Landscape of the 177 FRGs
The current study outlines its research approach using Fig. 1, which presents the overall research methodology. To eliminate batch effects from the dataset, we employed the "sva" R package. We used PCA to verify the consistency of the sample distribution prior to and after correction. Figure 2A displays the scatter distribution of the three datasets before batch effect removal, while Fig. 2B depicts the scatter distribution after correction, indicating the successful removal of all confounding factors from the rectified samples. To explore the differential expression levels of 177 FRGs between healthy and RIF patients, we applied the "limma" R package (Figs. 3A and 3B). Our analysis revealed 18 differentially expressed FRGs, including SERPINE1, S100A8, CCL11, TUBA1A, IFNG, FN1, MMP9, S100A9, CP, CCL2, FCGR2A, TIMP1, and CCL5, which exhibited over-expression in patients with RIF. Conversely, ALB, PLG, EPHX1, VEGFA, and SLC26A9 were down-regulated (Tab S2). Figures 3C and 3D illustrated the functional and molecular pathways associated with these FRGs (Supplementary Tables 3 and 4).
Recognition of of hub FRGs
To further refine the FRGs associated with RIF, we utilized a range of analytical tools, including the SVM-RFE algorithm, RF algorithm, LASSO analysis, and PPI network analysis, to identify specific genes with key regulatory roles. The ten-fold cross-validation curve demonstrated that the accuracy of SVM-RFE was highest when 12 FRGs were selected (Fig. 4A). Subsequently, the RF model identified 11 FRGs (Fig. 4B) and the LASSO algorithm identified 10 FRGs among the 18 differentially expressed FRGs (Fig. 4C). We explored the interaction network of these 18 FRGs using the String database and visualized the information using the Cytoscape software (Fig S1). Applying the MCODE algorithm, we identified 11 FRGs (Fig. 4D). From these three analytical approaches and the PPI network analysis, we identified CCL5, TIMP1, ALB, and IFNG as the four FRGs that were related to RIF (Fig. 4E).
Development of a nomogram model
Building upon the identification of the four FRGs, we developed a nomogram to forecast the occurrence of RIF (Fig. 4F). Our calibration curve analysis indicated that the nomogram possesses accurate predictive ability (Fig S2A). Predicated on the DCA curve, the nomogram holds promise for RIF patients, with the red line remaining above the gray and black lines within the range of 0 to 1 (Fig S2B). Our clinical impact curve revealed that the nomogram model possesses robust predictive strength (Fig S2C). Furthermore, the AUC values of the four FRGs exceeded 0.75, indicating the nomogram's strong sensitivity and specificity in predicting RIF prevalence (Fig. 5).
Recognition of subtype classification
Basing our analysis on the four key FRGs, we identified two distinct FRG patterns (FRG clusters A and B) using the "ConsensusClusterPlus" package and consensus clustering method in R software (Fig. 6A and Fig S3). We generated histograms to visualize the differences in the expression levels of these four FRGs between the FRG clusters (Fig. 6B). We observed that CCL5, TIMP1, and IFNG were more highly expressed in FRG cluster A, while ALB was expressed more highly in FRG cluster B. Our PCA analysis demonstrated that the FRG clusters can be fully distinguished from the expression of four FRGs (Fig. 6C). Moreover, by employing ssGSEA, we plotted histograms to visualize the differences in immune cell (Fig. 6D). Our results indicate that apart from neutrophils, type 17 T helper cells, and immature dendritic cells, the abundance of other immune cells was significantly higher in FRG cluster A (p < 0.05). Additionally, we evaluated the correlation between the four FRGs and immune cells, classifying the four key FRGs into groups with different expressions and then constructing a histogram (Fig. 6E). Lastly, we evaluated the association between immune cell infiltration and four FRGs, using the same classification strategy to group the FRGs based on their median expression level before plotting the resulting histogram (Fig. 7A).
Function enrichment
462 DEGs were identified in the two FRG patterns using a threshold of |logFC| ≥ 1 and an adjusted p value < 0.05 (Tab S5). To gain further insights into the potential roles and molecular pathways of these DEGs in RIF, we conducted GO and KEGG analyses. The GO analysis revealed that the DEGs were enriched in biological processes related to cellular amino acid metabolic process, alpha-amino acid metabolic process, and small molecule catabolic process. In terms of cellular components, the DEGs were associated with the apical part of the cell, apical plasma membrane, and brush border. The molecular function analysis indicated secondary active transmembrane transporter activity, solute:sodium symporter activity, and active ion transmembrane transporter activity as relevant terms (Fig. 7B and Tab S6). Furthermore, the KEGG enrichment analysis demonstrated that the DEGs were highly enriched in pathways such as phagosome, viral protein interaction with cytokine and cytokine receptor, rheumatoid arthritis, staphylococcus aureus infection, hematopoietic cell lineage, and chemokine signaling pathway (Fig. 7C and Tab S7).
Genetic patterns and immune landscape
To validate the FRG patterns, we divided RIF patients into different genomic subtypes from the 462 DEGs. This resulted in two different gene patterns, gene cluster A and gene cluster B, which corresponded to the grouping of FRG patterns (Fig. 8A). PCA analysis revealed that the two gene patterns were fully distinct, further validating the grouping approach (Fig. 8C). Figures 8B and 8D showed that the differential expression levels of the four key FRGs between gene cluster A and gene cluster B and the immune cell infiltration between the two gene patterns were similar to those observed in the FRG clusters. This again validates the correctness of our consensus clustering approach for grouping.
The role of FRG and gene patterns in RIF
To quantify the FRG cluster, we employed a PCA algorithm to compare FRG scores between two distinct FRG clusters or gene clusters. The analysis revealed that FRG cluster A, or gene cluster A, exhibited significantly higher FRG scores compared to FRG cluster B, or gene cluster B (Fig. 9A). The relationship between FRG patterns, gene patterns, and FRG scores is graphically represented in the Sankey plot (Fig. 9B). In order to deepen our understanding of the connection between the FRG cluster and RIF, we further examined the relationship between different clusters and the expression levels of genes closely associated with RIF development. These genes included ATP6V1B1, AGTR1, IL6, TNF, SLC22A12, VEGFA, PKD2, B2M, WT1, and HIF1A. Our findings indicated that FRG cluster A is strongly linked to RIF characteristics (Figs. 9C and D).
Discussion
With a global prevalence of 8–16%, CKD has emerged as the third most prominent cause of premature death globally, trailing only behind AIDS and diabetes [27]. CKD progression is caused by various risk factors, including hyperlipidemia, diabetes, obesity, age, and lifestyle [28]. Fibrosis is a pathological state that exists in various types of tissue injury, mostly due to an uncontrolled tissue repair response, and is commonly found in chronic inflammatory diseases [29]. RIF, commonly seen as a pathomorphological feature of CKD, is also a key factor in determining the progress of renal failure. The pathogenesis of RIF is closely related to the abnormal regulation of the immune system, which mainly includes cytokine imbalance, abnormal polarization of immune cells, and tissue damage [30]. RIF is regarded as one of the most reliable predictors of the development of CKD [31]. At present, there are few effective treatment options available for CKD and its related complications, resulting in a grim outlook for patients' prognoses. Hence, delaying and inhibiting RIF damage might provide novel targets and directions for the treatment of CKD.
In the research, a total of 124 healthy and 175 RIF patients were retrieved from the GEO databases. From the genecard database, we screened 177 FRGs with relative scores greater than 10. We pre-processed the data to eliminate batch effects using the "ComBat" packages of R. In order to determine the FRGs linked to RIF, we performed differential analysis between healthy and RIF patients. We conducted enrichment analysis to explore the molecular functions of the differentially expressed FRGs. Besides, we used SVM-RFE, RF, LASSO, and PPI network analysis to identify hub FRGs. CCL5, TIMP1, ALB, and IFNG were identified as being remarkably related to RIF by the combined analysis approach. Based on the four candidate FRGs, we formulated a nomogram to forecast the occurrence of RIF. The result showed that the model possesses accurate predictive ability. In addition, we adapted the consensus cluster method to recognize the different patterns of FRGs. We found that CCL5, TIMP1, and IFNG were higher expressed in FRG cluster A, while ALB was higher expressed in FRG cluster B. We utilized ssGSEA analysis to predict the immune landscape of RIF, and the result was a significantly higher abundance of other immune cells in FRG cluster A except neutrophils, type 17 T helper cells, and immature dendritic cells. Finally, we used the PCA algorithm to compare FRG scores between distinct FRG clusters or gene clusters. Our findings suggested that FRG cluster A is strongly related to RIF characteristics.
CCL5 (also called RANTES) is a chemotactic cytokine involved in regulating fibroproliferation after inflammatory injury. In the early stages of immune diseases, CCL5 mainly affects the inflammatory response and immune cell aggregation. When the degree of disease is aggravated, it encourages the proliferation and migration of fibroblasts, collagen synthesis, and deposition of matrix proteins, thus promoting the progression of fibrosis [32; 33]. CCL5 expression levels correlate with the progression of inflammatory disease fibrosis and might serve as a predictive biomarker for the occurrence and prognosis of related inflammatory diseases [34]. A previous study observed that CCL5 promoted liver fibrosis in non-alcoholic fatty liver disease [35]. CCL5 was also found to be engaged in the pathology of chronic renal fibrosis disease [36]. Recent studies revealed CCL5 as an inflammatory chemokine potentially implicated in the progression of CKD [37]. The results of our analyses are broadly similar to the findings of the above studies, indicating that CCL5 might have a major role in the fibrotic lesions of diseases.
TIMP1 is an endogenous protease inhibitor that inhibits matrix metalloproteinase secretion, thereby reducing ECM protein degradation. Earlier studies found that serum levels of TIMP1 were strongly related to the developmental stage of liver fibrosis in hepatopathic patients, indicating that it may be a potential predictor of the severity of liver fibrosis [38]. Aberrant expression of TIMP1 may promote the progression of chronic renal fibrosis [39]. Furthermore, targeting TIMP1 was revealed to be potentially effective for antimyocardial fibrosis therapies [40]. The anti-tumour effect of TIMP1 has been recognized by worldwide scholars. Recent studies have shown that it can facilitate the growth of cancer cells under certain circumstance, and might have anti-apoptotic effects as well [41; 42]. At present, less studies have been conducted on the association between TIMP1 and RIF, and our results provide a reference for exploring the relationship between TIMP1 and RIF. Dysregulation of IFNG production/signaling is closely associated with diverse pathological changes [43]. Earlier studies have found that IFNG provided good therapeutic results in idiopathic pulmonary fibrosis [44]. Besides, IFNG has a key effect on the process of RIF and progression to CKD [45]. Current research remains controversial on the role of IFN, which is considered to be two-fold in tumour immunotherapy. On the one hand, IFNG prevents tumour growth by inducing tumour cells into dormancy; on the other hand, tumour cells continue to progress in some cancer patients despite high serum levels of IFNG after immunotherapy [46; 47]. A recent review suggested that IFN-Ⅰ promotes RIF and facilitates renal microangiopathy [48]. The evidence of studies on the pathogenesis of IFN and RIF is relatively scarce, and we need to further explore their relationships.
However, some limitations of this study were present. Firstly, the sample size included in this study was relatively small, and some analytical bias may be possible. Secondly, owing to the lack of complete clinical information (age, gender, etc.) in the included datasets, we could not explore the association of different clinical data with different subgroups. Lastly, because of the lack of tissue samples, our study did not perform basic and clinical studies for validation of the findings, and we will further validate the analysis results in the future.