The GSE195799 dataset, procured from the Gene Expression Omnibus (GEO) repository (http://www.ncbi.nlm.nih.gov/geo/), was based on the GPL24247 Illumina NovaSeq 6000 platform (16). The scRNA-seq was conducted on CD45+ immune cells isolated from the kidney of Type 1 diabetic OVE26 mice 3 and 7 months of age. The single nucleus RNA sequencing (snRNA-seq) data matrix GSE131882 (14), encompasses samples from three individuals with early human DKD alongside three control subjects.

The tubulointerstitial transcriptomic data series of human DKD, wherein GSE30122, derived from the GPL571 platform, serves as the training cohort, and includes 10 DKD and 24 control individuals. For subsequent verification, GSE104954 is employed as the external validation, consisting of 17 DKD and 21 control participants.

We utilized the R package “Seurat” (version 5.1.0) for the processing of scRNA-seq and snRNA-seq data. In advance of conducting downstream analyses, filtered cells meet the following criteria: cells with gene expression exceeding 200 but less than 10,000, and mitochondrial gene expression under 10%. We employed the FindVariableFeatures function to identify the top 2000 high variability genes. The processes of cell clustering were carried out by the FindNeighbors and FindClusters functions. Additionally, the t-Distributed Stochastic Neighbor Embedding (t-SNE) and Uniform Manifold Approximation and Projection (UMAP) for reducing dimensions were implemented by the RunTSNE and RunUMAP functions.

Cell annotation is a vital procedure of the transcriptomic data, we conducted this by consulting CellMarker 2.0 (http://bio-bigdata.hrbmu.edu.cn/CellMarker/) and previous research. Moreover, the FindAllMarkers function was used to screen differentially expressed genes (DEGs) for each cluster. Multiple volcano graphs were generated to illustrate the significantly altered genes for each cell type.

Inter-cellular communication is constant to cope with numerous physiological and pathological triggers. We conducted cellular communication analysis with the CellChat R package (version 1.6.1). Using the CellChatDB database, we determined the likelihood of secreted signaling, ECM-receptor, and cell-cell contact interactions among cell types and highlighted overexpressed ligands and receptors.

With monocle2 (version 2.30.1), we engaged in pseudotime trajectory analysis of scRNA-seq data to delve into the dynamics of gene expression patterns (17). The process involves data preprocessing, selecting key genes, dimension reduction, and the linchpin step is to apply the orderCells function to rebuild the temporal trajectory of cell differentiation or development.

Upon interacting with the tissue microenvironment, macrophages do not completely polarize to a solitary phenotype but exist in a dynamic state within various subpopulations (18). MacSpectrum (https://macspectrum.uconn.edu), leveraging single-cell transcriptomics, has devised two algorithms that can discern the heterogeneity of macrophages (19). The Macrophage Polarization Index (MPI) is based on “polarization signature genes” that signify the inflammatory condition, and the Activation-Induced Macrophage Differentiation Index (AMDI) sorts “activation-induced differentiation signature genes” to index the relative stage of differentiation maturity. This stratifies macrophages into M1-like, M2-like, transitional, and pre-activation phenotypes, effectively mapping the changing subpopulation composition.

The sva package was applied to mitigate batch effects, discarding probes without gene symbols and averaging those with multiple symbols. By employing the limma package (20) to identify DEGs (|logFC| > 1 and adjusted p < 0.05), we harnessed ggplot2 and pheatmap to underscore the prominently altered genes about tubulointerstitial injury.

The clusterProfiler package is an indispensable tool for functional enrichment analysis. Gene Ontology (GO) (21) identifies significantly enriched biological processes (BP), cellular components (CC), and molecular functions (MF); and Gene Set Enrichment Analysis (GSEA) (22) calculates the enrichment scores for gene sets, thereby enhancing our comprehension of molecular interactions.

By mapping DEGs to the STRING database (https://cn.string-db.org/), we integrated established and predicted interactions to build the protein-protein interaction (PPI) network. Cytoscape (version 3.8.0) software enabled the visualization of the PPI network and leveraged the plugin to calculate degree. Furthermore, we selected Least

(LASSO) regression analysis to select hub genes. Their diagnostic effectiveness was evaluated using receiver operating characteristic (ROC) curve.

The expression of key genes was verified against the external validation dataset GSE104954. Furthermore, we implemented the Nephroseq V5 tool (https://nephroseq.org/) evaluate the correlation between key genes and clinical parameters in DKD patients.

Male C57BL/6 mice, aged between 8 and 9 weeks, were sourced from Huafukang Animal Center (Beijing, China) and housed in a specific pathogen-free environment with a 12-hour light/dark cycle. Blood glucose and body weight were monitored bi-weekly. The experimental group received an 8-week high-fat diet to induce insulin resistance, followed by intraperitoneal injections of 60mg/kg streptozotocin for five consecutive days to damage pancreatic beta cells. A sustained blood glucose level exceeding 16.7 mmol/L from tail vein monitoring is the criterion for a successful diabetes model. After an additional 12 weeks of high-fat diet feeding, the mice were euthanized, and blood, urine, and kidney tissue samples were collected. The study was approved by the Ethics Committee of Tianjin Medical University Chu Hsien-I Memorial Hospital (Tianjin, China).

Upon the addition of RIPA lysis buffer, protease inhibitors, and phosphatase inhibitors to the kidney tissue, the mixture was homogenized and then centrifuged at 12,000 rpm for 10 minutes at 4°C. The resulting supernatant was collected and set aside for subsequent procedures. Protein quantification was performed using a Bicinchoninic Acid protein assay kit. Following this, proteins were separated by molecular weights using sodium dodecyl sulfate polyacrylamide gel electrophoresis. After transferring the isolated proteins onto the PVDF membrane, it was blocked with 5% skim milk at room temperature for 2 hours to block non-specific binding sites. Afterwards, the primary antibody incubation was performed at 4°C overnight, encompassing antibodies for CD36 (A19016, ABclonal, China), ITGAM (ab133357, Abcam, Cambridge, UK), and CX3CR1 (13885-1-AP, Proteintech, China). Following a wash step, the secondary antibody was incubated for 1 hour. After thoroughly treating the membrane with ECL reagent, it was photographed in a gel imaging system, and the bands were analyzed using Image J software.

Kidney tissue RNA was extracted with Trizol reagent, and the purity and concentration were measured. Following reverse transcription into cDNA, the SYBR Green Fast qPCR Mix kit (ABclonal, China) was used to amplify the target genes in a 10μl reaction system. The primer sequences for the target genes are listed in Supplementary Table S1 . During data processing, GAPDH was utilized as an internal reference, and the 2^-ΔΔCt approach was employed for semi-quantitative analysis to compute the relative mRNA expression of the target genes. Each experiment was repeated independently three times.

To induce apoptosis in human proximal tubular epithelial cells, trypsinized cells were irradiated with UV light for 30 minutes. Apoptotic cells were collected by centrifugation, washed with serum-free DMEM, and thoroughly resuspended in Dio fluorescent dye (UElandy). Following a 20-minute incubation, the cells were centrifuged and washed again to yield labeled apoptotic cells. The Raw264.7 macrophages were cultured in low glucose (5.5mM) and high glucose (30mM) for 24 hours. Afterward, they were dyed with Did dye (UElandy). The labeled apoptotic cells and macrophages were co-incubated at a 5:1 ratio for 2 hours, followed by imaging with a fluorescence microscope to assess the engulfment of apoptotic cells. The efferocytosis index, defined as the ratio of macrophages that have engulfed apoptotic cells to the total macrophages count, was used for analysis.

R software (version 4.3.2) and GraphPad Prism (version 10.2.0) were utilized for statistical analyses. Comparison between two groups was conducted using the Wilcoxon or Student’s t-test. A p-value less than 0.05 was considered statistically significant.

The GSE131882 dataset comprises kidney samples from three individuals with early human DKD and three control subjects. After discarding low-quality cells, we obtained 12,862 cells from the diabetes group and 17,410 cells from the control group ( Supplementary Figure S1 , Supplementary Table S2 ). Subsequent dimensionality reduction and clustering yielded 33 clusters ( Figure 1A ). Cell annotation, guided by established literature and marker genes, culminating in the identification of 11 cell types: proximal convoluted tubule (PCT), parietal epithelial cells (PEC), loop of Henle (LOH), distal convoluted tubule (DCT), connecting tubule (CT), type A intercalated cells (ICA), type B intercalated cells (ICB), podocytes (PODO), endothelial cells (ENDO), mesangial cells (MES), leukocytes (LEUK) (14) ( Figure 1B ). Figures 1C, D present the distinctive features of cell marker genes using violin charts and heat maps. Figure 1E displays a volcano plot, which highlights the dynamics of gene expression profiles across various cell types. An upsurge is observed in the expression of IL7R, ITGAL, and IKZF3, simultaneously with a pronounced decrease in the expression of NTN4, SULT1C2, and LINC00472 within leukocytes. Moreover, we noted a reduction in the quantity of certain renal tubular cells and a contrasting rise in the infiltration of leukocytes as DKD progresses ( Figure 1F ).

By synthesizing secreted signaling, ECM-Receptor, and cell-cell contact, and incorporating prior insights into the interactions, Cellchat forecasts the signaling pathways that mediate cellular communication, enhancing our grasp on how cells behave within tissue environment. As depicted in Figure 2A , the strength and count of interactions between leukocytes and glomerular and tubular cells are augmented in DKD versus the control. Immune cell infiltration is essential in mediating diabetic kidney injury (23), as macrophages, operating as both ligands and receptors, communicate with glomerular and tubular cells, leading to glomerulosclerosis and interstitial fibrosis ( Figure 2B ).

CD80 and CD86, serving as markers on macrophages, engage with CD28 or CTLA-4 to govern the immune response through direct cell-cell contact ( Figures 2C, D ). Hyperglycemia leads to an enhanced expression of chemokines in kidney tissue, such as CCL2, CCL3, CCL4, CCL5, and CXCL12, involving in immune cell recruitment and modulating the inflammation ( Figures 2E, F ). Additionally, the FN1 signaling pathway features robust interactions between renal tubules and glomeruli ( Figure 2G ). The TGFΒ pathway also extends its influence on leukocytes ( Figure 2H ). The malfunctioning of the collagen and TGFΒ pathways might contribute to excessive ECM accumulation. Figure 2I visualizes the possible interactions of signaling pathways with leukocytes acting as the ligand.

With significant plasticity, macrophage phenotypes can alter in response to hyperglycemia. Strategically controlling the polarization of macrophages to strike a balance between M1 and M2 could be pivotal for combating DKD. Therefore, we employed scRNA-seq of the diabetic kidneys from OVE26 and WT mice to delve into the distinctive functions and mechanisms of macrophages in the progression of DKD.

We selected the GSE195799 dataset to delve into the CD45-positive immune cell subpopulations. CD45+ myeloid cells were isolated from the kidneys of 3-month-old mice. After quality control, data normalization, and dimensionality reduction, the data was categorized into 25 clusters ( Supplementary Figure S2 ). By leveraging and integrating characteristics from multiple experimental platforms for the inference of cell functions (13, 16, 24, 25), 18 unique types of immune cells infiltrating the kidney were identified ( Figure 3A ). The most prominent macrophages can be separated into “Resident Mac”, “triggering receptor expressed on myeloid cells 2 (Trem2)-high expressing (Trem2^hi) Mac”, “Mannose receptor C-type 1 (Mrc-1)-high expressing (Mrc1^hi) Mac”, “Inflammatory Mac”, “Interferon (IFN) gene signature-high expressing (IFN^hi) Mac”, “Proliferating Mac”, and “Infiltration Mac”.

First, we referred to these cell surface markers (C1qa, C1qb, C1qc, Lst1, and Cx3cr1) to pick out resident macrophages (25). Building on the shared genes of resident macrophages, five other macrophage subpopulations exhibit different biological roles due to the expression of other outstanding cell markers. Trem2^hi and Mrc1^hi Mac perform functions like M2 macrophages, including anti-inflammatory, phagocytic, tissue repair, and immunoregulatory functions. Conversely, IFN^hi and inflammatory Mac respond to interferon and secrete more pro-inflammatory factors (such as TNF-Α, IL-1Β, IL-6, etc.) and chemokines (such as Ccl3, Cxcl2, Cxcl9, etc.). Proliferating Mac activates genes related to DNA replication, repair, and cell cycle. Infiltration Mac were further subdivided into ly6c^hi (Ear2, Ace) and ly6c^low (Plac8, Ccl24) subpopulations. Additionally, dendritic cells (DC) were subdivided into cDC1, cDC2, pDC (plasmacytoid DC), and Ccr7^hi DC. We also manually annotated clusters into “NK Cell”, “T Cell”, “Neutrophil”, “Plasma cell”, “innate lymphoid cells (ILCs)”, and “B Cell” ( Figure 3B ). As illustrated in Figure 3C , in the diabetic kidneys of 3-month-old OVE26 mice, the proportion of resident and inflammatory macrophages slightly increased, whereas the counts of Ly6c^low infiltration macrophages, Ccr7^hiDC, and ILCs displayed a downward trend.

As diabetic kidney disease progresses, 7-month-old OVE26 mice show pronounced glomerular enlargement and mesangial expansion, leading to significant proteinuria (16). Simultaneously, the accumulation of immune cells is evident. Therefore, scRNA-seq analysis of the CD45-enriched myeloid cells helps elucidate the dynamic transformation of macrophages. Following the execution of the analysis procedures ( Supplementary Figure S3 ), we identified immune cell subpopulations in the kidneys of 7-month-old mice that are generally like those identified previously, but with a completely different overall proportion ( Figures 3D, E ). Fibronectin 1 (Fn1)-high expressing (Fn1^hi) Mac also possesses the characteristics of infiltration Mac (Ear2). In the OVE26 group, the proportion of resident macrophages decreased to less than 10%, while the number of Trem2^hi and Mrc1^hi macrophages with M2 characteristics experienced a sharp rise. Furthermore, DCs, plasma cells, and B cells also diminished ( Figure 3F ).

The plasticity of macrophages enables them to respond swiftly, intricately, and diversely under pathological conditions (26). However, the traditional classification does not adequately annotate the intricate features of macrophages. As such, we introduce an innovative and extensive annotation platform, MacSpectrum (19), which is based on two newly devised algorithms: the MPI and AMDI. By precisely mapping each cell to the “inflammatory” or “terminal maturation” status, MacSpectrum achieves a full spectrum and high-resolution annotation of macrophage subpopulations in vivo. With these algorithms, MacSpectrum categorizes macrophages into M1-like, M2-like, transitional, and pre-activation phenotypes ( Figure 4A ).

The dynamic changes of activation profiles in the total macrophage population of kidneys of varying ages are displayed in Figure 4B . Most macrophages are in the “pre-activation” phenotype across the different stages of the WT group. At 3 months of age, more than half of the macrophages exhibit lower MPI and AMDI. The transformation at 7 months old is reflected in the rise of M2-like macrophages in the WT group (7.84% at 3 months versus 24.00% at 7 months). Additionally, we noted a decline in the M1-like phenotype of macrophages within the OVE26 cohort at later time points (45.89% at 3 months versus 27.05% at 7 months), accompanied by a rise in the M2-like phenotype (11.02% at 3 months versus 22.47% at 7 months) ( Figure 4B ). Upon a meticulous examination of each subset, it was evident that nearly all groups followed a comparable pattern. The results offer additional support for the opinion that macrophages exist in a continuously activated state ( Figures 4C–H ).

Subsequently, we engaged in pseudo-temporal analysis to reveal the fate trajectories of macrophages within the kidneys of 7-month-old mice, and to reenact the timeline of cellular evolution using monocle2. The trajectory tree orders the cells based on pseudotime, with the deepest color representing the root ( Figures 5A–C ). At branch node 2, the trajectory splits into two cell fates. Figure 5D outlines the differential expression of genes across macrophage fates; genes like Apoe, Malat1, Ctsc, and Wwp1 are prominently expressed in fate 1, while Fn1, Ace, Itgal, and Fgr achieve peak levels in fate 2.

We performed GO enrichment analysis on the DEGs at 3 and 7 months of age to further comprehend the molecular functions of macrophage subpopulations ( Figures 6A, B ). They manifest both general biological functions and pinpoint pathways unique to certain subgroups. As demonstrated by the enrichment analysis results, the BP in the 3-month-old group are primarily centered on regulation of the cell cycle, vacuolar transport, antigen processing and presentation, and receptor-mediated endocytosis ( Figure 6A ). Nevertheless, the 7-month-old mice exhibit significant BP including regulation of autophagy, phagocytosis, leukocyte migration, and chemotaxis. Resident Mac and infiltrating Mac are particularly involved in myeloid cell differentiation and several inflammation-related pathways ( Figure 6A ). Immune cells communicate through direct interactions or secreted mediators, with CCL and macrophage migration inhibitory factor being one of the most prominent signaling pathways ( Figure 6C ).

As inflammation pathways gain prominence in the progression of DKD, M2 macrophages particularly exert anti-inflammatory effects through efferocytosis. Efferocytosis, the process by which macrophages engulf and clear dying cells, prevents the secondary release of harmful substances after cell death, thus facilitating the resolution of inflammation and the repair of tissues (27). To delve deeper into the hubs, We collated a list of genes associated with efferocytosis from literature ( Supplementary Table S3 ) (27–29), and subsequently used a Venn diagram to intersect this list with the DEGs of macrophage subpopulations, ultimately yielding 17 genes ( Figure 6D ). Cytoscape was employed to construct a network of protein interactions, where the nodes with greater degrees are colored redder. CD36, ITGAM, and CX3CR1 are among the nodes with higher degrees ( Figure 6E ). Figures 6F–H individually depict the expression of the above hub genes within phagocytes, where CD36 is prominently expressed in both Mrc1^hi Mac and Infiltrating Mac, while CX3CR1 is expressed across different macrophages.

At the outset, scRNA-seq was used to preliminarily identify candidate genes, which displayed distinctive biological functions in specific cell types. We then focused on the GSE30122 training dataset, leveraging this gene expression profile to uncover the crucial molecules linked to the progression of DKD. We set the logFC cutoff value at 1, yielding 459 upregulated and 82 downregulated DEGs ( Figure 7A ). Figure 7B is a heatmap showcasing genes that exhibit significantly varied expression. GSEA analysis highlighted the “phagosome” pathway as the most enriched, alongside other significant pathways such as cell adhesion molecules, chemokine, and NOD-like receptor signaling pathway ( Figure 7C ). We once again intersected the efferocytosis gene list with the DEGs from GSE30122 to identify common genes ( Figure 7D ). Following this, we applied LASSO regression to model fitting, selected, and assessed the diagnostic features for DN ( Figure 7E ). Considering the node count in the PPI network, we ended up with CD36, ITGAM, and CX3CR1 as the final genes of interest. We reviewed the expression levels of the hub genes, which, as shown in Figure 7F , exhibited an increasing trend in the DKD group. As shown in Figure 7G , the ROC curve illustrates superior performance in diagnosing DKD, especially with ITGAM and CX3CR1 boasting AUC values of 98.3%.

To explore the critical molecule associated with efferocytosis in DKD, the CIBERSORT algorithm is applicable to gauge the abundance of 22 immune cell types in the renal tubulointerstitial tissue ( Figure 7H ). We detected rises in the levels of monocytes, M1 and M2 macrophages, resting mast cells, and neutrophils within the DKD group, while in the control group, there was an upsurge in regulatory T cells, M0 macrophages, and activated mast cells ( Figure 7I ). Additionally, as shown by the correlation heatmap ( Figure 7J ), eosinophils and T cells follicular helper exhibited a significant positive correlation (r = 0.79), and conversely, a negative correlation was observed between M1 macrophages and regulatory T cells (r = -0.39). Among these immune cells, each pivotal gene displayed a positive relationship with M1 macrophages, with CD36 additionally correlating positively with M2 macrophages ( Figure 7K ).

We chose GSE104954 as the external validation dataset, and in line with the training set results, the hub gene showed a similarly increased level in diabetic kidneys, demonstrating the model’s robust assessment performance ( Figures 8A–C ). Following this, we turned to the Nephroseq database to gauge the prognostic importance of key molecules. An increase in CD36 mRNA deposits in the renal tubulointerstitial corresponded to a downward trend in GFR (r = -0.75, p = 0.008). Overexpression of ITGAM (r = 0.60, p = 0.049), and CX3CR1 (r = 0.65, p = 0.032) was positively linked to proteinuria in DN ( Figures 8D–F ). Aligning with database outcomes, our research employing qPCR and Western blot assays has identified a significant upregulation of hub gene expression in DKD kidney tissues ( Figures 9A–C ).

Altered Macrophage Behavior and Efferocytosis Decline in DKD

According to prior analyses, OVE26 mice experienced greater infiltration of Trem2^hi, Mrc1^hi, inflammatory, IFN^hi, and Fn1^hi macrophages in the kidney ( Figure 3 ). Similarly, in the diabetic mouse model induced by a high-fat diet combined with STZ, we observed upregulated expression of Trem2, Mrc1, iNOS, IL-1Β, CCL3, and Fn1 ( Figure 10A ). Additionally, we detected a decline in the expression of efferocytosis-related molecules MERTK, AXL, and MFGE8 in DKD ( Figure 10B ). The phagocytic capability of macrophages to engulf apoptotic cells was also diminished after high glucose exposure ( Figure 10C ).