In the present study, two DKD datasets from the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/) (35) were screened: GSE30529 (GPL571) containing 10 cases of DKD and 12 controls; and GSE30122 (GPL571) as a validation dataset containing 19 cases of DKD and 50 controls (Table SI) (36,37). Serial matrix files of both datasets, containing gene expression data, were downloaded for subsequent analysis. The boxplot function from the 'graphics' package in R (version 3.6.2; https://rdocumentation.org/packages/graphics/versions/3.6.2) was used to visualize and evaluate the distribution of gene expression across samples. Differential genes were analyzed using the 'limma' R package (version 3.64.1; https://bioinf.wehi.edu.au/limma/) (38), with thresholds set at log2 (fold change) >0.58 or log2 (fold change) <-0.58, and P<0.05 for identifying differentially expressed genes (DEGs). Heatmaps and volcano plots were subsequently generated employing the 'pheatmap' (version 1.0.12; https://cran.r-project.org/web/packages/pheatmap/) and 'ggplot2' (version 3.3.5; https://cran.r-project.org/web/packages/ggplot2/) R packages, respectively, to display the DEGs.

The 'WGCNA' R package (version 1.73; https://cran.r-project.org/web/packages/WGCNA/) was employed to identify gene modules associated with DKD phenotypes in the GSE30529 dataset. A soft-thresholding power was selected to achieve a scale-free network topology (R²=0.85). Subsequently, hierarchical clustering analysis grouped genes into distinct modules, with a minimum size of 30 genes per module. To identify biologically relevant modules, the 'WGCNA' R package was used to evaluate the associations between module eigengenes and clinical traits. Genes within these trait-associated modules were defined as key drivers if they met the criteria of module membership (MM) >0.8 and gene significance (GS) >0.2. Venn diagram analysis was used to examine the intersection between key driver genes and previously identified DEGs to yield the final target gene set. The Venn diagram was generated using the 'VennDiagram' package (version 1.7.3; https://cran.r-project.org/web/packages/VennDiagram/).

GSEA was applied to evaluate gene distribution patterns in DKD within the GSE30529 dataset and to investigate their related phenotypes (39). The 'clusterProfiler' R package (version 3.0.4; http://www.bioconductor.org/packages/clusterProfiler/) was used for functional enrichment analysis of the HALLMARK gene set modules with the reference gene set: c2.cp.all. v2022.1.Hs.symbols.gmt. Absolute net enrichment scores >1, false discovery rates q<0.25 and P<0.05 were considered to represent significant enrichment.

The 'cluster-Profiler' R package was also used to analyze GO and KEGG pathway enrichment (40,41). Gene-related molecular functions (MFs), cellular components (CCs), biological processes (BPs) and gene-related signaling pathways were assessed.

In the present study, three ML algorithms: Random forest (RF) (42), least absolute shrinkage and selection operator (LASSO) logistic regression (43), and support vector machine-recursive feature elimination (SVM-RFE) (44), were employed to identify key biomarkers for DKD. The RF approach was performed with the 'random-Forest' R package (version 4.7-12; https://cran.r-project.org/web/packages/randomForest/). SVM-RFE was performed with the R package 'e1071' (version 1.7-16; https://cran.r-project.org/web/packages/e1071/), whereas LASSO logistic regression was conducted using the 'glmnet' R package (version 4.1-10; https://www.rdocumentation.org/packages/glmnet/), selecting the optimal model according to the minimum λ value. Parameter optimization was carried out using 10-fold cross-validation, with the model reaching the lowest biased likelihood deviance criterion. Genes exhibiting common traits across the three classification models were selected for further investigation. Additionally, a PPI network was built using the STRING database (http://string-db.org/) and was subsequently visualized with Cytoscape 3.9.1 software (https://cytoscape.org/). Differential genes were assessed using three algorithms: Degree, closeness and maximum clique centrality, within the cytoHubba plugin (version 0.1; https://apps.cytoscape.org/apps/cytohubba). The top 10 genes from each method were selected and their overlaps determined. Ultimately, genes common to these intersections were designated as hub genes. UpSet plots and petal diagrams was generated using the 'ggplot2' package.

Hub gene expression was validated using the GSE30122 dataset, followed by the use of the Nephroseq v5 tool (https://www.nephroseq.org/), which was employed to analyze the correlations between serum creatinine (Scr), eGFR and identified hub genes (45). Scatter plots illustrating these correlations were re-plotted using the 'ggplot2' R package. A nomogram model for diagnosing DKD was developed using the rms package (version 8.1-0; https://hbiostat.org/R/rms/) based on hub genes. Decision curve analysis (DCA) (version 3.8-3; https://github.com/therneau/survival) demonstrated the clinical utility of the nomogram model.

The CIBERSORTx algorithm (https://cibersortx.stanford.edu/) (46) was employed to quantify the infiltration levels of various immune cell subtypes within the DKD gene expression profiles. Subsequently, based on the relative abundance of 22 types of infiltrating immune cells, pairwise Pearson correlation coefficients were calculated. The 'ggplot2' R package was utilized to visualize the correlation results. Principal component analysis (PCA) was subsequently applied to compare immune cell infiltration patterns between the DKD and control groups, with visualization performed using the 'ggplot2' R package. Furthermore, the interrelationships among the 22 immune cell subsets were illustrated in a heatmap.

A total of 12 healthy male C57BL/6J mice (age, 8 weeks; weight, 20-22 g), six male C57BLKS/J db/db mice and six male C57BLKS/J db/m mice (age, 8 weeks; weight, 20-22 g) were purchased from Nanjing Junke Biological Engineering Co., Ltd. Due to their leptin receptor deficiency, db/db mice are commonly used as a classic animal model for spontaneous type 2 diabetes, with their littermate heterozygous db/m mice serving as a control group (47). The mice were housed and experimented on in the animal laboratory at Shanghai Rat & Mouse Biotech., Ltd. (Shanghai, China). The mice were maintained a temperature of 24°C and relative humidity of 50-60%, under a 12-h light/dark cycle. The mice had free access to standard laboratory chow and clean drinking water, and underwent a 1-week acclimation period. C57BL/6J mice were randomly allocated to two groups: The sham group (n=6) was maintained on a standard diet, whereas the DKD model group (n=6) received a high-sugar, high-fat diet (67% maintenance chow, 10% lard, 20% sucrose, 2.5% cholesterol and 0.5% sodium bile acid). Under anesthesia, induced with 2% isoflurane and maintained with 1.5% isoflurane, left uninephrectomy (UNx) was performed.

After 1 week of postoperative recovery, the mice were fasted for 12 h and received a single intraperitoneal injection of streptozotocin (STZ; 40 mg/kg body weight; Sigma-Aldrich; Merck KGaA) dissolved in sodium citrate buffer (pH 4.5). Control animals received an equivalent volume of sodium citrate buffer alone. After 4 consecutive days of injection, plasma glucose levels were assessed from 10-µl tail vein blood samples using an AccuChek glucometer (Roche Diagnostics) over 3 consecutive days. A diabetic mouse model was confirmed when plasma glucose concentrations reached >16.7 mmol/l. The early DKD mouse model was established by further defining diabetic models based on the urine albumin-to-creatinine ratio (UACR). 24-h urine samples collected via metabolic cages for analysis. When the UACR increases and meets the criteria for early DKD, the model is considered successfully established. All animal procedures were approved by the Institutional Animal Care and Use Committee of Shanghai Rat & Mouse Biotech., Ltd. (approval no. LL-2024-060).

The health and body weight of the mice were monitored regularly. Mice were weighed every Wednesday, and fasting blood glucose levels were measured after an 8-12-h fast with free access to water. Urine samples were collected at the conclusion of the 16-week experimental period. At the end of the experimental period, the mice were euthanized via intravenous injection of an overdose of pentobarbital (100 mg/kg). Subsequently, 0.8-1.0 ml blood was collected from the retro-orbital sinus and kidney tissue was obtained for subsequent analysis. To reduce variability arising from environmental or individual factors, each sample was collected in triplicate.

HK-2 cells (an immortalized proximal tubule epithelial cell line from a normal adult human kidney) were obtained from The Cell Bank of Type Culture Collection of The Chinese Academy of Sciences. The cells were cultured in DMEM/F12 (cat. no. 11320033) supplemented with 10% FBS (cat. no. A5670701) and 1% penicillin/streptomycin (cat. no. 15140148) (all from Gibco; Thermo Fisher Scientific, Inc.). The cells were cultured in a 37°C, 5% CO₂ incubator and subjected to a 24-h low glucose (LG) environment at 5.5 mmol/l glucose. Cells were then divided into the following groups: i) LG group: Treated with 5.5 mmol/l glucose plus 24.5 mmol/l mannitol for 24 h to serve as an osmotic control; ii) high-glucose (HG) group: stimulated with 30 mmol/l glucose for 24 h. Mannitol (cat. no. ST2362; Beyotime Biotechnology) was used as an osmotic control to prevent hyperosmolarity.

Cells were transfected with siRNAs targeting apoptotic peptidase activating factor 1 (APAF1; cat. no. HY-RS00801), ADAM metallopeptidase domain 10 (ADAM10; cat. no. HY-RS00265) and spleen-associated tyrosine kinase (SYK; cat. no. HY-RS14075) (all from MedChemExpress). Cells were cultured in 6-well plates for transfection. When cells reached 60% confluence, they were transfected using Lipofectamine^® 3000 (Invitrogen; Thermo Fisher Scientific, Inc.) according to the manufacturer's instructions. Briefly cells were transfected with siRNAs (final concentration: 50 nM) in a 5% CO₂ incubator at 37°C. After incubation of the cells with the siRNAs for 6 h, the medium was replaced with conventional complete medium containing 10% FBS, and the cells were further cultured at 37°C under 5% CO₂. Subsequent functional assays and relevant parameter measurements were conducted 48 h post-transfection. The sense and antisense sequences of the siRNAs (including the non-targeting negative control; MedChemExpress) are shown in Table I.

For renal function testing and cytokine detection, 0.8-1.0 ml blood drawn from the retro-orbital sinus of euthanized mice was collected into an anticoagulant-free centrifuge tube, which was allowed to stand at room temperature for 30 min, before being centrifuged at 1,500 × g for 10-15 min at 4°C. The supernatant was thus separated to obtain the serum for subsequent analyses. Blood urea nitrogen (BUN) was quantified using a urea assay based on the urease method (BUN; cat. no. C013-2-1; Nanjing Jiancheng Bioengineering Institute), whereas Scr was assessed via a creatinine oxidase assay (Scr; cat. no. C011-2-1; Nanjing Jiancheng Bioengineering Institute). At the end of the 16-week study, mice were housed in metabolic cages for 24-h urine collection. The urine samples were thoroughly mixed and centrifuged at 2,000 × g for 10 min at 4°C. The resulting supernatant was carefully collected, and urinary protein levels were determined using UACR assay kit (microalbumin; cat. no. C035-2-1; Nanjing Jiancheng Bioengineering Institute, Inc.) via the turbidimetric method. All kits were used in accordance with the manufacturer's protocol.

Renal tissues from the mice were collected and fixed in 4% paraformaldehyde solution at room temperature (20-25°C) for 24-h, followed by paraffin embedding. After preparing 3-µm renal tissue sections, multiple staining techniques including hematoxylin and eosin (H&E), and Masson's trichrome staining were performed. For H&E staining, samples were stained with hematoxylin for 5-10 min, followed by eosin counterstaining of the cytoplasm for 1-3 min. The degree of tubular damage was recorded using the Paller tubular injury scoring system (48), wherein 10 randomly selected tubules per high-power field were observed: Marked tubular dilatation, flattened or swollen cells scored 1 point; brush border damage or loss scored 1 or 2 points; cast formation scored 2 points; and necrotic cells shed into the tubular lumen (without forming casts or fragments) scored 1 point. The median score was calculated for each histopathological section by selecting 10 fields of view (representing 100 tubules).

For Masson's trichrome staining, first, the dewaxed tissues were incubated overnight in Bouin's fixative (solution A) at room temperature, then processed at 65°C for 30 min. Subsequently, the nuclei were stained with a mixture of Wiegers' stain A (solution B) and Wiegers' stain B (solution C) for 1 min, followed by differentiation with 1% hydrochloric acid in ethanol. The processed tissue was then treated with phosphomolybdic acid (solution E) for 1 min. Subsequently, muscle fibers were stained red with Ponceau red (solution D) for 8 min, while collagen fibers were stained blue with aniline blue (solution F) for 8-30 sec. Differentiation with 1% glacial acetic acid enhanced contrast between the two colors. The Masson's trichrome staining kit (cat. no. G1340) was obtained from Beijing Solarbio Science & Technology Co., Ltd. Pathological images were captured using a BX51 brightfield microscope (Olympus Corporation) and were semi-quantitatively analyzed with Image-Pro Plus 6.0 software (Media Cybernetics, Inc.).

Expression levels of the three hub genes identified through the analyses were validated in both blood samples from patients with DKD, as well as in kidney tissue samples from DKD and sham mice (n=6/group). Renal tissues were flash-frozen in liquid nitrogen upon collection to preserve RNA integrity and maintained at −80°C prior to processing. For the detection of gene expression in cells, the cells were collected 48 h after siRNA transfection. Total RNA was isolated using TRIzon reagent (cat. no. CW0580S; Jiangsu CoWin Biotech Co., Ltd.) according to the manufacturer's protocol. RT was performed using Evo M-MLV RT Premix (cat. no. AG11706; Hunan Accurate Bio-Medical Technology Co., Ltd.), with all procedures strictly adhering to the manufacturer's instructions. qPCR was performed using an ABI 7500 system (Applied Biosystems; Thermo Fisher Scientific, Inc.) with the SYBR Green Premix Pro Taq HS qPCR kit (cat. no. AG11701; Accurate Biology). The following thermal cycling conditions were adhered to: 95°C pre-denaturation for 30 sec, followed by 40 cycles at 95°C for 5 sec and 60°C for 30 sec. Curve analysis was performed after amplification. Gene expression was quantified using the 2^−ΔΔCq method with GAPDH as the reference gene (49). The primer sequences of the hub genes used in RT-qPCR are shown in Table II.

Kidney tissue and HK-2 cell proteins were extracted using RIPA lysis buffer (cat. no. P0013B; Beyotime Biotechnology) and quantified using the BCA protein assay. Protein samples were denatured according to the 5X protein loading buffer protocol. The denatured proteins (30 µg/lane) were separated using sodium dodecyl sulfate-polyacrylamide gel electrophoresis on a 10% resolving gel and 5% concentrating gel. The voltage was set at 80 V for 20 min, then increased to 100 V for 40 min. After electrophoresis, the proteins were transferred to PVDF membranes, which were subsequently blocked with 5% non-fat milk for 1 h at room temperature. Membranes were then incubated overnight at 4°C with primary antibodies against SYK (1:1,000; cat. no. ab40781), APAF1 (1:1,000; cat. no. ab234436), ADAM10 (1:1,000; cat. no. ab242389), fibronectin (1:5,000; cat. no. ab45688), vimentin (1:2,500; cat. no. ab137321), Snail (1:1,000; cat. no. ab180714) and GAPDH (1:10,000; cat. no. ab181602) (all from Abcam), and β-actin (1:10,000; cat. no. 6009-1-1; Proteintech Group, Inc.). This was followed by incubation with anti-rabbit secondary antibodies (1:5,000; cat. no. 7074) and anti-mouse secondary antibodies (1:5,000; cat. no. 7076) (both from Cell Signaling Technology, Inc.) for 1 h at room temperature. Protein bands were visualized using an ultrasensitive ECL chemiluminescence kit (cat. no. P0018S; Beyotime Biotechnology) and documented with a gel imaging system. The gray values of the strips were analyzed semi-quantitatively using Image-Pro Plus 6.0 software.

The present study was approved by the Ethics Committee of Seventh People's Hospital of Shanghai University of Traditional Chinese Medicine (approval no. 2024-7th-HIRB-097) and adhered to ethical guidelines consistent with the principles of The Declaration of Helsinki. All participants provided written informed consent. Patients with DKD and healthy controls were consecutively recruited from the outpatient clinic and inpatient ward of the Department of Nephrology, Seventh People's Hospital of Shanghai University of Traditional Chinese Medicine. Between July 2025 and August 2025, 15 patients with DKD and 15 healthy controls were enrolled in the study. Peripheral venous blood samples were collected from all participants for total RNA extraction and RT-qPCR was used to detect the expression of three hub genes. The inclusion criteria were established based on the KDIGO 2022 Clinical Practice Guideline for the Management of Chronic Kidney Disease in Diabetes (50), as follows; i) UACR ≥30 mg/g; ii) eGFR <60 ml/min/1.73 m² for ≥3 months; and iii) renal biopsy demonstrating pathological changes consistent with DKD. The exclusion criteria were as follows: ⅰ) Individuals <18 years of age or >65 years of age; ⅱ) women who were pregnant or breastfeeding; ⅲ) individuals with a history of diabetic ketoacidosis or urinary tract infection within the past month; and ⅳ) patients diagnosed with primary diseases affecting other systems (including cardiac, hepatic, renal and hematopoietic systems).

The R software 'ggpubr' package (version 0.6.2; https://cran.r-project.org/web/packages/ggpubr/) was used to perform statistical analyses. Statistical comparisons between groups were performed using either the parametric Student's t-test or the nonparametric Mann-Whitney U test, depending on the data distribution. Multi-group comparisons were performed using one-way analysis of variance; when significant differences were detected, Tukey's or Bonferroni-corrected multiple comparisons were conducted. The correlation between hub gene expression and eGFR was assessed using Spearman's or Pearson's rank correlation coefficient analysis (Nephroseq v5 platform). P<0.05 was considered to indicate a statistically significance difference. To evaluate the diagnostic potential of candidate genes, ROC analysis was performed utilizing the 'pROC' package (version 1.19.0.1; https://www.rdocumentation.org/packages/pROC/) in R. Diagnostic performance was quantified by calculating the area under the ROC curve (AUC), with genes exhibiting AUC values >0.85 considered to have significant diagnostic capability. Visualization of data distributions and ROC curves was accomplished using the 'ggplot2' R package.

For the present investigation, transcriptomics data from the GSE30529 dataset were obtained from the GEO repository. The quality of data normalization was visualized using boxplot analysis (Fig. 1A). Analysis demonstrated that sample median values exhibited consistent horizontal alignment, validating the high quality of the microarray data and ensuring comparability between the experimental groups. Differential expression analysis was performed using the 'limma' R package, which identified 1,281 DEGs. Among these, 817 genes showed increased expression, whereas 464 demonstrated reduced expression. The distribution of DEGs was illustrated using a volcano plot (Fig. 1B) and hierarchical clustering heatmap (Fig. 1C).

Gene co-expression network analysis was implemented using WGCNA methodology to elucidate patterns of gene coordination. For network construction, a scale-free topology criterion with R²=0.85 as the soft-thresholding parameter was employed, which optimized biological network fidelity (Fig. 1D). Module detection was performed through a combination of average linkage hierarchical clustering and dynamic branch cutting algorithms, yielding 22 discrete modules designated by distinct colors (Fig. 1E). The relationships between identified modules across control and DKD specimens were visualized using a correlation heatmap (Fig. 1F). Subsequently, MM and GS metrics were calculated to prioritize functionally relevant genes, resulting in 420 candidate hub genes. The overlap between these hub genes and the 1,281 DEGs identified by differential expression analysis was determined through Venn diagram analysis, which identified 49 genes of interest (Fig. 1G).

To identify relevant biological pathways associated with the disease phenotype, GSEA was performed comparing transcriptomics profiles between renal tissue from patients with DKD and healthy control individuals. The HALLMARK analysis in the GSE30529 dataset revealed the top 10 significantly enriched biological signaling pathways associated with DKD (Fig. 2A). These pathways include 'Integrin cell surface interactions', 'Antigen processing cross presentation', 'Neutrophil degranulation', 'Interferon signaling', 'Extracellular matrix organization', 'Core matrisome', 'Signaling by interleukins', 'Leishmania infection', 'TYROBP causal network in microglia' and 'Allograft rejection', suggesting their potential regulatory roles in the pathogenesis of diabetic nephropathy (Fig. 2B-K).

To elucidate the biological significance of the identified key genes, comprehensive functional characterization was performed. GO enrichment analysis revealed 457 significantly enriched terms, categorized into 351 BP, 50 CC and 56 MF terms. The most statistically significant terms in each category were selected for detailed examination (Fig. 2M-O). Within the BP category, prominent enriched pathways included 'neutrophil degranulation', 'neutrophil activation involved in immune responses', 'response to nutrient levels', 'T cell activation', 'response to nutrient', 'antigen processing and presentation', 'antigen processing and presentation of peptide antigen', 'regulation of innate immune response', 'positive regulation of defense response' and 'response to drug'. CC terms were mainly related to 'endocytic vesicle', 'phagocytic vesicle', 'specific granule', 'tertiary granule', 'recycling endosome', 'secretory granule membrane', 'glutamatergic synapse', 'focal adhesion', 'cell-substrate junction' and 'specific granule membrane'. MF terms included 'GTPase activity', 'lipase activity', 'magnesium ion binding', 'G protein-coupled receptor binding', 'Wnt-activated receptor activity', 'H4 histone acetyltransferase activity', 'apolipoprotein binding', 'mannose binding', 'triglyceride lipase activity' and 'Wnt-protein binding'.

Pathway enrichment analysis using the KEGG database identified 118 significantly associated signaling cascades for the candidate genes. After applying a statistical threshold of P<0.05, 11 pathways emerged as particularly relevant. These pathways encompassed neurodegenerative disorders ('Alzheimer disease'), infectious disease responses ('Tuberculosis', 'Epstein-Barr virus infection', 'Human cytomegalovirus infection' and 'Epithelial cell signaling in Helicobacter pylori infection'), immune functions ('Phagosome', 'Antigen processing and presentation' and 'IL-17 signaling pathway'), metabolic processes ('Glycerolipid metabolism' and 'Lipid and atherosclerosis') and neurological signaling ('GABAergic synapse') (Fig. 2L).

For examination of molecular interactions, the 49 identified DEGs were submitted to the STRING database to generate a PPI network. Network visualization was accomplished using Cytoscape software (Fig. 3A). To prioritize genes based on topological properties, three distinct algorithmic approaches were used (Fig. 3B-D). Further refinement of critical gene identification was performed through the application of three independent ML methodologies to extract feature genes of biological relevance. Feature selection was performed using the LASSO regression technique, which identified 13 genes with significant predictive capacity from the pool of variables demonstrating statistical significance in univariate analysis (Fig. 3E and F). The SVM-RFE recursive method was employed for core gene screening, identifying 16 predictive genes (Fig. 3G). Additionally, RF with feature selection, a robust and interpretable method, was used to adjust the number of classification trees and the importance of the screened genes, resulting in the selection of the top 19 hub genes in descending order (Fig. 3H). Integration of these six algorithms highlighted SYK, ADAM10 and APAF1 as hub genes associated with DKD (Fig. 3I and J).

A nomogram model was subsequently developed for diagnosing DKD based on the marker genes SYK, ADAM10 and APAF1, using the rms software package (Fig. 3K). DCA indicated that the nomogram model provided substantial clinical benefits (Fig. 3L). ROC curves demonstrated that the AUC values for SYK, ADAM10 and APAF1 were 0.950, 0.942 and 0.950, respectively, demonstrating their promise for biomarker development (Fig. 3M) and highlighting their high predictive accuracy. External validation using the GSE30122 dataset (quality of data normalization shown in Fig. S1) revealed significantly elevated transcript abundance of SYK, ADAM10 and APAF1 in the renal tissues of patients with DKD compared with that in healthy control samples (Fig. 3N-P). To assess the role of these hub genes in the context of DKD, hub gene analysis was performed using the Nephroseq v5 bioinformatics platform; the results showed that the expression levels of SYK, ADAM10 and APAF1 were positively correlated with Scr (Fig. 3Q-S) and negatively correlated with GFR (Fig. 3T-V).

Immunological profile analysis revealed marked distinctions in cellular composition between DKD specimens and normal renal tissue. Dimensionality reduction through PCA demonstrated clear separation between pathological and healthy samples, emphasizing the importance of immune system dysregulation in DKD progression (Fig. 4A). Computational deconvolution of the GSE30529 transcriptomic dataset using CIBERSORTx algorithms identified significant alterations in specific lymphocyte populations, particularly activated memory CD4⁺ T cells and T follicular helper cells (P<0.05; Fig. 4C and D). Additionally, correlation network analysis among 22 distinct immune cell subpopulations within glomerular compartments revealed intricate interconnections, indicating a sophisticated and highly coordinated immunoregulatory landscape in DKD pathophysiology (Fig. 4B).

To further elucidate the immunomodulatory mechanisms, associations between the expression profiles of hub genes (SYK, ADAM10 and APAF1) and signature transcripts of various immune cell subsets were assessed. Comprehensive correlation matrices (Fig. 4E-G) revealed that SYK expression exhibited significant positive associations with naïve B lymphocytes, T follicular helper cells, plasma cells and quiescent dendritic cells, while showing inverse relationships with activated memory CD4⁺ T cells, monocytes and memory B lymphocytes (all P<0.05). The ADAM10 transcript abundance demonstrated significant positive correlation with activated mast cells and resting natural killer cells, whereas significant negative associations were observed with activated natural killer cells, activated memory CD4⁺ T cells, M0 macrophages, monocytes, CD8⁺ T lymphocytes and resting mast cells (all P<0.05). For APAF1, expression levels positively correlated with resting dendritic cells, but negatively correlated with M0 macrophages, activated memory CD4⁺ T cells, γ-δ T cells, monocytes and memory B lymphocytes (all P<0.05). These relationship patterns suggest that these three hub genes may function as critical modulators of immune cell dynamics in the pathogenesis of DKD.

To elucidate the expression profiles of hub candidate genes during the progression of DKD, fasting blood glucose levels were monitored ensuring they remained >16.7 mmol/l, and two distinct mouse models of DKD were successfully established (Fig. 5A). Subsequently, key clinical parameters in both DKD and control groups were evaluated, including the UACR, blood glucose, body weight, BUN and Scr. The results revealed that the UNx/STZ/HFD-induced DKD group exhibited higher UACR, blood glucose, BUN and Scr levels compared with the sham counterparts, coupled with substantial weight loss (Fig. 5B-F). Likewise, the db/db mice showed elevated UACR, body weight, blood glucose, BUN and Scr levels compared with the db/m control group (Fig. 5B-F).

Histopathological analysis was performed on renal tissue sections, employing H&E staining and Masson's trichrome staining. H&E staining identified hallmark pathological features of DKD, including glomerular hyperplasia, vacuolization of tubular epithelial cells and thickening of the glomerular basement membrane, when compared with controls. Masson's trichrome staining further revealed pronounced interstitial fibrosis in both DKD groups (Fig. 5G). Quantitative analysis results of H&E and Masson staining are shown in Fig. 5H and I.

To corroborate these histological observations, western blotting and RT-qPCR analyses were performed on renal tissues from both DKD mouse models. RT-qPCR results demonstrated a significant upregulation in the mRNA expression levels of SYK, ADAM10 and APAF1 in both DKD models (Fig. 5J-L). Western blot analysis further validated these findings, showing significantly increased protein expression of SYK, ADAM10 and APAF1 in the UNx/STZ/HFD model (Fig. 5M and N), with a similar pattern observed in the db/db mice (Fig. 5O and P). These results align with the bioinformatics predictions, providing robust evidence for the involvement of these genes in DKD progression.

To further investigate the role of hub genes in renal fibrosis, an in vitro model was established using HG-induced HK-2 cells. Western blot analysis revealed that, compared with in the LG group, three hub genes-SYK, ADAM10 and APAF1, alongside increased expression of fibrosis-related markers, including fibronectin, vimentin, and Snail, in HG-treated HK-2 cells (Fig. 6B, C, E, F, H and I). These findings suggested that the aberrant upregulation of hub genes may contribute to the progression of renal fibrosis.

To explore the functional relevance of these hub genes, siRNAs were utilized to knock down their expression. Compared with in the HG group, knockdown of SYK using SYK-siRNA resulted in a significant reduction in fibrosis marker expression (Fig. 6A-C). Similarly, silencing ADAM10 with ADAM10-siRNA led to a significant decrease in fibrosis markers (Fig. 6D-F). Furthermore, silencing APAF1 with APAF1-siRNA reversed the upregulation of fibrosis markers (Fig. 6G-I). These results provide strong evidence that hub genes serve a pivotal role in the regulation of renal fibrosis, and their targeted suppression can effectively attenuate fibrosis associated with DKD. This highlights the therapeutic potential of hub gene inhibition in mitigating the progression of renal fibrosis.

To further assess the diagnostic potential of hub genes in human DKD, the expression levels of three hub genes, SYK, ADAM10 and APAF1, were evaluated in peripheral blood samples from both patients with DKD and non-DKD (healthy) individuals using RT-qPCR. The results revealed significantly elevated expression levels of SYK, ADAM10 and APAF1 in the DKD cohort compared with those in the non-DKD group (P<0.05; Fig. 7A-C). These findings are consistent with the results of the bioinformatics analysis and provide robust evidence supporting the involvement of these genes in DKD progression. Furthermore, these results underscore the potential of these hub genes as valuable diagnostic biomarkers, providing new options for future clinical research and therapeutic exploration.