Single-cell transcriptomics and machine-learning reveal M1 macrophage-driven progression from minimal change disease to focal segmental glomerulosclerosis

Abstract

Minimal change disease (MCD) and focal segmental glomerulosclerosis (FSGS) are two key nephrotic syndrome types with significant clinical implications. MCD predominantly affects children, while FSGS is more common in adults, often leading to irreversible kidney dysfunction. Despite shared features like podocyte injury and immune dysregulation, their pathological and clinical presentations differ. Understanding gene expression changes in these diseases could reveal new therapeutic targets. Single-cell transcriptomic datasets (GSE213030 and GSE176465) were analyzed to investigate cellular interactions in MCD and FSGS. Machine learning algorithms developed diagnostic models, and immune subtypes were identified for detailed subtype analysis. Key genes were validated using qRT-PCR and immunohistochemical staining in a mouse model, focusing on their association with M1 macrophage activation. Integrated single-cell analysis identified six key genes (PTPRC, ACTR2, MYO1F, UBB, CSF1R, and LYN) central to macrophage activation. These genes were closely linked to M1 macrophage activation, as confirmed through transcriptomic profiling and spatial co-expression patterns in Sprague–Dawley (SD) rat models. Machine learning models validated their predictive value in disease progression from MCD to FSGS. This study highlights six hub genes as potential biomarkers for predicting MCD-to-FSGS progression. Their roles in macrophage activation suggest these genes may serve as novel therapeutic targets for personalized treatment strategies, particularly for patients at high risk of disease transition.

Graphical Abstract

Overview of the integrative analysis and validation workflow. Part of the picture is taken from GDP by Figdraw.

1. Introduction

Minimal change disease (MCD) and focal segmental glomerulosclerosis (FSGS) are two major pathological types of nephrotic syndrome, which hold significant clinical importance [1,2]. MCD is the most common cause of nephrotic syndrome in children, marked by mild glomerular damage that is often difficult to detect under light microscopy [3,4], and it responds well to steroid therapy. FSGS is more common in adults, characterized by focal glomerular sclerosis accompanied by notable proteinuria [5,6]. FSGS typically progresses rapidly, is resistant to steroids, prone to recurrence, and may ultimately develop into chronic kidney disease (CKD) [7].

Both diseases share common features including podocyte injury and immune dysregulation, which disrupt the glomerular filtration barrier and lead to proteinuria [8,9]. Notably, some MCD patients may progress to FSGS during disease course [10], with persistent heavy proteinuria considered a major risk factor for this transition [11]. Persistent immune system dysregulation, especially the imbalance of immune cells and cytokines, may play a key role in the transition from MCD to FSGS [12]^. Certain genetic variations, such as mutations in the APOL1 gene, are associated with FSGS pathogenesis, and these genes may also contribute to the risk of MCD progressing to FSGS [13].

Given the significant clinical overlap but differing treatment strategies and prognoses between MCD and FSGS, accurate early distinction is crucial for optimal treatment [9,14]^. With advances in bioinformatics and machine learning technologies, the exploration of inflammatory immune microenvironments and inflammation-related biomarkers has become increasingly diverse. In recent years, the application of artificial intelligence in nephrology has provided new tools for disease mechanism research and biomarker discovery [15]. On the one hand, clustering analysis and other techniques can be used to identify immune cells in the inflammatory immune microenvironment, helping us better visualize changes in this environment [16]. On the other hand, single-cell analysis and machine learning are being used to explore potential gene targets related to the disease progression from MCD to FSGS and to identify new genetic variations linked to disease progression [17,18]. In this study, we combined single-cell analysis and machine learning to explore potential genetic targets for MCD to FSGS progression and validated related genes as biomarkers through animal models.

2. Methods

2.1. Data acquisition

Gene expression datasets for MCD and FSGS were retrieved from the Gene Expression Omnibus (GEO) database (www.ncbi.nlm.nih.gov/geo). Eight standard glomerular tissue samples and 14 MCD tissue samples were selected from the MCD dataset of GSE216841 [14]. Membranous nephropathy samples in the dataset were excluded from this study. In the dataset GSE129973,20 standard glomerular tissue samples and 20 FSGS tissue samples were collected. We searched for GSE219185 as the validation set for machine learning, and the MCD tissue samples in GSE216841 and GSE129973 were combined with the information of FSGS samples as the training set for machine learning. Gene expression profiling arrays of GSE216841 and GSE129973 were based on the GPL20301 platform (Illumina), respectively. HiSeq 4000 (Homo sapiens)) and GPL17586 platforms ([HTA-2_0] Affymetrix Human Transcriptome Array 2.0 [transcript (gene) version]), respectively, and gene expression profiling arrays for GSE219185 were based on platforms GPL18573 (Illumina NextSeq 500 (Homo sapiens)). Single-cell transcriptome data were obtained from datasets GSE213030 and GSE176465.GSE213030 contains single-cell transcriptome data for 10 MCD patients. GSE176465 contains single-cell transcriptome data for 24 FSGS patients.

2.2. Data preparation

Raw gene expression data from MCD and FSGS samples were processed utilizing R software version 4.2.2 (R Core Team, University of Vienna, Vienna, Austria) along with Bioconductor Packages. Transcriptome files were normalized via Platform File Probe to ID and transformed into FPKM format files. The datasets GS213030 and GSE176465 were converted into Seurat objects using the Seurat R package version 4.3.0 (Satija Lab, New York Genome Center, New York, United States). Sample numbering was employed to mitigate batch effects. During this phase, the top 2,000 variable genes were filtered for subsequent analysis and normalized using the harmony function to correct batch effects in the matrix. To reduce dimensionality, principal component analysis (PCA) was conducted on the integrated data matrix. The Elbowplot function of Seurat was used to select the top 20 principal components for further analysis. Major cell clusters were identified with the Seurat FindClusters function, setting the resolution of the MCD-related dataset GSE213030 to 1 (res = 1). These clusters were visualized using 2D tSNE or UMAP plots. Conventional markers, as reported in previous studies, categorized each cell into recognized biological cell types. Initially, cells were grouped into six primary cell types, which were subsequently subdivided into subsets and further into subclusters to examine cell type heterogeneity. The Seurat FindAllMarkers function identified highly variable genes within each cell type. For the FSGS-associated dataset GSE176465, the resolution was set to 1.5 (res = 1.5), and it was visualized using tSNE or UMAP plots. Cells were also grouped into six primary cell types, which were further divided into subsets and subclusters to explore cell type heterogeneity. Highly variable genes in each cell type were identified using the Seurat FindAllMarkers function.

2.3. Different expression genes (DEGS)

We analyzed the MCD data according to the differences between normal glomeruli and MCD using the ‘limma package’ with the screening criteria of logFC > 0.5 and p < 0.05, and obtained and visualized the differential genes. A series of differential genes associated with MCD disease were obtained. At the same time, we analyzed the FSGS data according to the differences between normal glomeruli and FSGS. We used the ‘limma package’, the screening conditions were logFC > 0.5, p < 0.05, and the differential genes were obtained and visualized. A series of differential genes associated with FSGS disease were obtained.

2.4. Cellchat

The ‘cellchat’ package was used to explore the cellular interactions among the clusters.

2.5. GSEA enrichment

We used the R package ‘GSEAbase’ version 1.58.0 (Bioconductor, Fred Hutchinson Cancer Research Center, Seattle, United States) to perform GSEA enrichment analysis on the DEGs related to the two diseases, utilizing the gene sets from various subpopulations in the single-cell transcriptome.

2.6. Immunoinfiltration analysis

The immunological microenvironments of MCD (GSE216841) and FSGS (GSE129973) were analyzed using the CIBERSORT algorithm with the ‘IOBR’ package version 0.99.9 (Southern Medical University, Guangzhou, China). p Values from 0 to 0.001 were marked as ‘***’, those from 0.001 to 0.01 were marked as ‘**’, p values between 0.01 and 0.05 were marked as ‘*’, and p values above 0.05 were marked as ‘NS’.

2.7. Venn and PPI

The genes in the Macrophages cluster of MCD and the Macrophages cluster of FSGS were extracted and intersected by the Venn diagram. Subsequently, the ‘STRINGdb’ package version 2.8.4 (Swiss Institute of Bioinformatics, Zurich, Switzerland) was employed to build the protein–protein interaction (PPI) network by setting the confidence threshold to 400 and using the ‘between’ value. The PPI network was constructed, and the shared genes in the network were screened by Betweenness Centrality (BC). The genes exhibiting the highest number of interactions were identified as hub genes, potentially playing a pivotal role in the transition from MCD to FSGS.

2.8. Enrichment analysis

R packages ‘clusterProfiler’ version 4.6.2 (Southern Medical University, Guangzhou, China), ‘org.Hs.eg.db’ version 3.15.0 (Bioconductor, Boston, United States), and ‘DOSE’ version 3.24.2 (Southern Medical University, Guangzhou, China) were used to analyze for DO, GO, and KEGG enrichment, respectively, and the results with p less than 0.05 are presented as histograms.

2.9. Multi-machine learning

Using the R package ‘mlr3verse’, MCD and FSGS data from datasets GSE216841 and GSE129973 were merged and a machine learning diagnostic model was built with matching clinical data, and a validation model was built using GSE219185 and matching clinical data. Multiple machine learning models were used, and the most efficient model was used to predict the critical role of hub genes in disease diagnosis.

2.10. NMF of hub genes in macrophages

In order to investigate how hub genes affect macrophages in the inflammatory immune microenvironment, we conducted a dimensionality reduction analysis to specifically examine the expression patterns of hub genes in both diseases. Using the scRNA expression matrix as our starting point, we applied additional screening steps and then performed dimensionality reduction clustering. Finally, we use non-negative matrix factorization (NMF) algorithm to identify different macrophage subtypes.

2.11. NMF hub gene-related subtype SCENIC analysis

We used the aertslab/SCENIC package on GitHub to study the gene regulatory network of transcription factor (tf) in FSGS. We used two gene motif ranks, specifically hg19-tss-center-10 kb and hg19–500 bp upstream from the RcisTarget database. These Ranking allows us to identify transcription start sites and Explore gene regulatory networks in scRNA-seq data The operating system. For subsequent analysis, tf with adjusted p value was used less than 0.05 corrected by the Benjamini–Hochberg method was selected for further investigation.

2.12. M1/M2 macrophage polarization score

We used the ‘scMetabolism’ package and AddModuleScore function of ‘Seurat’ package to score the macrophage subtypes using the M1/M2 macrophage-related gene set, and the scoring results were displayed using violin diagram and heat map.

2.13. Rats modeling

To investigate the role of the pivotal genes screened by machine learning in the disease, this study purchased 20 4–5-week-old male juvenile Sprague–Dawley (SD) rats weighing 100–140 grams from Liaoning Changsheng Biotechnology Co., Ltd. for the experiment. All rats were acclimated to the facility at the First Affiliated Hospital of Jinzhou Medical University for 7 d. The environmental conditions were as follows: temperature 22 ± 2 °C, relative humidity 50 ± 5%, and 12-h day-night light cycle. Each polycarbonate cage housed 3–4 rats. The iron-waste corn husk bedding was replaced every 2 d and sterilized under high pressure. The rats had free access to standard feed and sterile-filtered drinking water. Using the random number table method in Excel, 20 rats were divided into 3 groups (FSGS group, MCD group, and normal control group). The grouping was carried out by personnel who did not participate in the subsequent experiments. The rats were anesthetized by intraperitoneal injection of 1% pentobarbital sodium (50 mg/kg body weight; Sigma-Aldrich, St. Louis, MO), and the depth of anesthesia was assessed by the absence of paw withdrawal reflex. To establish a model of FSGS, we first performed a right nephrectomy, followed by an injection of Adriamycin administered through the tail vein at a dosage of 4 mg/kg and 3 mg/kg in the second and sixth postoperative weeks, respectively, and assessed the renal pathological changes up to the 70th day. This modified two-step low-dose protocol, adapted from previously published models, aimed to induce a progressive and reproducible FSGS phenotype while minimizing acute mortality associated with single high-dose injection models [19,20]. The renal tissue pathological scoring and immunohistochemical results analysis were independently conducted by 2 pathologists who were unaware of the grouping information. The assessment was carried out in a blinded manner. Diagnostic criteria of FSGS were light microscopic observation of segmental sclerosis of glomeruli, and the sclerosis might be seen in the portal portion of capillary collaterals or periphery, as well as at the corticomedullary junction, accompanied by localized tubular atrophy or interstitial fibrosis. On the other hand, a model of MCD was established by a single tail vein injection of 100 mg/kg of puromycin aminonucleoside, and samples were taken ten days later for electron microscopic observation. The primary lesions observed in this model included fusion of podocyte peduncles, granular-like or vacuolar degeneration of the renal tubules, tubular dilatation, and severe vacuolar degeneration, often accompanied by severe interstitial edema. These data were analyzed by standard statistical methods, and the results showed that both models were established by the experimental expectations, providing a valid model for subsequent mechanistic studies. After the experimental endpoint, rats were euthanized by an overdose of pentobarbital sodium (150 mg/kg body weight, intraperitoneal injection), followed by cervical dislocation to confirm death. This procedure is consistent with the 2020 AVMA Guidelines for the Euthanasia of Animals. Has followed the ARRIVE guidelines.

2.14. PCR

Total RNA was extracted from cultured cells using Trizol reagent (Beyotime, Shanghai, China), followed by cDNA synthesis using NovoScript^® Plus 1st Strand cDNA Synthesis SuperMix (Novoprotein Scientific Inc., Shanghai, China). Subsequently, qRT-PCR was performed with SYBR High-Sensitivity qPCR SuperMix (Novoprotein Scientific Inc., Shanghai, China), and the transcriptional levels were normalized to the internal control gene, GAPDH and β-actin. The primer sequences used were as follows:

GAPDH: F-CCTTCCGTGTTCCTACCC R-CAACCTGGTCCTCAGTGTAG

β-actin: F-GGCTGTATTCCCCTCCATCG R-CCAGTTGGTAACAATGCCATGT

ACTR2: F-GTGATGAGGCAAGTGAGC R-TGGGAGGTTCTGTAAGTAAA

CSF1R: F-GTGGCTGTGAAGATGCTAA R-GCTCCCAAGAGGTTGACTA

LYN: F-GGTGCGAAGTTCCCTATC R-TCATCACATCTGCGTTGG

MYO1F: F-ACTGGCAGAGTCACAACG R-ATTTCTCGGTCAGTGAAGTAG

PTPRC: F-TGGTCCTCCTTATGAAAC R-TAACTGAATCTCCCTCGT

UBB: F-GAGCCCAGTGACACCATC R-GAGTGCGGCCATCTTCTA

2.15. Experimental techniques

Standard histological and molecular techniques were employed including hematoxylin-eosin (HE) staining, periodic acid-Schiff (PAS) staining, transmission electron microscopy (TEM), immunohistochemistry (IHC), and multiplex immunofluorescence (mIF) staining. Detailed protocols are provided in Supplementary Methods.

3. Result

3.1. Single-cell landscape

We identified podocytes (PODXL+), immune cells (PTPRC+), principal cells (AQP2+), Loops of Henle (SLC12A1+), mesangial cells (FHL2+), tubular cells (LRP2+) in the single-cell data of the MCD (Figure 1(A)). We confirmed the accuracy of the cellular subpopulation annotation using the expression degrees of the relevant marker genes (Figure 1(C)). In the single-cell data from FSGS, we identified Mesangial cells (FHL2+), myeloid cells (CD68+), Tubular cells (SPP1+), Myofibroblast cells (MYL9+), Principal cells (AQP2+), and T cells (CD3D+) (Figure 1(B). We also confirmed the accuracy of the cellular subpopulation annotation (Figure 1(D)).

Figure 1.

The UMAP plots to identify each cell type in Minimal Change Disease (A). The tSNE plots to identify each cell type in focal segmental glomerulosclerosis (FSGS) (B). Characteristic Markers of Each Cell Subgroup in Minimal Change Disease (C). Characteristic Markers of Each Cell Subgroup in focal segmental glomerulosclerosis (FSGS) (D).

3.2. Differential gene analysis of FSGS was concentrated in myeloid cells

Differential expression analysis highlighted significant transcriptional differences between MCD and normal kidney tissues, identifying key up-and down-regulated genes for MCD pathology (Figure 2(A)). The annotated subpopulation-associated hypervariable genes in the single-cell transcriptome of the MCD group were extracted to form a subpopulation-associated gene set. By gene set enrichment analysis (GSEA), we found that differential genes in the BULK transcriptome were centrally enriched on the podocyte cell clusters of the single-cell subpopulation (Figure 2(B)). Differential transcriptomics between FSGS and normal kidney tissues similarly showed significant variability, from which we obtained key up-and down-regulated genes in FSGS pathology (Figure 2(C)). By also extracting the relevant subpopulation gene sets for differential gene GSEA enrichment from cellular subpopulations after FSGS single-cell transcriptome annotation, we found that FSGS differential genes were mainly enriched in myeloid lineage, myofibroblasts, and T cells (Figure 2(D–F)). These enrichment results coincide with our pathologic knowledge of the disease at this stage, in which the pathological manifestations of MCD are mainly reflected in podocytes cytopathic lesions. In contrast, the pathological manifestations of FSGS are primarily focused on fibrosis of renal tubules. With this, we are more convinced of the accuracy of the single-cell subpopulation annotation.

Figure 2.

The DEGs in GSE216841 are shown in a volcano plot (A). The gene set enrichment analysis (GSEA) suggested that GSE216841 showed enrichment for podocytes (B). The DEGs in GSE129973 are shown in a volcano plot (C). The gene set enrichment analysis (GSEA) suggested that GSE129973 showed enrichment for Myeloid, Myofibroblast, T cells (D–F). The CellChat diagram indicated that immune cells exert an influence on mesangial cells and podocytes (G). GSVA enrichment analysis demonstrates the enrichment results of the genome for each cell subgroup of MCD (H).

3.3. Macrophage immune dysfunction was observed in both MCD and FSGS

Cellular communication analysis revealed complex interactions between different cell types in MCD, in which immune cells had a significant effect on the mesangial cells and podocytes cells, suggesting that immune-mediated mechanisms play a role in MCD pathology (Figure 2(G)). And then signature pathway analysis identified several critical pathways enriched in MCD, we focused on the enrichment of the gene concentration of the immune cell-cell subpopulation, this cell subpopulation was enriched in differential genes for xenobiotic metabolism, reactive oxygen species (ROS) response, adipose formation, fatty acid metabolism, oxidative phosphorylation, gamma-interferon response, allograft rejection pathway, IL-6-JACK-STAT pathway, and other biological functions, reflecting a solid immune cell and immune factor-mediated effect (Figure 2(H)). Meanwhile, cell communication analysis of MCD subpopulations revealed receptor-ligand interactions specific to each cell cluster, from which we could find that APP-CD74 receptor-ligand generates strong interactions, acting between principal cells and mesangial cells (Figure 3(A)). In turn, CD74 is a macrophage-associated cellular marker, so we hypothesized that the changes in the inflammatory internal environment of MCD might be related to macrophage dysfunction. Meanwhile, in the cell communication analysis of FSGS subpopulations, we found that myeloid cells had the most pronounced effect on renal tubular cells, which was probably related to the development of fibrosis (Figure 3(B)), and in the communication analysis of myofibroblasts we found similar conclusions, myofibroblasts and myeloid cells had the most robust liaison with the renal tubular cells (Figure 3(C)). At the same time, T cells also produced some connectivity (Figure 3(D)). In the gene set variation analysis (GSVA) of different cell clusters, we found that the myeloid clusters reflected the activation of the homograft response, angiogenesis, TNFA signaling pathway activation, estrogen response, TGF-beta signaling pathway activation, p53 pathway activation, apoptosis, APICAL-SURFACE, and hypoxic function (Figure 3(E)), and similar enrichment results were obtained in the T-cell clusters. the myofibroblast cluster was mainly enriched for activation of the MYC pathway with epithelial-mesenchymal transition, whereas in the renal tubular cluster, the pathways of MYC pathway activation, epithelial-mesenchymal transition, reactive oxides, unfolded protein response, fatty acid metabolism, and lipogenesis were enriched, which also validate the pathologic explorations of FSGS at this stage. And the analysis of receptor–ligand interactions of FSGS-specific cell subpopulations revealed inflammatory factor interactions in the FSGS internal environment, such as myofibroblasts interacting with myeloid cells via MIF-(CD74 + CXCR4), MIF-(CD74 + CD44), and APP-CD74 receptor–ligand interactions (Figure 3(F)). We explored the immune cell infiltration in MCD versus normal renal tissues using the CIBERSORT algorithm using the BULK transcriptome data and found that there was a significant difference between the two, with a significant increase in the infiltration of M1 macrophages, and memory B-cells in MCD, suggesting that immune microenvironment dysregulation is involved in the disease progression (Figure 4(A)). Then we took the same approach to calculate the immune cell infiltration in MCD versus FSGS, and the concluding question suggested that there were more M0 macrophage, M1 macrophage, M2 macrophage fine, CD4+ T cell, and CD8+ T cell infiltration in FSGS than in MCD. This suggests that macrophage activation tends to increase stepwise in the development of normal tissue from MCD to FSGS (Figure 4(B)).

Figure 3.

The CellChat diagram illustrates the receptor-ligand interactions for each cell subgroup of MCD (A). The CellChat diagram demonstrates the interactions between myeloid cells, myofibroblasts, and T cells with other cell subgroups (B–D). GSVA enrichment analysis demonstrates the enrichment results of the genome for each cell subgroup of FSGS (E). The CellChat diagram illustrates the receptor-ligand interactions for each cell subgroup of FSGS (F).

Figure 4.

Immune infiltration analysis reveals the differences between normal tissues and minimal change disease (MCD) (A). Immune infiltration analysis reveals the differences between normal minimal change disease (MCD) and focal segmental glomerulosclerosis (FSGS) (B). The tSNE plots to identify each Myeloid cell type in minimal change disease (MCD) (C). The tSNE plots to identify each Myeloid cell type in focal segmental glomerulosclerosis (FSGS) (D).

3.4. Macrophages subpopulation landscape

To further explore the role of macrophages in the inflammatory microenvironment, we extracted the Immune cells cluster from the MCD single cell cluster. They were subdivided again by the identified immune cell markers, and Macrophages (LYZ+), Plasma cells (IGHG1+), T cells (CD3E+), and mast cells (KIT+) were obtained (Figure 4(C)). Meanwhile, the myeloid cell clusters were further subdivided in the single-cell RNA sequencing subpopulation of FSGS samples and the Macrophages clusters (CD74+) and monocytes (S100A9+) clusters were obtained (Figure 4(D)). We extracted the highly variable genes in the macrophage subpopulations of both MCD and FSGS into two gene sets for subsequent analysis.

3.5. Macrophage activation results in cell structure disorder

Venn diagram analysis showed that there were 29 overlapping genes between macrophage differential genes of MCD and FSGS (Figure 5(A)). We performed DO enrichment analysis on these overlapping genes and found that these genes were highly associated with glomerulonephritis and nephritis (Figure 5(B)). And then we further highlighted the functions enriched in the overlapping genes by GO enrichment analysis, in which we found that these genes affected biological processes, such as neuroinflammatory response, cation transport across membranes, and establishment of organelle localization, and were associated with cellular components, such as cell-substrate junction, focal adhesion, membrane microdomain, and membrane rafts, and were associated with the extracellular skeleton, proteoglycan binding, cytokine binding, actin binding, myofilament protein binding, and other molecular functions (Figure 5(C)). KEGG pathway enrichment analysis of overlapping genes identified functions involved in MCD and FSGS as Fc gamma R-mediated phagocytosis, Viral protein interaction with cytokine and cytokine receptor, Tight junction, Platelet activation, Chemokine signaling pathway, Rap 1 signaling pathway, cytokine–cytokine receptor interaction, Leukocyte transendothelial migration, Phagosome, Cell adhesion molecules, and other functions were related (Figure 5(D)). These functional enrichment results suggest that macrophage activation may disrupt the corresponding cellular structures and biological functions through the above-involved pathways, thus leading to the progression of MCD to FSGS.

Figure 5.

The Venn plot shows the intersected 29 genes (A). DO enrichment analysis of overlap genes (B). GO enrichment analysis of overlap genes (C). KEGG enrichment analysis of overlap genes (D). Hub genes are shown in the protein–protein interaction (PPI) plot (E). Expression differences of hub genes between MCD and FSGS (F).

3.6. Hub gene promoting macrophages activation

From the overlapping genes we constructed a PPI network revealing a complex interaction network, and identified the critical hub genes as CSF1R, MYO1F, UBB, PTPRC, ACTR2, and LYN (Figure 5(E)). And then we performed differential expression analysis of these hub genes, and the results showed that the remaining genes except UBB had significant expression differences between MCD and FSGS, and all were highly expressed in FSGS (Figure 5(F)). These genes may be the key genes that guide macrophage activation and promote the development of MCD to FSGS. We then evaluated the interactions between these essential genes in the joint kidney tissue versus MCD group and joint kidney tissue versus FSGS group, respectively. The analysis revealed a complex network of interactions, and the results showed that the remaining genes, except UBB, demonstrated coordinated rent with each other in the development of the two diseases, especially the three genes, PTPRC with LYN and CSF1R, interacted with the rest of the genes more significantly (Figure 6(A,B)). We then explored the role of critical genes for MCD and FSGS groups, i.e., the immune microenvironment during disease transformation. The results showed that ACTR2 showed positive correlation with naive B cells and dendritic cells, and CSF1R showed positive correlation with M1 macrophages, plasma cells, CD4 memory cell activation, and CD8 cells, and negative correlation with CD4 memory cell dormancy. LYN showed a positive correlation with monocytes and CD4 memory cell activation, MYO1F showed a positive correlation with plasma cells, CD4 memory cell activation, and CD8T cells, and a negative correlation with CD4 memory cell dormancy, PTPRC showed a positive correlation with dendritic cell dormancy, M1 macrophage activation, CD8T cells, CD4 memory cell activation, and plasma cells, and PTPRC showed positive correlation with M1 macrophage activation, CD8T cells, CD4 memory cell activation, CD8T cells, CD8T cells, CD4 memory cell activation, and CD8T cells: Cell activation, positive correlation for plasma cells, and negative correlation for CD4 memory cell dormancy. UBB did not exhibit any immune cell correlation (Figure 6(C)).

Figure 6.

Interactions of hub genes in MCD (A). Interactions of hub genes in FSGS (B). The role of hub genes in the activation of immune cells during the transition from MCD to FSGS (C). Machine learn model comparison of the six hub genes (D). Receiver operating characteristic (ROC) curve of the multiple machine learning models (E). ROC curve of the naive Bayes model in the external validation set (F).

3.7. Hub gene showed certain predictive value in disease transformation

We used multiple machine learning to model the essential genes and obtained models for modeling, among which we selected the naive Bayes model with the best benefit (AUC 0.71) for subsequent validation (Figure 6(D)). The analysis of subject job characteristics (ROC) curves of the machine learning model showed certain accuracy and reliability of the classical Bayesian model, and the confidence intervals showed predictive solid power (Figure 6(E)). Validation of the classic Bayesian machine learning model on independent datasets confirmed its reliability, with a predictive performance as high as 0.936, indicating the model’s robustness in predicting disease states (Figure 6(F)). Multiple machine learning modeling re-emphasized the importance of core genes in the pathogenesis of MCD and FSGS. It demonstrated the potential of machine learning models to accurately classify and predict disease outcomes based on genetic and transcriptomic data.

3.8. Hub gene subtype macrophages in FSGS showed higher immune activation

We used the NMF algorithm to re-cluster macrophages in the two diseases with dimensionality reduction (Figure 7(A,F)), and reclassified the macrophages in the two diseases into hub gene characteristic subtypes based on gene expression levels in different clusters (Figure 7(B,G)). Studies indicated that Hub gene-infiltrating macrophages showed a higher cellular communication effect on myeloid cells and T cells than MCD in FSGS (Figure 7(C,H,I)). Moreover, Hub gene infiltration + Macrophages group showed a higher cellular communication effect than other subtypes (Figure 7(J)). Subtypes of macrophages in MCD did not show a tendency toward M1 or M2 polarization (Figure 7(D,E)). However, subtype macrophages in FSGS showed a higher M1 macrophage polarization tendency (Figure 8(A,C)). At the same time, we found that USF2 (11 g), SPI1 (32 g), and MEF2C_extended (16 g) tfs were activated in the Hub gene infiltration + Macrophages subtype (Figure 8(B)). All metabolic activities in this group were enhanced except Nitrogen metabolism and Linoleic acid metabolism (Figure 8(D)).

Figure 7.

NMF of macrophages subtype in MCD (A). Heatmap of hub gene in NMF cluster in MCD (B). Cellchat of macrophages subtype in MCD. (C). Violin plot of M1/M2 gene set expression of macrophages subtype in MCD (D). Expression of M1/M2 gene set of macrophages subtype in MCD (E). NMF of macrophages subtype in FSGS (F). Heatmap of hub gene in NMF cluster in FSGS (G). Cellchat of macrophages subtype in FSGS (H–I). Signaling pattern plot of macrophages subtype in FSGS (J).

Figure 8.

Violin plot of M1/M2 gene set expression of macrophages subtype in FSGS (A). Heat maps of macrophage subtype associated TF in FSGS (B). Expression of M1/M2 gene set of macrophages subtype in FSGS (C). Metabolic pathway about macrophages subtype in FSGS (D).

3.9. The hub gene has a higher M1 macrophage activation rate in FSGS than in MCD

We successfully constructed mouse models of MCD and FSGS, and isolated kidney tissues of corresponding mouse models for subsequent experiments (Figure S1(A,B)). qRT-PCR analysis of ACTR2, CSF1R, LYN, MYO1F, PTPRC, and UBB genes showed significant differences between the other groups, in line with the trend of our biosynthesis analysis, except that there was no difference between the normal group, MCD group, and FSGS group in UBB gene and no difference between the MCD group and FSGS group in LYN (Figures 2 and 9(A)). Immunofluorescence detection showed that the fluorescence intensity of CD86, CTR2, CSF1R, MYO1F, PTPRC, and LYN genes was significantly different in the normal group, MCD group, and FSGS group, indicating that the expression levels of these genes were different in the normal group, MCD group, and FSGS group. In addition, the molecular marker CD86 of M1 macrophages had co-spatial expression and proportional expression relationship with our hub gene in normal group, MCD group, and FSGS group (Figure 9(B,C)).

Figure 9.

qRT-PCR results of hub genes (A). Multiple immunofluorescence intensity statistics (B). Immunofluorescence assay results in normal, MCD, and FSGS tissues (C).

4. Discussion

In this study, we utilized a combined approach of single-cell transcriptomics and bulk transcriptomics to investigate the connection between M1 macrophage activation and the key genes involved in the disease progression from MCD to FSGS. Through this integrative analysis, we identified a novel perspective, allowing us to observe the pathological characteristics of MCD and FSGS more intuitively.

Podocyte injury is a critical factor contributing to proteinuria in MCD [21]. Using single-cell mapping and enrichment analysis of differentially expressed genes from transcriptomics data, we precisely identified the cells undergoing injury and performed immune infiltration analysis, obtaining results consistent with current research findings [22]. These include significant activation of helper T cells and memory T cells in MCD. Additionally, we observed a considerable increase in M1 macrophage activation in MCD. This may be due to M1 macrophages engaging in antigen presentation and T-cell activation, forming a positive feedback loop that amplifies inflammation, leading to upregulation of podocyte CD80 expression and exacerbation of foot process effacement [10,23].

Pathologically, FSGS is characterized by segmental glomerular sclerosis, which is associated with podocyte loss and abnormal extracellular matrix accumulation [24]. Through single-cell mapping and integrative transcriptomic analysis, we found higher activation of helper T cells and M1 macrophages in FSGS compared to MCD, consistent with conclusions from experimental research [25]. The activation of these immune cells likely plays a critical role in the progression from MCD to FSGS [26,27]. Some studies suggest that macrophages play a significant role in this transition [28–30]. However, specific mechanisms and molecular targets through which macrophages contribute to disease progression remain underexplored. Therefore, we utilized bioinformatics tools to investigate the role of macrophages in disease progression.

Based on insights gained through a bioinformatics perspective, we extracted differentially expressed genes from macrophage clusters in both diseases and identified overlapping genes using Venn diagram analysis. Through PPI network analysis, we identified hub genes associated with both diseases to study whether changes in these genes within macrophages influence disease progression. This approach is common in cancer research but rarely applied to inflammation studies. Using this method, we identified six hub genes: PTPRC, ACTR2, MYO1F, UBB, CSF1R, and LYN.

Protein Tyrosine Phosphatase Receptor Type C (PTPRC), also known as CD45, is a transmembrane protein tyrosine phosphatase primarily expressed in hematopoietic cells. PTPRC enhances the ability of macrophages to recognize and respond to pathogens by regulating the JAK/STAT and NF-κB signaling pathways [31]. Its role in MCD and FSGS remains unknown.

Actin-Related Protein 2 (ACTR2) is a key component of the ARP2/3 protein complex, primarily involved in cytoskeletal remodeling. These studies have shown that ACTR2 regulates the branching of actin filaments, impacting cell migration, adhesion, and phagocytosis [32]. Its expression is associated with the Yes-associated protein (YAP) signaling pathway, influencing M1 and M2 macrophage polarization [32,33]. However, studies linking ACTR2 to MCD and FSGS remain limited.

Myosin 1 F (MYO1F), a non-muscle class I myosin, is widely expressed in immune cells and plays a critical role in regulating immune cell polarization and pro-inflammatory responses, thanks to its unique ATPase activity and actin-binding capabilities [34]. MYO1F promotes the secretion of pro-inflammatory cytokines, such as IL-1β, TNF-α, and IL-6 by M1 macrophages through activation of the NF-κB signaling pathway, while its deficiency inhibits M1 macrophage polarization [35]. In inflammatory microenvironments, MYO1F amplifies pro-inflammatory signaling by interacting with cytoskeletal proteins and signaling molecules, leading to sustained activation of M1 macrophages and contributing to the pathological progression of chronic inflammatory diseases [36]. But, its relevance to MCD and FSGS requires further exploration.

UBB (Ubiquitin B) encodes ubiquitin, a key component of the ubiquitin-proteasome system. UBB plays an essential role in maintaining cellular homeostasis, immune regulation, and responses to oxidative stress [37]. By modulating NF-κB pathway activity, UBB promotes the secretion of pro-inflammatory cytokines (e.g. TNF-α, IL-6, and IL-1β) by M1 macrophages, enhancing inflammation [37,38]. Its role in MCD and FSGS remains unclear. Although not differentially expressed at the mRNA level in our analysis, UBB may exert influence through post-translational modifications affecting protein stability of other hub genes or key inflammatory mediators, representing a potential regulatory layer in the macrophage activation circuit.

Colony-Stimulating Factor 1 Receptor (CSF1R) is a transmembrane tyrosine kinase receptor primarily expressed in macrophages and their progenitors. CSF1R plays a key role in immune homeostasis and the regulation of inflammatory diseases [39]. Its signaling enhances the secretion of pro-inflammatory cytokines (e.g. TNF-α, IL-6, and IL-1β) by M1 macrophages via NF-κB activation [40] and promotes macrophage migration and localization to inflammatory sites by regulating the cytoskeleton [41]. In chronic inflammation, excessive CSF1R signaling leads to sustained M1 macrophage activation, exacerbating tissue damage [42]. However, its role in inflammation progression remains unclear.

Lyn Tyrosine Kinase (LYN) is a Src family non-receptor tyrosine kinase that regulates immune cell functions by phosphorylating substrate proteins [43]. LYN interacts with TLR2 and TLR4 to activate the NF-κB pathway, promoting the secretion of pro-inflammatory cytokines (e.g. TNF-α, IL-6, and IL-1β) by M1 macrophages. It also enhances ROS production, amplifying the bactericidal and pro-inflammatory abilities of M1 macrophages [44,45]. LYN’s role is well-documented in infectious diseases and inflammatory pathologies such as atherosclerosis [46], but its involvement in MCD and FSGS has not been reported.

Our subsequent analyses hypothesized potential coordination among these six M1 macrophage polarization-associated genes (as shown in Figure 6(A–C)). These six genes synergistically activate the NF-κB signaling pathway, amplifying chronic MCD-induced damage, ultimately leading to the development of FSGS. High expression of PTPRC activates MYO1F, CSF1R, and LYN. Concurrently, UBB, once activated by CSF1R, suppresses ACTR2 activity, prolonging M1 macrophage-mediated inflammatory damage. While UBB also exerts a slight inhibitory effect on MYO1F activity, this inhibition is offset by strong activation from other genes, cumulatively resulting in enhanced inflammatory activity. Based on the results of our enrichment analysis (as shown in Figure 5(B–D)), activated M1 macrophages disrupt the cytoskeleton and adhesion structures of podocytes through the ROS pathway and the activation of inflammatory cytokines, leading to substantial podocyte detachment. Additionally, M1 macrophages secrete pro-fibrotic factors, initiating and accelerating the process of glomerular fibrosis.

With the advancement of computational algorithms, numerous studies utilizing machine learning to investigate cancer have been published [47]. In the medical field, machine learning is widely applied in disease diagnosis, treatment outcome prediction, and multi-omics data analysis [48]. Similarly, we employed machine learning to further analyze the roles of these six key genes within the inflammatory microenvironment. However, single models are typically only effective for specific types of data (e.g. linear models for linear data) and often lack robustness when handling multimodal or complex data [49]. By integrating multiple machine learning methods, we combined the strengths of different models to improve prediction accuracy and robustness [50,51]. Using diverse machine learning approaches, we developed a predictive model for the transition from MCD to FSGS. Studies on predictive models for such inflammatory transitions are currently limited. Our model demonstrated a high-risk association, further confirming that the activation of these six genes represents a high-risk factor for the progression of MCD to FSGS.

Additionally, using the NMF algorithm, we constructed macrophage-related weight matrices based on the six key genes as foundational components [52,53]. This enabled us to explore the activation effects of these six genes on macrophages in both inflammatory conditions. Our analysis revealed that macrophage activation in the inflammatory microenvironment of MCD was relatively mild, with no significant M1 polarization tendency. In contrast, the immune microenvironment of FSGS showed intense M1 macrophage activation. Furthermore, clusters characterized by infiltration of the key genes displayed strong activation of pathways related to MIF, SPP1, Galectin, CCL, and the complement system. These pathways are closely related to the polarization and activation of macrophages.

Through the NMF algorithm, we also discovered that the metabolic functionality of the hub gene-macrophage cluster was enhanced across various pathways, except for those involving linoleic acid and nitrogen metabolism. Transcription factor prediction analysis revealed strong activation of the SPI1 tf in the hub gene-macrophage cluster. SPI1, a member of the ETS tf family, is critical in hematopoiesis, particularly in the differentiation and development of myeloid and lymphoid lineages. SPI1 predominantly functions in immune cells. By integrating multiple computational algorithms, we conducted a more detailed analysis of the six key genes, uncovering additional insights into potential mechanisms underlying disease progression.

In our experiments, we selected the mIF technique [54], which allows for the precise identification of the spatial relationship between target genes and target cells. In other studies, involving models, such as cardiovascular injury, mIF has been used to evaluate the dynamic distribution of macrophages during tissue repair, revealing how the balance between M1 and M2 macrophages influences the regeneration process [55]. Therefore, we hypothesized that this technique could also be applied to our disease progression model.

After establishing animal models of MCD and FSGS, we initially used PCR to test the expression of the M1 macrophage marker CD86 and the six hub genes in both diseases. Subsequently, we applied the mIF technique to further investigate the co-expression relationship between the hub genes and CD86.

Our data collectively suggest the existence of a ‘pre-FSGS inflammatory tipping point’ within the renal microenvironment. In this model, chronic but subcritical inflammation in MCD, characterized by mild M1 macrophage activation, is contained. However, sustained signaling through the macrophage activation circuit formed by PTPRC, ACTR2, MYO1F, CSF1R, and LYN (potentially modulated by UBB) may amplify NF-κB signaling beyond a threshold. This shift pushes the system into a self-sustaining pro-fibrotic state, marking the transition to FSGS. This circuit-based framework moves beyond viewing individual genes as independent markers and instead emphasizes their synergistic role in driving a pathological state transition.

Despite the novel findings of our study, there are still some limitations. The two-step low-dose Adriamycin-induced FSGS model used in this study mainly simulates the characteristics of glomerulosclerosis and inflammatory microenvironment, rather than the natural progression from MCD to FSGS. It may not fully recapitulate the pathophysiological mechanisms of clinical disease transformation. Future studies will adopt a sequential model of FSGS induced by MCD recurrence to further validate the dynamic role of hub genes in disease progression. In addition, due to the complexity of inflammation-related experiments and financial constraints, we were only able to perform basic experiments to demonstrate the correlation between genes and disease without thoroughly validating the specific underlying mechanisms. As a result, the progression mechanisms of the disease were only speculated based on bioinformatics analyses. However, we plan to address this limitation in future studies by building upon this research as a foundation for more in-depth investigations.

Building on these findings, future research should transition from retrospective biomarker identification to prospective validation. Priority steps include: (1) Establishing longitudinal cohorts of MCD patients with serial biopsies to validate the predictive value of the six-gene signature and macrophage activation state for FSGS progression. (2) Employing spatial transcriptomics on human tissue to confirm the co-localization and cellular interaction of the hub gene circuit within the glomerular niche. (3) Developing genetically engineered or inducible animal models that more faithfully mimic the human progression from MCD to FSGS, to experimentally test the causality of the proposed macrophage circuit. (4) Integrating our molecular signature with clinical parameters and AI-enabled phenotyping to build refined, personalized risk prediction models. Ultimately, these efforts aim to translate the identified circuit into actionable biology, informing the development of targeted interventions to intercept the progression at the ‘pre-FSGS tipping point’.

In summary, this study integrated single-cell transcriptomics, bulk transcriptomics, various machine learning methods, and the NMF algorithm to identify six hub genes. These genes may interact to activate the NF-κB signaling pathway, driving M1 macrophage polarization, which amplifies chronic MCD-related damage and plays a crucial role in the progression to FSGS. The research established a high-risk gene signature model for chronic MCD. By implementing precision treatment strategies for these high-risk MCD patients, further deterioration into FSGS can potentially be prevented. Additionally, the findings provide a bioinformatics-based molecular mechanism that may explain the progression from MCD to FSGS.

Animal Ethics Committee

This study was performed in line with the principles of the Declaration of Helsinki. Approval was granted by the Ethics Committee of Jinzhou Medical University (Project number: JYTFUDF201759).

Funding Statement

This work was supported by the Natural Science Foundation of Liaoning Province (Grant Number 2022-MS-383 to Xiang-fei Cui).

Disclosure statement

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Data availability statement

The data that support the findings of this study are openly available in figshare at http://doi.org/10.6084/m9.figshare.30521045, reference number 30521045.