Identification of immune-related biomarkers in peripheral blood of schizophrenia using bioinformatic methods and machine learning algorithms

Abstract

Schizophrenia is a group of severe neurodevelopmental disorders. Identification of peripheral diagnostic biomarkers is an effective approach to improving diagnosis of schizophrenia. In this study, four datasets of schizophrenia patients' blood or serum samples were downloaded from the GEO database and merged and de-batched for the analyses of differentially expressed genes (DEGs) and weighted gene co-expression network analysis (WCGNA). The WGCNA analysis showed that the cyan module, among 9 modules, was significantly related to schizophrenia, which subsequently yielded 317 schizophrenia-related key genes by comparing with the DEGs. The enrichment analyses on these key genes indicated a strong correlation with immune-related processes. The CIBERSORT algorithm was adopted to analyze immune cell infiltration, which revealed differences in eosinophils, M0 macrophages, resting mast cells, and gamma delta T cells. Furthermore, by comparing with the immune genes obtained from online databases, 95 immune-related key genes for schizophrenia were screened out. Moreover, machine learning algorithms including Random Forest, LASSO, and SVM-RFE were used to further screen immune-related hub genes of schizophrenia. Finally, CLIC3 was found as an immune-related hub gene of schizophrenia by the three machine learning algorithms. A schizophrenia rat model was established to validate CLIC3 expression and found that CLIC3 levels were reduced in the model rat plasma and brains in a brain-regional dependent manner, but can be reversed by an antipsychotic drug risperidone. In conclusion, using various bioinformatic and biological methods, this study found an immune-related hub gene of schizophrenia - CLIC3 that might be a potential diagnostic biomarker and therapeutic target for schizophrenia.

Keywords: CIBERSORT; CLIC3; LASSO; WGCNA; peripheral immune-related biomarkers; random forest; schizophrenia; support vector machine.

Figure 1

Flow chart of the present study. DEG, differentially expressed gene; LASSO, least absolute shrinkage and selection operator; PBMC, peripheral blood mononuclear cells; ROC, receiver operating characteristic; SCZ, schizophrenia; SVM-RFE, support vector machine – recursive feature elimination; WGCNA, weighted gene co-expression network analysis.

Figure 2

Normalization, merging, and de-batching of the four datasets. (A,B) Boxplots of the four datasets before and after removing the batch effect. (C,D) Density maps of the four datasets before and after removing the batch effect. (E,F) UMAP maps of the four datasets before and after removing the batch effect.

Figure 3

Weighted gene co-expression network analysis. (A,B) The values of soft-threshold power based on scale independence and mean connectivity of the weighted gene co-expression network analysis (WGCNA). (C) Nine co-expression modules of WGCNA. (D) Cluster dendrogram of genes of WGCNA. Each color represented a module, and the gray module included the genes that could not be classified into any module. (E) Heatmap of the eigengene network representing the relationships among the modules and the clinical trait status Heatmap of correlations between module characteristic genes (MEs) and phenotype of clinical traits (type of disease). Red represented correlation and green represented p-value. (F) Correlation of gene significance (GS) and module membership (MM) of the cyan module.

Figure 4

Differentially expressed gene analysis. (A) Hot map of the differentially expressed genes (DEGs). (B) Volcano map of the DEGs. (C) Venn diagram of the genes screened by weighted gene co-expression network analysis (WGCNA) and DEGs. −, p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Figure 5

Functional enrichment analyses. (A) Signaling pathways of the Kyoto Encyclopedia of Genes and Genomes (KEGG) enrichment analysis. (B) Biological processes of the Gene Ontology (GO) enrichment analysis. (C) Cellular components of the GO enrichment analysis. (D) Molecular function of the GO enrichment analysis (the color of the bubble represents the p-value, and the size of the bubble represents the number of genes).

Figure 6

Immune cell infiltration analysis. (A) Relative percentage of 22 groups of immune cells in each sample. (B) Differences in immune infiltration between schizophrenia and control samples, including 4 significantly different immune cell groups: eosinophils, M0 macrophages, resting mast cells, and gamma delta T cell. (C) Principal component analysis for the immune cell infiltration between schizophrenia subjects and healthy controls. (D) Venn diagram indicating 95 immune-related key genes for schizophrenia. −, p > 0.05; *p < 0.05; **p < 0.01; ***p < 0.001; ****p < 0.0001.

Figure 7

Identification of immune-related hub genes for schizophrenia. (A) The Random Forest analysis indicates 36 candidate hub genes. (B) The 36 candidate hub genes generated by the Random Forest algorithm are ranked by ‘MeanDecreaseAccuracy’ (left) and ‘MeanDecreaseGini’ (right), respectively. (C) The least absolute shrinkage and selection operator (LASSO) analysis indicates 35 candidate hub genes. (D) The support vector machines - recursive feature elimination (SVM-RFE) analysis indicates 5 candidate hub genes. (E) A Venn diagram indicating one hub gene of schizophrenia by intersecting the three groups of candidate genes revealed by the three machine-learning algorithms, respectively. (F) The mRNA expression of CLIC3 between the schizophrenia patients and healthy controls (****p < 0.0001). (G) The relationship between CLIC3 and immune cells. (H) The ROC curve of CLIC3 to assess the accuracy of CLIC3 to potentially differentiate between the schizophrenia patients and healthy controls.

Figure 8

Protein expression of CLIC3 in the rat plasma and brains. (A) CLIC3 expression in the plasma. (B) CLIC3 mRNA expression in schizophrenia patients’ cortex. (C) CLIC3 expression in the prefrontal cortex (PFC). (D) CLIC3 expression in the caudate putamen (CPu). (E) CLIC3 expression in the nucleus accumbens (NAc). (F) CLIC3 expression in the hippocampus (HIP). *p < 0.05; +mean value.