Primary membranous nephropathy (PMN) is a common cause of primary nephrotic syndrome in middle-aged and elderly patients and one of the leading causes of end-stage renal disease (ESRD) worldwide(1). The pathology of the glomeruli from patients with PMN features
Figure 7: External validation of the candidate diagnostic genes in the validation dataset. (a) The gene expression level of CEBPD. (b) The gene expression level of MYOZ2. (c) The gene expression level of ZNF189.
thickening of the capillary wall and formation of subepithelial “spikes” of basement membrane(4). Although the landmark discoveries of anti-phospholipase A2 receptor (PLA2R) has widely applied clinically, with the presence of anti-PLA2R in only 60–70% for patients with PMN(6), in most centers, a diagnostic renal biopsy remains the key method for diagnosis of PMN(4). In addition, previous studies have showed that immune cell infiltration may plays
significant roles in the development of PMN(11–14). Therefore, higher rate of diagnosis can be achieved with novel diagnostic biomarkers and mechanisms of PMN can be studied in depth with a fuller characterization of the pattern of immune cell infiltration into the kidney tissues from patients with PMN. In recent years, bioinformatics methods have been widely applied to discover potential diagnostic biomarkers and analyze the patterns of immune cell infiltration
in tissues. Nevertheless, few studies have concentrated on clarifying the relationships between candidate gene biomarkers and the immune cell infiltration in PMN. Therefore, in the present study, we tried to identify candidate diagnostic biomarkers for PMN and further analyzed the role of immune cell infiltration in PMN.
We downloaded the PMN expression profile datasets from the GEO database and identified a total of 86 DEGs. Go pathway analysis indicated that the DEGs were principally related to small molecule catabolic process, cellular amino acid metabolic process, apical part of cell, apical plasma membrane, and carboxylic acid binding. Do pathway analysis demonstrated that the DEGs were principally related to essential hypertension, kidney disease, and urinary system disease. In addition, GSEA showed that the following principal pathways were enriched: adhesion molecules, glycosylphosphatidylinositol (GPI) anchor biosynthesis, natural killer cell mediated cytotoxicity, type I diabetes mellitus, and viral myocarditis. A previous research has confirmed that PLA2R as the auto-antigen, presenting at the level of the foot process as well as on the apical plasma membrane of podocytes(18), enables continued deposit formation on the cell surface, where they begin to activate complement(19). Haddad et al.(20) showed that some specific cellular amino acid metabolic processes are the driver of formation of membrane attack complex (MAC) and podocyte damage after complement activation. In addition, a weakening of adhesion is one of the common features of podocyte damage, which could be affected by shedding of adhesion molecules through activity regulation of cell surface-expressed proteases(21). Above researches are consistent with our results and suggest that these processes and pathways may play significant roles in the development and progression of PMN.
Machine-learning is increasingly being used to help identify the potential diagnostic biomarkers and its satisfied accuracy has been verified in experiments of previous studies(8–10). Comparing with simply using the degree of gene difference as the screening standard, machine- learning algorithms screened the characteristic genes through the construction of the diagnostic model who has the most accuracy of diagnosis, which was more convincing for identifying the potential diagnostic genes. In the present study, three machine-learning algorithms were used and we took an intersection of the results for accuracy. The model of random forest (RF) refers to a non-parametric approach that uses an ensemble decision tree to achieve the classifying process under the supervision(22), in which random subsets are drawn from the data, with replacement(23). Support vector machine-recursive feature elimination(SVM-RFE) is a machine-learning method of pattern recognition and estimation of function(24) that achieves extensive applications to rank features and select the significant ones for classification(25). Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression refers to a machine-learning method that determines the variable by finding λ when the classification error is the smallest(26), which avoid the over-fitting in construction of model(27). In this study, we finally identified CEBPD and MYOZ2 as the diagnostic marker genes, which showed a satisfied efficacy of diagnosis in both the training set and the validate set.
CCAAT/enhancer-binding protein delta (CEBPD) belongs to the family of CCAAT/enhancer-binding protein, and the protein as transcription factors, its function covers many biological
processes including metabolism and immune responses, cell differentiation, proliferation (28). Yamaguchi et al(29) demonstrated that CEBPD regulates the expression of hypoxia inducible factor (HIF)-1α protein not only under hypoxia but also under inflammatory conditions. Li et al(30) indicated that interstitial HIF-1α may be involved in renal interstitial fibrosis in diabetic nephropathy under the regulation of CEBPD. Bechara et al(31) showed that induction of autoimmune nephritis is controlled by an insulin-like growth factor (IGF)-2 mRNA-binding protein (IMP)2/m6A epitranscriptomic axis through CEBPD and its family members. On account of activation of the complement system was the main pathological processes of PMN(19) which promote inflammation and thrombosis(32), we speculate that CEBPD may be involved in the regulation of immune processes in the pathogenesis of PMN. The protein encoded by myozenin (MYOZ)2 belongs to a sarcomeric protein family that tether calcineurin to alpha-actinin at the z-line of the sarcomere mainly in cardiac and skeletal muscle cells(33), which is important for calcineurin signalling. T cells being activated by T cell receptor (TcR)-calcium-calcineurin signalling pathway is a crucial part of adaptive immune response.(34) Previous researches have proved that calcineurin inhibitors affect the humoral immune response by interfering with T helper signals(35). In addition, calcineurin inhibitors (CNIs) can be applied to achieve remission in PMN clinically for its immunosuppressive and antiproteinuric effects(36). Therefore, we speculate that MYOZ2 may be involved in pathogenesis of PMN by regulating T cell receptor (TcR)-calcium-calcineurin signalling pathway. The results of previous studies have suggested that CEBPD and MYOZ2 may play significant roles in the inflammation and the progression of autoimmune nephropathy. Therefore, these two genes may represent meaning diagnostic biomarkers of PMN, but their efficacy for diagnostic purposes needs further clinical studies to confirm.
In the present study, CIBERSORT was used to comprehensively characterize the nature of the immune cell infiltration into the glomeruli of patients with PMN. We found that there was greater infiltration with B memory cells and neutrophils, and less infiltration with monocytes, which may be involved in the development and progression of PMN. Nagata et al(14) proved that infiltrating immune cells presence in the glomeruli of patients with multiple idiopathic nephrotic syndrome using indirect immunofluo rescence. In the previous study, a prominent accumulation of B cells in the secondary interstitial of involvement of PMN was detected(37), which may suggested that B cells and tertiary lymphoid organs may play a significant role in renal inflammation(13). In addition, Zheng et al(38) discovered that patients with PMN have elevated urinary monocyte chemotactic protein (MCP-1) level, which indicates that monocytes are prompted to migrate toward the chemokine source in patients with PMN(39). Combining with our results, above researches have shown that B cells and monocytes play significant roles in PMN and are worthy of further studies. However, studies revealing the role of neutrophils in PMN are deficient, and further experimental data are required. Besides, the present study revealed the interaction of 21 subtypes of infiltrated immune cells in patients with PMN. M1 macrophages are positively correlated with resting mast cells, and memory B cells are negatively correlated with naïve B cells and monocytes. What role, if any, these correlations may have in the pathogenesis of PMN remains unclear. Whether these cells are capable of producing proteinuria or whether these correlations influence the process of end-stage renal disease (ESRD) must await determination of their functional properties.
An analysis of the correlations between the expression of the two candidate diagnostic marker genes and the extent of infiltration with subtypes of immune cells revealed close positive correlations of CEBPD expression with the extent of memory B cells, neutrophils, and
delta gamma T cells infiltration. In addition, CEBPD expression negatively correlated with the
numbers of monocytes, activated NK cells and M1 macrophages, and the expression of MYOZ2 negatively correlated with the numbers of neutrophils and resting dendritic cells. Balamurugan et al(40) demonstrated that a positive feedback loop may forms after the activation of TLR4 gene expression by CEBPD in macrophages. Another previous study showed that CEBPD also can be stabilized by deletion of the E3 ubiquitin ligase Cop1 to decrease the expression of macrophage chemoattractant genes(41). Yan et al (42) indicated that mice without the CEBPD gene displayed significant attenuation of the neutrophils in bronchial alveolar lavage fluids. In addition, CEBPD influences the recruitment of monocytes by involving in IL-1β-mediated MCP-1 expression(43). Therefore, we speculated that the two candidate diagnostic marker genes may participate in the development and progression of PMN in a similar approach. However, this hypothesis requires further testing in preclinical and clinical contexts.
In the present study, novel bioinformatic methods was used to identify potential diagnostic biomarkers and analyze immune cell infiltration in kidney tissues from patients with PMN. Nevertheless, there were several limitations to the present study. First, clinical information was not available for the datasets that the influences of other disease factors on expression of genes could not be assessed. Second, samples of patients and controls in datasets were regional and small that cannot represent samples in other areas. Third, analysis in the present study was based on the second-degree data mining and samples were derived from open datasets. Therefore, further research is needed to validate the present finding.