Comprehensive identification of immune-associated biomarkers based on network and mRNA expression patterns in membranous glomerulonephritis

We download all the genes related with immune from AmiGO 2 version: 2.4.26 in Homo sapiens species, constituting 3068 immune-related genes from 651 records [19].

The MGN-related genes were downloaded from DisGeNET, which is a discovery platform containing one of the largest publicly available collections of genes and variants associated to human diseases. The database collected 561,119 gene-disease associations, between 17,074 genes and 20,370 diseases, disorders, traits, and clinical or abnormal human phenotypes [20]. At last, we got 90 MGN-related genes including the records labeled as membranous glomerulonephritis and idiopathic membranous glomerulonephritis.

The protein–protein interaction (PPI) data for constructing the human protein interaction network were download from the HPRD (Human Protein Reference Database) database [21]. More than 42,000 manually curated interactions between 9826 human genes were contained in the database.

First, a human PPI network associated with immune and MGN was constructed based on the HPRD data. Then, a sub-network of the human PPI network named the immune or MGN-directed neighbor network (IOMDN network), which including immune or MGN-related genes and their direct interacting genes in the network (referred to as immune neighbor genes) using all the immune-related genes and MGN-related genes from AmiGO and DisGeNET. Finally, we extracted all MGN-related genes to construct an MGN-related genes-directed network (MGND network) for subsequent analysis. We used cytoscape software to construct network and analyze the topological properties of nodes in both IOMDN and MGND network.

We identified all the network modules using MCODE follow the default parameters based on the MGND network constructed by the MGN-related genes (http://​apps.​cytoscape.​org/​apps/​mcode). MCODE is a tool which clusters a given network based on topology to find densely connected regions. We identified 20 modules and extracted the five modules with top numbers of nodes.

We obtained the expression data of MGN from GEO (Gene expression Omnibus). The MGN expression profile data included 21 diseases samples and 18 control samples (GSE99340) [22]. We calculated the Pearson correlation coefficients (PCCs) between two nodes either in MGND network or modules used by disease samples expression data. We also used unpaired t-test to get the p values of differential expression genes between disease and control samples for all the nodes in the modules.

Gene ontology (GO) and KEGG pathway enrichment analyses were performed by the DAVID functional annotation web server using default parameters [23]. We obtained enriched GO terms and pathways (p < 0.01).

With the Enrichr tool online web server using default parameters, drug enrichment was performed for all the genes in five modules. We obtained enriched drugs (p < 0.01) [24].

A sub-network of the human PPI network named the immune or MGN-directed neighbor network (IOMDN network) is constructed and the IOMDN network contains 6251 nodes and 19,983 edges (Fig. 1a). We find 56 immune-related genes that are themselves MGN-related genes such as MMP, CTLA4, CD19 and so on. We also discover 34 only MGN-related genes and 3011 only immune-related genes (Additional file 1: Table S1). We analyze the degree distribution of immune and MGN-related genes, immune-related genes, MGN-related genes and other genes, respectively. All the degree of these genes show the scale-free distribution (Fig. 1b). Especially, we further find a higher degree of genes (mean = 19.5) that were both related with immune and MGN in the IOMDN network and indicating complex links between the immune-related genes and MGN-related genes (Fig. 1c). The top five genes including TP53, EP300, SRC, CREBBP and GRB2 which are all immune-related genes have a direct connection and highest degree in the network, indicating that immune-related genes play hub roles in the IOMDN network (Fig. 1d). Notably, these five top immune-related genes are all interacted with some MGN-related genes (Fig. 1e). For example, immune-related gene SRC directly connects with MGN-related gene including TPRC6, CDKN1B, PRKCA and so on. All above results indicate that immune-related genes play crucial roles in membranous glomerulonephritis.

We construct a network named MGN-related genes-directed network (MGND network) to further explore the relationship between immune-related and MGN-related genes (Fig. 2a). This sub network is extracted from the IOMDN network by selecting the MGN-related genes and their directly connected genes. We find the MGND network also show scale-free distribution (Fig. 2b). There are 53 immune and MGN-related genes, 25 only MGN-related genes, 408 only immune-related genes and 619 other genes in MGND network (Fig. 2c). Compared to the topological features of nodes in IOMDN, the nodes in MGND have higher network density (avg. 0.007 vs 0.001), degree (avg. 8.009 vs 6.389) and clustering coefficient (0.154 vs 0.097). This result indicates that the MGND extracted from IOMDN is closer in structure and played a crucial role in the biogenesis of MGN.

Next, we integrate gene expression data for the MGND network to further study the relationships between MGN-related and immune-related genes on expression level. The interaction relationships in MGND hold 22.22% of interactions in IOMDN. 19.57% interactions are significantly co-expression in all the interactions of MGND network (Fig. 2d). We also find 643 (73.99%) significantly co-expressed interactions in MGND and show strong correlation (Pearson correlation coefficients > 0.5) (Fig. 2e). The strong correlations between the nodes in MGND network indicate that the association between immune-related genes and MGN-related genes not only show on topological structure but also show on expression pattern. We further find immune and MGN-related genes are significantly correlated with a large number of immune-related genes (Fig. 2f). For instance, immune and MGN-related gene NPHS1 is strong correlated with immune-related gene FYN (PCC = 0.88). In addition, these significant correlations contain both positive and negative correlation. The MGN-related gene PRKCA is positive correlated with MBP and ITPKB (PCC = 0.61, 0.6), however, negative correlated with PTPN6 and THOCA5 (PCC = − 0.48, − 0.74). The results show that the complex properties between the communication of MGN-related genes and immune-related genes.

We perform a module analysis for the MGND network to reveal the communication between immune-related genes and MGN-related genes. We extract five modules include top number of nodes and also analyze their expression patterns (Fig. 3). The first module contains three immune and MGN-related genes, 14 immune-related genes and six other genes. Integrating expression data, we find MGN-related gene PRKCA has strong negative correlation with immune-related gene PTPN6 (PCC = − 0.48, p = 0.027) (Fig. 3a). The second module included two immune and MGN-related genes, one MGN-related gene, eight immune-related genes and 10 other genes (Fig. 3b). The genes MMP9 and APP were also related with other genes on expression level. There were one immune and MGN-related gene, 15 immune-related genes and four other genes in the third module. The most of immune-related and MGN-related genes in third module had strong positive and negative correlations (Fig. 3c). For example, immune and MGN-related gene HRAS were negative correlated with two immune-genes. The fourth and fifth modules also contained strong interaction relationships (Fig. 3d, e). We also discovered the direction of correlation was complex in four modules (Fig. 3f).

After exploring the interactions between immune-related genes and MGN-related genes in five modules, we also consider if there are differential expressed genes for MGN patients and control samples. By integrating expression data, we discover diverse number of differential genes in all the modules. We find 20 differential genes including nine up-regulated and 11 down-regulated genes in all the 23 genes for the first module (Fig. 4a). Notably, the differential expressed MGN-related gene PRKCA is negative correlated with immune-related gene PTPN6. For the second module, 19 genes of all the 22 genes are differential expressed contained ten up-regulated and nine down-regulated genes (Fig. 4b). There are 12 differential genes including five up-regulated and seven down-regulated genes. We also discover that immune and MGN-related gene HRAS is strongly negative correlated with immune-related gene RASA1 (PCC = − 0.7, p = 0.00042) (Fig. 4c). The fourth and fifth module also show the similar phenomenon (Fig. 4d). The differential expression details of all genes in five modules are showed in Additional file 2: Table S2. Our results show that some immune-related and MGN-related genes are differential expressed between MGN patients and control samples and these differential expressed immune-related genes and MGN-related genes are co-expressed in MGN patients. It indicates that the immune-related genes are associated with MGN-related genes on expression level.

We perform GO enrichment analysis based on all the genes in the first four modules for MGN. We find these genes in diverse modules are enrichment in some different kinds of GO terms (Fig. 5a). The first and fourth modules are enrichment in some phosphorylation-associated GO terms such as positive regulation of protein phosphorylation, peptidyl-tyrosine phosphorylation and so on. Previous study reports that glomerular cluster in is associated with the phosphorylation of PKC-α/β regulation and good outcome of membranous glomerulonephritis in human [25]. It indicates that the phosphorylation of vascular permeability factor/VEGF-A is essential for the proper maintenance of glomerular filtration barrier and the glomerular endothelial fenestrae follow recent data from podocyte-specific knockout mice as well as studies using neutralizing antibodies [26‐29]. Besides, the genes in second module are major enrichment in some GO terms related with immune such as T cell receptor signaling pathway, B cell receptor signaling pathway, B cell differentiation and so on. As we know, some studies suggest that the inhibition of B cell function is associated with beneficial effects on proteinuria in MGN, and human studies clearly demonstrate that the inhibition of B cells with alkylating agents induces remission of the nephrotic syndrome [30, 31]. Experimental results show that T cell subset modulated immunoglobulin production in MGN [32]. The third module are almost enrichment in some GO terms including negative regulation of apoptotic process, positive regulation of neuron apoptotic process and so on. The result of experiment in MGN shows Haeme oxygenase (HO)-1 is considered treat the disease via anti-apoptotic and immunomodulatory effects [33].

Notably, we discover that the genes in these five modules are related with a pathway called B cell receptor signaling pathway (Fig. 5b). In this pathway, we find some immune-related genes including CD22, PTPN6, MAP2K1 and so on and these immune-related genes are key genes for the B cell receptor signaling pathway. In addition, we also find a MGN-related gene HRAS in the third module was play its role in pathway. We also construct an inflammation-related drugs and genes network which is extracted from drug enrichment result (Fig. 5c). The drugs in the network contain Cyclosporin, Indomethacin, Sulindac, Leflunomide and Nimesulide which all could treat inflammation. It also indicates that the essential associations among MGN, immune and inflammation on treatment. In addition, we also get the MGN-related papers for each gene in five modules according to incomplete statistics in pubmed. We found several genes which have been studied in a large number of MGN-related publications due to their important and diverse roles in MGN (Additional file 3: Table S3). The result suggests that the genes in modules can identify genes important in MGN.