INTRODUCTION
Schizophrenia is a mentally debilitating disorder with unknown etiology that primarily affects young adults between the ages of 16 and 30 and often persists throughout their lives [1]. Current research suggests that schizophrenia is associated with biological factors such as genetics, abnormal neural development, and neurotransmitter abnormalities, as well as psychosocial factors including personality, mental endurance, and stress [2-5]. Clinical observations indicate that patients with different subtypes of schizophrenia exhibit varying symptoms, which are complex and diverse. These symptoms often involve impairments in perception, thinking, emotion, volition, behavior, and cognitive functions. Furthermore, there are significant inter-individual differences in symptom presentation, and even within the same patient, different symptoms may manifest during different stages or phases of the illness [6-8]. Therefore, early detection and treatment are crucial in managing schizophrenia [9]. Currently, pharmacological and psychological therapies have shown some efficacy in improving the recovery rate of individuals with schizophrenia [10]. However, the underlying causes of schizophrenia remain incompletely understood [11]. Therefore, exploring various factors that may influence the occurrence and development of schizophrenia could contribute to the diagnosis and treatment of the disorder.
Mitochondria are the primary producers of cellular energy and play a crucial regulatory role in iron homeostasis, amino acid metabolism, and nucleotide synthesis [12-15]. In recent years, numerous studies have demonstrated the significant involvement of mitochondria in brain development and the pathogenesis of various mental disorders [16]. In the central nervous system, mitochondria generate membrane ATPases and provide abundant energy to support the influx and efflux of neurotransmitters. They also participate in synaptic transmission, neuronal growth, and sprouting [17]. Schizophrenia has been associated with impaired immune function, abnormal neuronal differentiation, and various neurotransmitter system abnormalities [18]. Therefore, investigating the relationship between mitochondria and schizophrenia is of great research significance. With the advancement of studies, researchers have increasingly focused on the role of mitochondria in the pathophysiology of schizophrenia [19,20]. Ni et al. [21] found that enhancing mitochondrial function can serve as a potential therapeutic target for psychiatric disorders. However, the role of mitochondrial metabolism-related genes (MRGs) in the diagnosis of schizophrenia remains unknown.
In this study, we downloaded three schizophrenia expression datasets from the Gene Expression Omnibus (GEO) database and obtained MRGs from relevant literature. Differential analysis and Venn analysis were performed to identify differentially expressed MRGs (DEMRGs) in schizophrenia. Subsequently, hub genes were selected from the DEMRGs, and a diagnostic model was constructed to identify key genes with diagnostic value. Additionally, candidate drugs potentially targeting hub genes were predicted, and potential miRNAs and transcription factors (TFs) were screened. Overall, this study not only predicted multiple diagnostic biomarkers for schizophrenia but also provided new insights for further research on miRNAs, TFs, and candidate drugs targeting MRGs.
METHODS
Data retrieval
Three expression datasets related to schizophrenia were downloaded from the GEO database: GSE21138 (normal: 29, disease: 30), GSE92538 (normal: 50, disease: 24), and GSE27383 (normal: 29, disease: 43). A total of 1,234 MRGs (Supplementary Table 1 in the online-only Data Supplement) were obtained from the previously published literature [22].
Identification and enrichment analysis of MRGs in schizophrenia
In the GSE21138 dataset, differentially expressed genes (DEGs) between normal and schizophrenia samples were screened using the limma package, with a threshold of p-value <0.05 and |LogFC| >0.1. The intersection between DEGs and MRGs was obtained through Venn analysis, resulting in the identification of DEMRGs specific to schizophrenia. The Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analyses of DEMRGs were performed using the clusterProfiler R package [23] to explore their potential functions.
Identification of hub genes in DEMRGs
The DEMRGs were input into the STRING database (https://string-db.org/) to construct a protein-protein interaction (PPI) network with a confidence score >0.4. The hub genes in DEMRGs were identified using the MCODE algorithm in the cytoHubba plugin of Cytoscape software [24], based on the PPI network data obtained from the STRING database.
Construction and validation of the diagnostic model based on hub genes
The LASSO model was established using the gene expression profiles of hub genes based on the glmnet R package. The minimum lambda value was used as a reference to determine the optimal variables to be included in the model. A logistic regression analysis was performed using the genes obtained from the LASSO model to construct a model formula that included the hub gene expression values and regression coefficients. The specific formula was as follows:
Receiver operating characteristic curve (ROC) curves were plotted using the pROC R package in each of the three datasets (GSE21138, GSE92538, and GSE27383) to evaluate the stability and sensitivity of the LASSO model. Additionally, ROC curves of hub genes were plotted in the three datasets to assess the accuracy of hub genes in diagnosing the disease.
Prediction of potential drugs
The online network tool Enrichr (https://maayanlab.cloud/Enrichr/) was used to explore the relationship between hub genes and drug sensitivity through access to the DSigDB database. The CellMiner database (https://discover.nci.nih.gov/cellminer/home.do) is a molecular and pharmacological database of 60 cell lines based on the National Cancer Institute, which contains transcriptome data for 60 cell lines and more than 100,000 natural products and compounds. Drugs approved by the U.S. Food and Drug Administration and drugs in clinical trials were selected for in-depth analysis. Then Spearman correlation analysis was conducted to determine the correlation between hub genes and drug sensitivity.
Construction of a miRNA-TF-hub gene network
The NetworkAnalyst website (https://www.networkanalyst.ca/) was used to create an interaction network between miRNA-TF-hub genes. The following parameters were specified: organism: Homo sapiens (human), collection ID type: official gene symbol, gene-miRNA interaction database: miRTarBase v8.0, and gene-TF interaction database: ENCODE.
RESULTS
Identification and analysis of DEMRGs
Through differential expression analysis, a total of 2,582 DEGs were identified in the GSE21138 dataset (Figure 1A). By conducting Venn analysis, we obtained 172 DEMRGs (Figure 1B). The results of enrichment analysis showed that DEMRGs were enriched in GO terms such as phospholipid metabolic process, phospholipid biosynthetic process, mitochondrial matrix, tricarboxylic acid cycle enzyme complex, and phosphatidylinositol phosphate phosphatase activity (Figure 1C). Furthermore, they were also found to be enriched in KEGG pathways including Carbon metabolism, peroxisome proliferator-activated receptor signaling pathway, citrate cycle (tricarboxylic acid cycle), phosphatidylinositol signaling system, biosynthesis of amino acids, alanine, aspartate and glutamate metabolism, tryptophan metabolism, and fatty acid degradation (Figure 1D).
Diagnostic model construction based on hub genes and model validation
Furthermore, using the MCODE algorithm, 18 hub genes selected were used to construct a LASSO logistic regression diagnostic model. The minimum lambda value was used as a reference during the process to determine the optimal variables to be included in the model. Eventually, six diagnostic model genes (ANGPTL4, CPT2, GLUD1, MED1, MED13L, and MED20) were identified (Figure 3A). The model formula was as follows:
Afterwards, the ROC curves of the model were plotted using the pROC R package in three datasets (GSE21138, GSE92538, and GSE27383). The results showed that the area under curve (AUC) values of the ROC curves in all three datasets were greater than 0.62, indicating that the diagnostic model constructed based on the six genes had good stability and sensitivity (Figure 3B). In addition, the evaluation of the diagnostic performance of the six individual genes revealed the following AUC values in GSE21138: 0.695 for ANGPTL4, 0.692 for CPT2, 0.714 for GLUD1, 0.685 for MED1, 0.738 for MED13L, and 0.692 for MED20 (Figure 4A); in GSE92538: 0.646 for ANGPTL4, 0.531 for CPT2, 0.571 for GLUD1, 0.67 for MED1, 0.631 for MED13L, and 0.595 for MED20 (Figure 4B); and in GSE27383: 0.499 for ANGPTL4, 0.493 for CPT2, 0.522 for GLUD1, 0.542 for MED1, 0.613 for MED13L, and 0.617 for MED20 (Figure 4C). These analyses collectively indicated that both the diagnostic model constructed using the six genes and the individual genes themselves exhibited good diagnostic performance.
Prediction of potential drugs
After screening using DSigDB, we obtained ten potential candidate drugs: rifampicin, fulvestrant, pentadecafluorooctanoic acid, eldecalcitol, aluminum chloride, digoxin, 2,2'-bipyridine, tesaglitazar, perfluoroundecanoic acid, and alfacalcidol (Figure 5A). Furthermore, the prediction results of the gene-drug interaction network showed that rifampicin may be associated with ANGPTL4 and MED1, fulvestrant with MED13L and MED1, pentadecafluorooctanoic acid with CPT2 and ANGPTL4, eldecalcitol with MED1, aluminum chloride with GLUD1, digoxin with MED13L and MED1, 2,2'-bipyridine with ANGPTL4, tesaglitazar with CPT2, and perfluoroundecanoic acid and GW0742 with ANGPTL4 (Figure 5B).
To further explore the clinical value of hub genes, we used CellMiner database to explore the correlation between hub genes and drug sensitivity. ANGPTL4 was correlated significantly and negatively with nilotinib, BP-102, vincristine, desfluoro-TAK-960, lexibulin, and OSU-03012. CPT2 was correlated significantly and negatively with dasatinib. MED13L was correlated significantly and negatively with vinorelbine. MED1 was correlated significantly and negatively with alectinib (Figure 6).
Construction of the miRNA-TF-gene interaction network
Through analysis, we identified a total of 34 miRNAs that potentially targeted and regulated the diagnostic model genes (ANGPTL4, CPT2, GLUD1, MED1, and MED20), as well as 137 TFs. Noteworthily, the TF SMAD5 may simultaneously interact with all five genes (ANGPTL4, CPT2, GLUD1, MED1, and MED20) (Figure 7).
DISCUSSION
Schizophrenia is a complex mental disorder with its etiology involving neurochemical and neurodevelopmental components [25]. Increasing evidence suggests that individuals with schizophrenia exhibit multifaceted mitochondrial dysfunctions [26,27]. For instance, a study by Beeraka et al. [28] revealed that mitochondrial DNA alterations, Nrf2 signaling pathway, dynamic changes in the dorsolateral prefrontal cortex, and oxidative stress activation contribute to the progression of schizophrenia [28]. Furthermore, it has been found that impaired cellular function, neuroplasticity, and disruption of brain circuits in individuals with schizophrenia may be attributed to compromised energy metabolism and increased oxidative stress, primarily regulated by mitochondria [17,29,30]. The aim of this study was to further explore the relationship between MRGs and schizophrenia, develop valuable diagnostic biomarkers and potential therapeutic targets for schizophrenia, and provide insights for future research in the diagnosis and treatment of individuals with schizophrenia.
Several biological processes, such as phospholipid biosynthetic process, phospholipid metabolic process, biosynthesis of amino acids, and alanine, aspartate, and glutamate metabolism, were found to be enriched among the identified DEMRGs in this study. Previous research has implicated disturbances in phospholipid metabolism in the pathogenesis of schizophrenia. Wang et al. [31], using untargeted liquid chromatography-mass spectrometry metabolomics, found differential expression of multiple phospholipids (phosphatidylcholines, lysophosphatidylcholines, phosphatidylethanolamines, lysophosphatidylethanolamines, and sphingomyelins) in individuals with schizophrenia compared to healthy controls, suggesting their association with the development of schizophrenia. Alanine and glutamate are both important amino acids that have been shown to be closely related to schizophrenia. Clinical research by Hatano et al. [32] revealed a correlation between increased plasma levels of alanine from the acute phase to the remission phase of schizophrenia and improvement in symptoms, suggesting that endogenous plasma alanine levels could serve as clinical markers for the severity and improvement of schizophrenia. Additionally, recent studies have implicated glutamate-mediated neurotransmission dysfunction in various neuropsychiatric disorders, including schizophrenia [33]. Meta-analysis of glutamate proton magnetic resonance spectroscopy studies has shown an elevation of glutamate metabolites in multiple brain regions in schizophrenia, suggesting that compounds that reduce glutamate transmission may hold therapeutic potential [34]. Taken together, these findings suggest that DEMRGs may regulate the disease progression of individuals with schizophrenia by modulating these biological processes.
ANGPTL4, CPT2, GLUD1, MED1, MED13L, and MED20 were identified as diagnostic biomarkers with diagnostic value in this study. Mediator complex (MED) is a large, evolutionarily conserved multi-protein complex that facilitates the interaction between TFs and RNA polymerase II in eukaryotes, and some MED subunits (such as MED13L, MED1, and MED20) have been found to undergo changes in the brain. Among them, MED13L and MED20 are associated with various genetic diseases, while MED1 is involved in post-stroke cognitive impairment [35]. Although the relationship between MED13L, MED1, MED20, and schizophrenia has not been discovered yet, research has found that MED1 is closely related to mitochondrial metabolism. Studies by Becerril et al. [36] and Li et al. [37] have shown that MED1 is involved in lipid synthesis, lipid metabolism, biosynthetic processes, glucose metabolism, and mitochondrial metabolic pathways. Enhancing MED1 can eliminate the promoting effect of miR-146a on lipid metabolism and mitochondrial function. ANGPTL4 is a secreted protein and an inhibitor of lipoprotein lipase-mediated plasma triglyceride clearance. It has been shown to be involved in the development of various diseases, including coronary artery disease, type 2 diabetes, and tumors [38-41]. Although the relationship between ANGPTL4 and schizophrenia has not been discovered yet, Wang et al. [42] found that many mitochondrial proteins are largely downregulated by ANGPTL4, and ANGPTL4 may induce its metabolic effects by regulating mitochondrial function and methionine. Carnitine palmitoyltransferase (CPT) 2 is a mitochondrial fatty acid oxidation enzyme that is involved in the entry of long-chain fatty acids into the mitochondria for β-oxidation and energy production [43,44]. Research by Virmani et al. [45] suggests that the activity of the CPT system (CPT1 and CPT2) is associated with Parkinson’s disease, Alzheimer’s disease, and schizophrenia, mainly related to changes in insulin balance in the brain. GLUD1 in the central nervous system is a crucial metabolic enzyme in glutamate metabolism and can lead to behavioral abnormalities and increased mPFC glutamate in schizophrenia patients [46]. Research by Yadav et al. [47] found that the absence of GLUD1 leads to abnormal emotional and social behavior. Therefore, based on these analyses, we speculated that ANGPTL4, CPT2, GLUD1, MED1, MED13L, and MED20 may be key genes influencing the progression of schizophrenia.
Through screening with DSigDB, we obtained ten potential candidate drugs: rifampicin, fulvestrant, pentadecafluorooctanoic acid, eldecalcitol, aluminum chloride, digoxin, 2,2'-bipyridine, tesaglitazar, perfluoroundecanoic acid, and alfacalcidol. Gene and drug targeting predictions showed that rifampicin may be associated with ANGPTL4 and MED1; fulvestrant may be associated with MED13L and MED1; pentadecafluorooctanoic acid may be associated with CPT2 and ANGPTL4; eldecalcitol may be associated with MED1; aluminum chloride may be associated with GLUD1; digoxin may be associated with MED13L and MED1; 2,2'-bipyridine may be associated with ANGPTL4; tesaglitazar may be associated with CPT2; and perfluoroundecanoic acid and GW0742 may be associated with ANGPTL4. However, there is limited research on the relationship between these drugs and schizophrenia or their target genes. Therefore, further research is still needed in this area.
In conclusion, this study identified six MRGs (ANGPTL4, CPT2, GLUD1, MED1, MED13L, and MED20) with diagnostic significance for schizophrenia through bioinformatics analysis. Additionally, several miRNAs, TFs, and candidate drugs targeting MRGs were identified. These findings provide potential research directions for understanding the pathogenesis of schizophrenia and exploring the possible drugs. However, this study is limited by the lack of corresponding clinical research and experimental confirmation.
