1]. The diagnosis of SCZ is heterogeneous and therefore it is highly subjective and inaccurate to classify its diversity based only on clinical symptoms [
2]. The objective definition of biological subtypes using neuroanatomical data is crucial for further progress on the disease. Of note, early-onset SCZ (EOS), a rare form of SCZ with the onset of the disease before the age of 18 years, exhibits a poorer prognosis [
3]. To date, the specific architectural brain changes and the biological pathogenesis of EOS have remained unclear [
4].
To date, numerous studies have focused on the application of magnetic resonance imaging (MRI) for detecting specific architectural changes in SCZ patients. Multiple brain abnormalities, including structural abnormalities in the cortical areas, slow-growing white matter, and disrupted functional connections [
5,
6], have been linked to altered social cognitions and executive functions, as well as sensorimotor processing in SCZ [
7,
8]. MRI studies of EOS revealed that EOS patients exhibited abnormal structural and functional integration [
9] and abnormal development of the ventral occipitotemporal gray matter network associated with the social perception system [
10]. This finding supports the hypotheses of abnormal neural development and dysconnectivity in EOS, but the specific abnormal pattern of brain connectivity and corresponding biological interpretation in EOS are absent.
Data-driven modeling of the objective neuroanatomical data showed two distinct but stable subtypes of brain atrophy in adult-SCZ that exhibited different clinical characteristics and therapeutic effects [
11]. Similarly, another study identified two distinct neuroanatomical subtypes of adult-SCZ based on volume measurements of gray matter, white matter, and cerebrospinal fluid, characterized by a broad volume reduction and larger basal ganglia and internal capsule [
12]. However, the majority of the studies involved only single MRI morphometric and anatomical features, such as cortical thickness, volume, and curvature. Multiple MRI parameters could be utilized to obtain a better classification. Morphometric similarity mapping (MSN), a classical analytic procedure, quantifies the similarity across multiple cortical areas by combining multiple MRI parameters to construct whole-brain morphometric networks for each subject [
13,
14]. MSN has been widely used to detect macroscale cortical structural abnormalities in several mental disorders. Additionally, heterogeneity through discriminant analysis (HYDRA), a novel machine learning approach, is used to cluster disease effects by modeling differences with healthy controls (HC) instead of directly clustering the patients [
15]. A combination of MSN with HYDRA is a promising technique and explores the neuroanatomical subtypes of EOS to deepen our understanding of the early pathophysiology of SCZ.
The neurodevelopmental theory of SCZ posited that environmental and genetic risk factors, ranging from prenatal periods to adolescent stages, remarkably influenced the typical developmental trajectory of brain tissues, leading to psychotic symptoms during adolescence or early adulthood [
16]. Furthermore, genetic factors are known to impact mental diseases by shaping brain connectivity through connectomes [
17], and corresponding brain-wide transcriptional expression atlases can serve as the bridge connecting the brain connectomes with biological functions [
18]. Furthermore, studies combining the MSN analysis and gene transcripts have revealed the potential relationships between macroscale architectural abnormalities and specific transcriptional expression patterns in different mental disorders, such as major depressive disorder (MDD) [
19], adult-SCZ [
20], and autism [
21]. However, further studies are warranted to understand the combined evaluation of MSN differences and regional gene expression patterns for EOS, explore the potential pathogenesis, and develop novel therapeutic targets for individuals with EOS.
We studied the connection between molecular mechanisms and structural alterations in different EOS subtypes by associating EOS-related MSN abnormalities with transcriptional data. First, the HYDRA method was used to identify two distinct EOS subtypes based on MSN characteristics. Different abnormal MSN patterns between EOS subtypes and HC were revealed. For each subtype, we investigated the relationship between EOS-related regional changes in the MSN and corresponding gene expression patterns using the Allen Human Brain Atlas (AHBA) and in situ hybridization (ISH) data to identify EOS-related genes. Subsequently, we performed Spearman’s correlation analysis between published differential gene expression (DGEs) patterns of other mental disorders and EOS-related MSN differences. Finally, functional enrichment analysis was applied to decipher the ontological pathways of EOS-related genes. Specific cell types were mapped to estimate their contribution to the transcriptomic relationship with EOS-related MSN changes in different EOS subtypes. In conclusion, our findings generated an in-depth understanding of EOS subtypes, revealing a complex connection between MSN macrostructural changes and specific transcriptional expression patterns.
Discussion
This study is the first to conduct an integrated investigation of neuroanatomical subtypes of EOS patients by exploring the relationships between structural alterations and molecular mechanisms using a combined analysis of MSN macrostructural changes and specific transcriptional expression patterns. The study included 100 initially diagnosed and untreated EOS patients, along with 106 age- and gender-matched HC cohorts. We employed the novel machine learning approach (HYDRA) to identify two distinct MSN-based EOS subtypes, each exhibiting different abnormal MSN patterns compared with HC individuals. A combination of the AHBA and ISH datasets was used to identify 69 SCZ-related genes and further investigate the potential relationships between EOS-related regional changes in the MSN strength and corresponding transcriptomic patterns. Moreover, significant Spearman’s correlations were noted between PLS1 weighted genes of EOS-related MSN strength differences and DGEs from ASD, adult-SCZ, and IBD, and distinct functional enrichment of ontological terms and pathways in EOS subtypes was studied. Furthermore, the expression of PLS weighted genes was mapped to the distribution of astrocytes and neuronal cells, and BPs associated with these cells were identified using the Metascape software. These findings demonstrate the changes in the MSN strength across different EOS subtypes, bridging the gap between neuroimaging and transcriptional patterns and promoting an integrative understanding of EOS.
EOS1 displayed lower VIQ and FSIQ scores compared to EOS2, suggesting that cognitive characteristics distinguished the two specific EOS subtypes. In addition, similar subtypes were successfully identified in external adult-SCZ datasets with spatial positive correlation of MSN strength and shared abnormal regions, indicating the existence of disease subtypes. Moreover, the
t-value distributions of the differences in the MSN strength between the two EOS subtypes and HCs were opposite. EOS1 displayed similarities to MDD [
19] and adult-SCZ [
20], with a typical distribution of differences in
t-values for MSN strength. Unlike adult-SCZ patients, EOS1 exhibited more abnormally increased MSN strength in the frontal and parietal regions, with involvement of the occipital lobe and superior temporal gyrus. Furthermore, EOS1 exhibited more abnormally decreased MSN strength in the superior frontal and temporal regions, along with the involvement of the insula and precentral gyrus. However, EOS2 displayed the opposite atypical difference in the
t-values of morphological similarity, with abnormally increased MSN strength only in the right paracentral gyrus and abnormally decreased MSN strength in the right superior frontal gyrus. EOS1 exhibited differences in MSN strength in multiple functional and von Economo atlas, whereas EOS2 displayed abnormalities only in MSN strength within motor networks. Previous studies have demonstrated that high/low MSN strength implies that the cellular structures of similar/differentiated networks could be more/less likely to form axonal connections to each other [
41]. Changes in the brain develop rapidly from childhood to adolescence as myelination and synapse pruning [
42,
43]. Consistent with these studies, our developmental trajectories analysis constructed a full landscape of MSN strength along with age development in different EOS subtypes. Abnormal morphological similarities in EOS1 suggested that EOS1 could be a typical subtype of EOS, whereas brain regions showed abnormal axonal connectivity before adulthood due to myelin and synaptic hypoplasia, with more severe scope and degree of abnormalities. Abnormal morphological similarities in EOS2 suggested EOS2 as an atypical subtype of EOS, demonstrating only a single abnormality of axonal connectivity. The abnormal MSN pattern of the two EOS subtypes indicated that, unlike the classification based on subjective clinical symptoms, the MSN-based classification of pure and drug-naïve EOS could enhance our understanding of the early neuropathological mechanism of SCZ.
The analysis of SCZ-related genes revealed significantly different transcriptional patterns in distinct EOS subtypes. In EOS1 patients, a substantial portion (46/69) of EOS-related genes were associated with regional changes in the MSN, whereas the EOS2 subtype exhibited only a few EOS-related genes (3/69), suggesting stronger connectivity between gene expression and structural abnormities in EOS1 than those in EOS2 patients. The discovered
CIT gene encodes a citron rho-interacting kinase, which is known to regulate cytokinesis and central nervous system development [
44]. A genetic mutation in the
CIT (rs10744743) gene has been reported to be linked to risks for SCZ by interacting with the
DISC1 gene [
45]. Similarly, the tachykinin precursor 1 (
TAC1) gene has been implicated in the process of neuron excitation and behavioral responses by encoding four primary tachykinins, including substance-P, neurokinin-A, neuropeptide-K, and neuropeptide-γ [
46,
47]. Interestingly, as shown in Fig.
3e, the
CIT and
TAC1 genes exhibited contrary associations with corresponding changes in the MSN strength among different EOS subtypes, implying that expressional dysregulation of similar SCZ-related pathogenic genes were consistent with macrostructural abnormalities in brain regions with different EOS subtypes. In addition, the genetic polymorphism of the
CINR1 gene, widely expressed in the central nervous system, correlated with the pathogenesis of SCZ and antipsychotic response [
48]. Furthermore, genome-wide association studies (GWAS) have identified variants of the
GRM7 gene as risk factors and responses to antipsychotic therapy in SCZ patients [
49]. We demonstrated a significant spatial correlation between the expression of these genes and structural changes in the MSN strength. The positive/negative correlations indicated that PLS1 positively/negatively weighted genes were overexpressed in the macrostructures where the MSN strength was increased/decreased in EOS patients. These findings support a potential explanation that changes in transcriptional patterns may precede regional macrostructural abnormalities, and clinical disorder may be the latest during the development of SCZ.
Previous GWAS studies have reported shared genetic commonalities among major psychiatric disorders, particularly SCZ, BD, ASD, MDD, and obsessive–compulsive disorder [
50,
51]. Moreover, a large-scale longitudinal study identified SCZ patients as high-risk groups for subsequent IBD development [
52]. Similarly, Florian et al. demonstrated the presence of shared susceptibility genes (
NR5A2, SATB2, and
PPP3CA) between SCZ and IBD [
53]. These studies suggest that common neural mechanisms or the brain-gut axis, facilitated by shared genetic commonalities could be linking multiple mental disorders. Furthermore, our MAGMA analysis identified enrichments of similar BPs at genetic levels in these neuropsychiatric disorders, particularly in energy metabolism and neural interaction signaling. Consistent with the potential genetic commonalities, our results revealed significant positive correlations between PLS1- weights and ASD-related and adult-SCZ-related DGE values in the EOS1 subtype and with IBD-related DGE values in the EOS2 subtype. In addition, these findings indicated that the EOS1 subtype exhibited characteristics of “classical SCZ” with similar psychic abnormalities, whereas the EOS2 subtype displayed the “non-classical SCZ” type with higher links to the genetic background of IBD.
Based on PLS1 weighted genes, the enrichment of BP terms and KEGG pathways provided insightful interpretations of transcriptional signatures concerning the regional changes in the MSN strength for different EOS subtypes. The PLS1 + weighted genes in EOS1 were enriched in similar ontological terms such as PLS1- weighted genes in EOS2 patients, including “regulation of cellular metabolic process”, “regulation of RNA metabolic process” and herpes simplex virus 1 (HSV-1) infection pathway. Multiple cellular metabolic processes have been associated with the dysfunction of the dorsal prefrontal cortex in SCZ [
54], Similarly, HSV-1 infection is potentially associated with the pathogenesis of SCZ, as evident from infection and incubation in autonomic neurons and peripheral sensory neurons [
55]. These results suggested that PLS1- weighted genes are prominently implicated in abnormal functions observed in EOS2 subtypes, and virtual infection could potentially explain the “nonclassical SCZ” status at the functional level. In addition to virtual infection, PLS1 + genes of EOS1 were enriched in calcium signaling and MAPK signaling pathways, which are implicated in the onset of SCZ by regulating neuronal excitability and neural development [
56,
57]. However, PLS1- weighted genes in EOS1 were predominantly enriched in synaptic signaling-related functions, such as serotonergic synapse, phagosome, gap junction, neurotrophin signaling pathway. Synaptic signaling, responsible for synaptic stability and maturation, is intricately related to the mechanism of SCZ by regulating neuron connectivity, co-transmission, and activity [
58]. This classical regulatory mechanism of SCZ can reasonably explain the “classical SCZ” status observed in EOS1 patients.
Cellular abnormalities have been known to play a vital role in the development of psychiatric disorders, including ASD, SCZ, BD, and MDD [
59]. We identified inhibitory and excitatory neurons as the largest proportion among the seven cell types in PLS1 + weighted genes, whereas astrocytes constituted the majority among PLS1- weighted genes, concerning the changes in MSN strength in EOS1 patients. The distribution of the above cell types aligned with that of published studies using single-cell RNA sequencing (scRNA-seq) in SCZ [
60], confirming transcriptional abnormities of these cells in the pathogenesis of SCZ. The imbalance between excitatory and inhibitory neurons is crucial for the pathogenesis of SCZ, especially GABAergic deficits, and excitation in glutamatergic systems [
61]. Moreover, astrocytes are known to critically influence pivotal processes of neurodevelopment and homeostasis, leading to SCZ pathogenesis, induced by synaptogenesis, glutamatergic signaling, and myelination [
62]. The functional enrichment analysis from previous studies has demonstrated that neurons were enriched in biological processes associated with signaling transport, as well as pathways of “calcium signaling”, “neuronal System” and “tyrosine kinases receptor for neuronal cells”. In contrast, astrocytes were enriched in the “regulation of epithelial cell proliferation”, “cell–cell adhesion” and “Transport of small molecules”, consistent with cellular functional annotation for SCZ. These findings provide us with a reliable pattern to investigate cellular transcriptional signatures in EOS patients.
Our study had several limitations. First, although machine learning methods were used to identify novel disease subtypes with optimal category numbers, the sample size of patients was relatively limited. Therefore, MSN macrostructural changes and specific transcriptional patterns in different EOS subtypes should be further estimated and validated through other congeneric studies. Second, we included a limited set of clinical parameters related to EOS symptoms. More comprehensive clinical signatures should be collected for correlation analyses with regional changes in the MSN strength from EOS subtypes, such as body mass index, hyperlipemia, and hyperglycemia. Finally, the transcriptional datasets, obtained from the AHBA database, only covered gene expression from two right hemisphere tissues, allowing us to include the datasets of only the left hemisphere in our analysis. Hence, there exists a lack of a corresponding connection between gene signatures and MSN macrostructural changes in the right hemisphere.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.