Abstract

   Diabetic nephropathy (DN) exhibits profound spatial metabolic
   heterogeneity across kidney regions, yet how compartmentalized pathways
   drive disease progression remains poorly defined. A deeper
   understanding of the organizational spatial environment and metabolic
   pathways of diabetic kidney damage will provide new insights to develop
   new therapies. By integrating high-resolution spatial multi-omics and
   single-cell transcriptomics, we mapped region-specific metabolic
   dysregulation in diabetic kidneys, identifying glutathione metabolism,
   pentose phosphate, and glycolytic pathways as zonally disrupted in
   cortical and medullary regions. Spatial metabolomics revealed distinct
   anatomical clustering of ten clinically associated metabolites, while
   spatial proteomic profiling uncovered sixty-four region-enriched
   proteins linked to these pathways. Specifically, depending on anatomic
   location, spatial protein signatures across multiple regions of
   diabetic mouse kidneys were enriched in each segmentation,
   respectively. Cross-species integration identified GPX3 as a
   fibroblast-enriched biomarker strongly correlated with kidney
   dysfunction and closely related to clinical indicators. Notably,
   astragaloside IV (ASIV) treatment reversed spatial metabolic
   perturbations in diabetic mice, restoring glutathione and glycolytic
   pathway activity in a compartment-specific manner. Single-cell analyses
   identified five cell types—endothelial cells, fibroblasts, epithelial
   cells, macrophages and neutrophils—and further revealed fibroblasts as
   key contributors to regulatory effects via GPX3 overexpression.
   Importantly, the higher expression of Gpx3 in fibroblasts compared to
   other cell types, Gpx3 (AUC = 0.995), was further validated,
   demonstrating the high sensitivity and specificity for DN patients.
   This multimodal atlas establishes the spatially resolved metabolic
   blueprint of DN, bridging molecular zoning with anatomical localization
   of renal tissue to unveil actionable therapeutic targets for metabolic
   disorders in kidney disease.

   Keywords: Spatial proteomics, Spatial metabolomics, Metabolic zonation,
   Metabolism, Metabolic pathway, Target

Graphical abstract

   [41]Image 1
   [42]Open in a new tab

Highlights

     * •
       Spatial multi-omics maps DN's metabolic zonation & actionable
       targets.
     * •
       Cross-species validation bridges murine models to human DN
       pathology.
     * •
       ASIV restores compartment-specific glutathione & glycolytic
       pathways.
     * •
       Single-cell atlas links fibroblasts to metabolic dysregulation via
       GPX3.
     * •
       GPX3 identified as fibroblast-enriched biomarker (AUC = 0.995) for
       DN.

1. Introduction

   Diabetic nephropathy (DN), a metabolic disorder-driven complication of
   diabetes characterized by progressive kidney damage ([43]Cefalu et al.,
   2024; [44]Chai et al., 2025; [45]Suzuki et al., 2024), arises from
   spatially heterogeneous dysfunction across renal subregions with
   distinct metabolic roles ([46]Guo, Dong, et al., 2024; [47]Lees et al.,
   2019). The structural complexity of the kidney and multifaceted injury
   mechanisms underlie current therapeutic limitations, necessitating
   multidimensional approaches to unravel its pathology. Recent advances
   in high-resolution spatial mapping integrate omics and imaging to
   decode the molecular networks governing disease progression
   ([48]Asowata et al., 2024; [49]Govind et al., 2022; [50]Li & Humphreys,
   2024b), offering dual promise and challenge to analyzing the
   comprehensive molecular profile of kidney tissue across mechanism and
   treatment.

   Spatial organization of metabolites within the kidney's 3D architecture
   critically governs DN progression, as compartmentalized metabolic
   activity directly shapes renal function and pathology ([51]Addario et
   al., 2024; [52]Kadotani et al., 2024; [53]Lee et al., 2024). This
   spatial regulation integrates multi-omic networks with tissue
   morphology, where metabolomics serves as the functional endpoint,
   synthesizing transcriptomics and proteomics to define
   disease-associated cellular states ([54]Cai et al., 2023; [55]Qiu et
   al., 2023). While spatially resolved metabolomics enables systematic
   mapping of metabolic perturbations and pathway dysregulation, current
   analyses remain fragmented across renal subregions, limiting
   mechanistic insights and therapeutic target identification. Single-cell
   transcriptomics has unveiled cell-type-specific injury mechanisms
   ([56]Juliar et al., 2024; [57]Li & Humphreys, 2024b; [58]Polonsky et
   al., 2024; [59]Zhang et al., 2024), yet its lack of spatial context
   obscured microenvironmental interactions. Recent advances in spatial
   multi-omics now enable multimodal reconstruction of regulatory networks
   across anatomical zones ([60]Cao et al., 2024; [61]Gopee et al., 2024;
   [62]Iglesia et al., 2024; [63]Ounadjela et al., 2024; [64]Qian et al.,
   2024). Nevertheless, the interplay between zonal metabolic
   reprogramming, cellular pathophysiology, and drug-responsive pathways
   in DN remains poorly resolved, highlighting the need for integrative
   frameworks to decode spatially targeted therapeutic opportunities.

   While spatial omics resolves metabolite localization in 2D tissue
   sections, 3D functional zonation mapping remains unachieved. Mass
   spectrometry imaging-based spatial metabolomics ([65]Vandergrift et
   al., 2025; [66]Ponzoni et al., 2024) advances multiplexed metabolite
   profiling while retaining tissue architecture, enabling systematic
   characterization of injury-associated microenvironmental remodeling.
   Integrating single-cell RNA sequencing (scRNA-seq) with spatial
   multi-omics bridges molecular gradients to cellular behaviors, offering
   unprecedented resolution to decoding the kidney injury mechanisms.
   These multimodal frameworks not only uncover spatially defined
   therapeutic metabolites but also pioneer next-generation strategies for
   renal compartment-specific diagnosis and intervention.

   Our current understanding of DN molecular mechanisms remains
   incomplete, necessitating integrative multi-omics and scRNA-seq
   approaches to elucidate disease progression pathways. Through spatial
   multi-omics mapping and renal zonation analysis, we systematically
   characterized protein-metabolic pathway enrichment patterns in diabetic
   murine kidneys. Crucially, spatial resolution of metabolic pathway
   localization within renal tissue compartments represents a pivotal
   advancement for deciphering DN pathogenesis.

2. Results

2.1. Machine learning-enhanced metabolomic profiling identifies diagnostic
biomarkers and dysregulated metabolic pathways in diabetic nephropathy

   Integrated strategy of scRNA-seq analyses and spatial metabolomics as
   well as spatial proteomics was shown in [67]Fig. S1.
   Principal-component analysis (PCA) scores revealed significant
   differences in urine metabolite expression between DN patients and
   healthy controls (HCs) ([68]Fig. 1(A)). In addition, hierarchical
   analysis of the metabolite signals in urine samples from DN patients
   and HCs also presented clear inter-group clustering ([69]Fig. 1(B)). We
   screened and analyzed the differentially expressed metabolites in DN
   patients and a total of 10 abundant metabolites were found ([70]Fig.
   1(C), [71]Table S2). Heatmap plot illustrated the relative abundance of
   the differentially abundant metabolites between the DN and HCs
   ([72]Fig. 1(D)). The KEGG pathway analysis results showed that citrate
   cycle, glutathione metabolism, pentose phosphate pathway,
   glycerophospholipid metabolism, fructose and mannose metabolism, fatty
   acid biosynthesis, glycolysis/gluconeogenesis were significantly
   enriched ([73]Fig. 1(E), [74]Table S3). Correlations analysis (Mantel
   test) on differentially abundant metabolites and the clinical
   biochemical indicators revealed homocitric acid and pyroglutamic acid
   were positive correlation (p < 0.05) with BUN, serum creatinine, eGFR
   and blood glucose, myristic acid was positive correlation (p < 0.05)
   with blood glucose, serum creatinine and eGFR ([75]Fig. 1(F), [76]Table
   S4). The predictive performances of each metabolite between the DN and
   HCs were illustrated via ROC curves ([77]Fig. 1(G)). Intriguingly, we
   observed AUC values on the prediction model demonstrated the high
   diagnostic capability of metabolite combination panel from potential
   biomarkers ([78]Fig. 1(H)). In [79]Fig. 1(I), the bar chart showed that
   top feature contribution on the performance evaluation of the
   differentially abundant metabolites in DN patients. Furthermore,
   [80]Fig. S2 showed the expression status on feature contribution of
   potential biomarkers for the prediction model evaluation.

Fig. 1.

   [81]Fig. 1
   [82]Open in a new tab

   Metabolic profiles and identification of differentially abundant
   metabolites in diabetic nephropathy patients using non-targeted
   metabolomics.

   (A) Principal-component analysis (PCA) scores of urine samples between
   the diabetic nephropathy (DN) patients (n = 29) and healthy controls
   (HCs) (n = 29). (B) Hierarchical clustering analysis of the metabolite
   signals in urine samples from DN patients (n = 29) and HCs (n = 29).
   (C) Volcano plot highlights significant differentially expressed
   metabolites between the DN and HCs. (D) Heatmap plot showing the
   relative abundance of the differentially abundant metabolites in the
   patients (n = 29) and HCs (n = 29). (E) Pathway enrichment analysis of
   differentially abundant metabolites in DN patients. (F) Correlations
   analysis (Mantel test) on differentially abundant metabolites and the
   clinical biochemical indicators (blood glucose, serum creatinine, blood
   uric acid, eGFR, BUN). (G) ROC curves illustrating the predictive
   performance of the differentially abundant metabolites among the DN
   patients (n = 29) and HCs (n = 29). (H) The diagnostic capability of
   potential biomarkers via receiver operating characteristic curves on
   comparison of AUC values of the prediction model demonstrating the
   performance of the metabolite combination panel, based on the logistic
   regression algorithm; ROC curves are generated by Monte-Carlo cross
   validation (MCCV), and the built-in PLS methods was used for the
   feature ranking and corresponding classification. (I) Top feature
   contribution on the performance evaluation of the prediction model.

2.2. Spatially resolved metabolomics unveils anatomical zonation and
pathway-specific metabolic remodeling in diabetic nephropathy mouse kidneys

   Principal component analysis of metabolites can significantly
   distinguish kidney tissue samples from db/m mice and transgenic db/db
   mice ([83]Fig. 2(A)). Targeted metabolomics revealed group clustering
   of the metabolite signals in kidney tissue samples from db/m mice and
   transgenic db/db mice ([84]Fig. 2(B)). Heatmap of the clustering
   analysis of tissue samples indicated metabolites changes between db/m
   mice and transgenic db/db mice ([85]Fig. 2(C)). High-spatial-resolution
   MALDI-MSI was applied to detect the differentially abundant metabolites
   from mice kidney tissue at spatial resolution. Spatially resolved
   metabolomics mass spectrometry imaging (MSI) revealed the unique
   metabolite biomarkers signature ([86]Fig. 2(D)). [87]Fig. 2(E) displays
   the schematic diagram for kidney organizational annotations (Cor,
   kidney cortex including its vasculature; OM, outer stripe of kidney
   medulla including its vasculature; IM, the inner stripe of kidney
   medulla consisting of some vasculature, the main arteries and veins
   within the renal parenchyma, papilla and pelvis, part of the external
   renal vessels, part of the ureter, and uniform surrounding tissue).
   Heatmap of spatial segmentation analysis then highlighted metabolite
   changes of mice kidney samples from db/m and transgenic db/db mice
   ([88]Fig. 2(F)).

Fig. 2.

   [89]Fig. 2
   [90]Open in a new tab

   Spatially resolved metabolomics highlights spatial segmentation
   profiling analysis using mass spectrometry imaging.

   (A) Principal component analysis of metabolites in kidney tissue
   samples from db/m mice and transgenic db/db mice. (B) Hierarchical
   clustering analysis of the metabolite signals in kidney tissue samples
   from db/m mice and transgenic db/db mice. (C) Heatmap of the clustering
   analysis of tissue samples between db/m mice and transgenic db/db mice.
   (D) Mass spectrometry imaging (MSI) analysis of metabolite biomarkers
   (Ⅰ, cysteinylglycine; Ⅱ, pyroglutamic acid; Ⅲ, stearic acid; Ⅳ,
   dihydroxyacetone phosphate; Ⅴ, dodecanoic acid; Ⅵ, myristic acid; Ⅶ,
   homocitric acid; Ⅷ, citric acid; Ⅸ, gluconic acid; Ⅹ, deoxyribose
   1-phosphate) in the kidney tissue of db/m mice (upper) and transgenic
   db/db mice (below). (E) Schematic diagram for kidney organizational
   annotations (Cor, kidney cortex including its vasculature; OM, outer
   stripe of kidney medulla including its vasculature; IM, the inner
   stripe of kidney medulla consisting of some vasculature, the main
   arteries and veins within the renal parenchyma, papilla and pelvis,
   part of the external renal vessels, part of the ureter, and uniform
   surrounding tissue). (F) Heatmap of spatial segmentation analysis of
   differential metabolites in kidney tissue samples from db/m and
   transgenic db/db mice. (G) Spatial feature plot of cysteinylglycine
   (Ⅰ), pyroglutamic acid (Ⅱ) and stearic acid (Ⅲ) in mice kidney tissue
   (Cor) segmentation by feature extraction, similarity evaluation,
   clustering algorithms. (H) Pathway enrichment analysis on metabolite
   expression for mice kidney tissue (Cor) region in transgenic db/db
   mice. (I) MALDI-MS imaging for spatial localization and distribution of
   dihydroxyacetone phosphate (Ⅳ), dodecanoic acid (Ⅴ), myristic acid (Ⅵ)
   and homocitric acid (Ⅶ) in mice kidney tissue (OM) segmentation
   analysis through by feature extraction, similarity evaluation,
   clustering algorithms, and result visualization. (J) Pathway enrichment
   analysis for metabolites in mice kidney tissue (OM) region of
   transgenic db/db mice. (K) Spatial distribution of citric acid (Ⅷ),
   gluconic acid (Ⅸ) and deoxyribose 1-phosphate (Ⅹ) in mice kidney tissue
   (IM) with MALDI-MS imaging analysis and dimensionality reduction and
   segmentation. (L) Biological functions and metabolite pathway
   enrichment analysis for mice kidney tissue (IM) region. (M) Sankey
   diagram indicates the relationship between metabolites and pathways.

   To explore the anatomical region difference at the MSI-based spatially
   resolved metabolomics, similarly, the spatial metabolome was divided
   into Cor, OM, IM based on kidney organizational annotations sections.
   Spatial feature plots of cysteinylglycine, pyroglutamic acid and
   stearic acid were specific for mice kidney tissue (Cor) segmentation by
   feature extraction, similarity evaluation, clustering algorithms
   ([91]Fig. 2(G)), mainly enriched in glutathione metabolism ([92]Fig.
   2(H), [93]Table S5). Additionally, for dihydroxyacetone phosphate,
   dodecanoic acid, myristic acid and homocitric acid, the spatial
   location of the metabolic clusters showed zonation distributions in the
   mice kidney tissue (OM) segmentation analysis ([94]Fig. 2(I)),
   depending on the anatomical structure of the kidney, were associated
   with pentose phosphate pathway ([95]Fig. 2(J), [96]Table S6). MALDI-MS
   imaging analysis and dimensionality reduction showed the spatial
   distribution of citric acid, gluconic acid and deoxyribose 1-phosphate
   in mice kidney tissue (IM) segmentation ([97]Fig. 2(K)), enriched in
   glycolysis/gluconeogenesis pathway ([98]Fig. 2(L), [99]Table S7). We
   found these distinct metabolites were associated with characterization
   zonation and the corresponding metabolic pathway of mice kidney,
   maintaining a close relationship between metabolites and pathways
   within the kidney tissue segmentation profiling ([100]Fig. 2(M),
   [101]Table S8).

2.3. Spatial proteomic network analysis of multi-regional kidney zonation
reveals compartmentalized functional modules via WGCNA

   Original H&E staining image for spatial proteome of kidney multiregion
   was shown in [102]Fig. 3(A). We used WGCNA to construct distinct
   protein co-expression modules in different clusters for spatial
   proteome of kidney multi-region. The hierarchical clustering analysis
   of protein expression profiles in mouse kidney cortex (Cor) region
   identified distinct expression modules ([103]Fig. 3(B)). Subsequent
   topological overlap matrix analysis demonstrated significant
   correlations between these modules ([104]Fig. 3(C)), with the brown
   module showing particularly strong associations with other modules
   through protein co-expression patterns ([105]Fig. 3(D)). A
   protein-protein interaction network was constructed using hub proteins
   from this brown module ([106]Fig. 3(E)), whose enriched genes were
   functionally linked to antigen processing/presentation of exogenous
   peptides, interferon-beta response, and intermediate filament
   organization ([107]Fig. 3(F)). Parallel analysis in the outer medulla
   (OM) region revealed protein expression modules through hierarchical
   clustering ([108]Fig. S3A), with module correlations quantified by
   heatmap and dendrogram analyses ([109]Fig. S3B and C). The red module
   emerged as a central hub showing high connectivity with other modules.
   Its network architecture was mapped using key hub proteins ([110]Fig.
   S3D), and functional enrichment analysis associated these red module
   genes with intermediate filament dynamics, keratinization processes,
   and keratinocyte differentiation ([111]Fig. S3E). For the inner medulla
   (IM) region, hierarchical clustering similarly resolved protein
   expression modules ([112]Fig. S3F) with inter-module correlations
   displayed in matrix format ([113]Fig. S3G). Adjacency heatmaps and
   eigengene analysis further characterized module preservation across
   samples ([114]Fig. S3H). The purple module's hub proteins formed a
   distinct interaction network ([115]Fig. S3I), while its feature genes
   showed enrichment in glycosaminoglycan biosynthesis pathways ([116]Fig.
   S3J). Across all kidney regions, module analysis revealed
   region-specific patterns of protein co-expression, highlighting
   compartmentalized biological functions-ortical modules emphasizing
   immune-related processes, outer medullary modules focusing on
   structural differentiation, and inner medullary modules associated with
   extracellular matrix metabolism. The integrated network approach
   employing WGCNA successfully identified region-specific protein
   interaction modules, providing a systems-level perspective on renal
   zonation at the proteomic level.

Fig. 3.

   [117]Fig. 3
   [118]Open in a new tab

   Weighted co-expression network analysis (WGCNA) for spatial proteome of
   kidney multiregion.

   (A) Original H&E staining image for spatial proteome of kidney
   multiregion. (B) Hierarchical clustering tree and module detection
   based on protein/peptide expression patterns in mice kidney tissue
   (Cor) segmentation; each colour represents a module. (C) Topological
   overlap matrix for correlation between different protein modules and
   protein expression in mice kidney tissue (Cor) segmentation. (D)
   Dendrogram of consensus module eigengenes and heatmap of module
   adjacencies based on protein expression patterns in mice kidney tissue
   (Cor) segmentation. The heatmap shows the correlations (positive or
   negative) between the identified modules. (E) Network diagrams
   constructed using the hub proteins in the brown module based on protein
   expression patterns in mice kidney tissue (Cor) segmentation. The brown
   module was highly correlated with DN and the genes in the brown module
   are labeled as the WGCNA-hub genes. (F) Biological processes enrichment
   of the feature genes of brown module in mice kidney tissue (Cor)
   segmentation.

2.4. Spatial proteome identifies differential proteins linked to metabolic
pathways of multi-regional kidney

   Fuzzy c-means algorithm analysis tool was used to examine the potential
   functions of these spatial proteins by determining GO/KEGG/Domain, and
   primarily involved in carboxylic acid transmembrane transporter
   activity, anion transmembrane transporter activity, small molecule
   metabolic process, etc ([119]Fig. 4(A)). A total of 64 differentially
   expressed proteins were screened by the circular volcano plot from
   kidney tissue multiregion ([120]Fig. 4(B), [121]Table S9). As shown in
   [122]Fig. 4(C), the heatmap demonstrated the differentially expressed
   proteins in kidney tissue (Cor) segmentation. Protein-protein
   interaction network (number of nodes: 63, number of edges: 63, average
   node degree: 2, expected number of edges: 21 PPI enrichment p-value:
   1.47e−13) of the differentially expressed proteins constructed and
   identified in the kidney tissue (Cor) segmentation ([123]Fig. 4(D)).
   Pathway enrichment revealed that the differentially expressed proteins
   in kidney tissue (Cor) segmentation were significantly related to the
   functions “glutathione metabolism”, “metabolism of xenobiotics”, “fatty
   acid biosynthesis” and “renin-angiotensin system” ([124]Fig. 4(E),
   [125]Table S10). Heatmap showed 46 differential proteins in kidney
   tissue (OM) segmentation ([126]Fig. 4(F), [127]Table S11). Protein
   networks (number of nodes: 45, number of edges: 46, average node
   degree: 2.04, expected number of edges: 7, PPI enrichment p-value:
   <1.0e−16) displayed the interacted proteins in the kidney tissue (OM)
   segmentation ([128]Fig. 4(G)). They were related to carbon metabolism,
   glucagon signaling pathway, pentose phosphate pathway, fructose and
   mannose metabolism in the pathway category in ​kidney tissue (OM)
   segmentation ([129]Fig. 4(H), [130]Table S12). Heatmap analysis
   presented 24 differentially expressed proteins in kidney tissue (IM)
   segmentation ([131]Fig. 4(I), [132]Table S13). The highly connected
   protein networks (number of nodes: 24, number of edges: 17, average
   node degree: 1.42, expected number of edges: 2, PPI enrichment p-value:
   2.52e−10) was related to kidney tissue (IM) segmentation ([133]Fig.
   4(J)). KEGG analysis indicated that “glycolysis”, “glycine, serine and
   threonine metabolism”, and “biosynthesis of amino acid” pathways were
   enriched in kidney tissue (IM) segmentation ([134]Fig. 4(K), [135]Table
   S14).

Fig. 4.

   [136]Fig. 4
   [137]Open in a new tab

   Spatial protein characteristics of diabetic mice kidney multiregion
   segmentation.

   (A) Summary of protein expression pattern clustering analysis based on
   the Fuzzy c-means algorithm analysis of GO/KEGG/Domain. (B) Circular
   volcano plot for the differentially expressed proteins from kidney
   tissue multiregion. (C) Heatmap showing the differentially expressed
   proteins in kidney tissue (Cor) segmentation. (D) Protein-protein
   interaction network of the differentially expressed proteins identified
   in the kidney tissue (Cor) segmentation. (E) Functional enrichment
   analysis of differentially expressed proteins in kidney tissue (Cor)
   segmentation. (F) Heatmap display the differentially expressed proteins
   in kidney tissue (OM) segmentation. (G) Networks of highly connected
   proteins related to kidney tissue (OM) segmentation. (H) KEGG
   enrichment analysis of the differentially expressed proteins in kidney
   tissue (OM) segmentation. (I) Heatmap analysis of the differentially
   expressed proteins in kidney tissue (IM) segmentation. (J) Protein
   networks displaying the interacted proteins in the kidney tissue (IM)
   segmentation. (K) Pathway enrichment analysis of the differentially
   expressed proteins in kidney tissue (IM) segmentation.

2.5. Proteomics with clinical data integration links protein-clustered
pathways in diabetic nephropathy

   The PCA scores ([138]Fig. 5(A)) and three-dimensional score ([139]Fig.
   S4A) of urine samples between DN patients and HCs based on the targeted
   proteome had the significant separation state. Heatmap ([140]Fig. S4B)
   and horizontal clustering analysis ([141]Fig. 5(B)) visualized the
   clustering of targeted proteome ([142]Table S15) between DN patients
   and HCs. The target proteins identified in the urine samples of DN
   patients can build a tight protein interaction network ([143]Fig. 5(C),
   number of nodes:97, number of edges: 244, average node degree: 5.03,
   expected number of edges:64, PPI enrichment p-value: < 1.0e−16).
   Molecular function of the target proteins identified in the urine
   samples enriched in the extracellular matrix structural constituent,
   growth factor binding, serine-type peptidase activity, antioxidant
   activity ([144]Fig. S4C). KEGG pathway enrichment of the target
   proteins identified in the urine sample was close with complement and
   coagulation cascades, protein digestion and absorption, metabolic
   pathways, and glycolysis/gluconeogenesis ([145]Fig. S4D). Enrichment of
   biological functions of the target proteins identified in the urine
   samples enriched in the “blood coagulation, fibrin clotformation”,
   “zymogen activation”, “cellular detoxification”, “response to
   wounding”, “anatomical structure homeostasis”, “leukocyte mediated
   immunity”, “immune effector process” ([146]Fig. S4E). Based on the
   pathway enrichment analysis ([147]Table S16) of the differential
   proteins and potential metabolites, we found that Eno1, Bpgm, Fbp2 and
   dihydroxyacetone phosphate actively participated in glycolysis or
   gluconeogenesis (p = 0.0011); Gpx3, Gsto1, cysteinylglycine and
   pyroglutamic acid participated in glutathione metabolism (p = 0.0060);
   Fbp2, deoxyribose 1-phosphate and gluconic acid actively participated
   in pentose phosphate pathway (p = 0.0239). Reactome pathway enrichment
   analysis manifested the targeted proteome had a close relation with
   collagen degradation, assembly of collagen fibrils, collagen chain
   trimerization, degradation of the extracellular matrix, etc ([148]Fig.
   S4F). Sangi correlation diagram then showed the close cooperation
   between metabolites and proteins ([149]Fig. 5(D)). BPGM were positively
   correlated with blood glucose, eGFR; ENO1 were positive correlation
   with GFR, blood glucose, serum creatinine, blood uric acid, and BUN;
   GPX3 was positively correlated with blood uric acid, while negatively
   correlated with eGFR, BUN, blood glucose, and serum creatinine
   ([150]Fig. 5(E), [151]Table S17).

Fig. 5.

   [152]Fig. 5
   [153]Open in a new tab

   Targeted proteomics and clinical data integration identified key
   proteins of DN phenotype.

   (A) PCA scores of urine samples between DN patients and HCs based on
   the targeted proteome. (B) Horizontal clustering analysis of all the
   urine samples from DN patients and HCs. (C) String for protein-protein
   interaction network of the target proteins identified in the urine
   samples of DN patients. (D) Sangi correlation diagram of the
   differential proteins and potential metabolites based on the pathway
   enrichment analysis. (E) Correlation analysis of the key proteins and
   clinical indices from DN patients, blue dots indicate negative
   correlations and red dots indicate positive correlations.

2.6. Spatially resolved metabolomics unravels MET/ASIV-driven
compartment-specific metabolic pathway restoration in diabetic nephropathy
mouse kidneys

   To understand spatial metabolic regulation in DN, we conducted a
   comprehensive analysis of spatial metabolite imaging data from diabetic
   mouse kidney with MET and astragaloside IV (ASIV) treatment. Firstly,
   PCA scores plot ([154]Fig. 6(A)) and 3D visualization plot ([155]Fig.
   6(B)) of the targeted metabolites were clearly separated on the
   transgenic db/db mice and db/m mice, while after MET and ASIV
   treatment, and its overall state is in the middle. Heatmap
   visualization clarifies its regulatory role of MET and ASIV treatment
   for kidney samples of db/m mice based on the relative intensity of
   metabolite signatures ([156]Fig. 6(C)). VIP scores of PLS-DA model
   ranked the metabolite biomarkers following ASIV administration in
   transgenic db/db mice ([157]Fig. 6(D)). High-resolution MALDI-MSI
   revealed the distribution of metabolite biomarkers in diabetic kidney
   tissue sections after MET and ASIV treatment ([158]Fig. 6(E)). MALDI-MS
   imaging via feature extraction, similarity evaluation, clustering
   algorithms showed spatial segmentation of cysteinylglycine,
   pyroglutamic acid and stearic acid in kidney tissue (Cor) segmentation
   from diabetic mice kidney sections after MET and ASIV treatment
   ([159]Fig. S5A). MALDI-MS imaging analysis by feature extraction,
   similarity evaluation, clustering algorithms, exhibited spatial
   localization and distribution of dihydroxyacetone phosphate (Ⅳ),
   dodecanoic acid (Ⅴ), myristic acid (Ⅵ) and homocitric acid (Ⅶ) in
   kidney tissue (OM) segmentation in diabetic mice kidney sections after
   MET and ASIV treatment ([160]Fig. S5B). Spatial segmentation of citric
   acid, gluconic acid and deoxyribose 1-phosphate was demonstrated in
   kidney tissue (IM) segmentation from diabetic mice kidney sections
   after MET and ASIV treatment ([161]Fig. S5C). Heatmap showed the
   characteristic metabolite expression in region-specific localization of
   mice kidney tissue with MET and ASIV treatment by spatial metabolomics
   ([162]Fig. S5D). Pathway enrichment analysis of ASIV altered
   differential metabolites in Cor region of mice kidney tissue
   predominantly associated with glutathione metabolism, fatty acid
   biosynthesis, biosynthesis of unsaturated fatty acids ([163]Fig. S5E).
   KEGG enrichment analysis of ASIV changed the marker metabolites in OM
   region of mice kidney tissue were characteristically expressed in
   pentose phosphate pathway, citrate cycle, glucagon signaling pathway
   ([164]Fig. S5F). Pathway enrichment analysis of ASIV affected the
   metabolites in IM region of mice kidney tissue enriched in the
   glycolysis and glycerophospholipid metabolism ([165]Fig. S5G).

Fig. 6.

   [166]Fig. 6
   [167]Open in a new tab

   Deciphering metabolic atlas of diabetic mouse kidney with ASIV
   treatment by spatial metabolomics.

   (A) PCA scores plot of metabolites in transgenic db/db mice after MET
   and ASIV treatment. (B) 3D Visualization on scores plot of metabolites
   in transgenic db/db mice after MET and ASIV treatment. (C) Heatmap
   visualization based on the relative intensity of metabolite signatures
   of kidney samples of db/m mice, transgenic db/db mice, MET and ASIV
   treatment. (D) VIP scores of PLS-DA model for ranking the metabolite
   biomarkers following ASIV administration in transgenic db/db mice. (E)
   high-resolution matrix-assisted laser desorption ionization-mass
   spectrometry imaging (MALDI-MSI) imaging distribution of metabolite
   biomarkers in db/m mice (1st row), db/db mice (2nd row) kidney tissue
   sections in diabetic mice kidney sections after MET (3rd row) and ASIV
   treatment (4th row) (Ⅰ, cysteinylglycine; Ⅱ, pyroglutamic acid; Ⅲ,
   stearic acid; Ⅳ, dihydroxyacetone phosphate; Ⅴ, dodecanoic acid; Ⅵ,
   myristic acid; Ⅶ, homocitric acid; Ⅷ, citric acid; Ⅸ, gluconic acid; Ⅹ,
   deoxyribose 1-phosphate).

2.7. scRNA-seq atlas unveil Gpx3 as a cell-type-specific biomarker in
ASIV-treated diabetic nephropathy

   Next, we used scRNA-seq analysis to investigate and verify specific
   markers in diabetic mouse kidney with ASIV treatment. Following the
   tSNE visualization ([168]Fig. 7(A)) and dimensionality reduction with
   UMAP ([169]Fig. 7(B)), single cell atlas was constructed from
   transcriptome data of db/m mice, db/db mice and diabetic mice kidney
   after ASIV treatment. Meanwhile, we further performed UMAP analysis
   with scRNA-seq samples to further confirm cell types based on the
   different cell clusters, and finally, five distinct cell types,
   including endothelial cells, epithelial cells, fibroblasts, macrophages
   and neutrophils were determined by single-cell RNA sequencing
   ([170]Fig. 7(C)). Heatmap showed the expression of key marker genes in
   bio-samples from each group and the key marker expression of db/m mice
   and ASIV treatment was closest ([171]Fig. 7(D)). Additionally, we
   observed the significant decreases in the relative abundance of key
   marker, such as Gpx3, Bpgm and Eno1 in ASIV treatment compared to db/db
   mice ([172]Fig. 7(E)). We also established the spatial feature maps
   that displayed the key marker expression pattern ([173]Fig. 7(F)). Five
   clusters in all scRNA-seq data were identified and generated
   representative UMAP maps. Then, five cell types were classified from
   the expression of key marker genes from scRNA-seq data and displayed in
   UMAP maps ([174]Fig. S6A). Importantly, each type of kidney cell
   exhibited the distinct genetic features, such as unique expression of
   Gpx3 ([175]Fig. S6B), Bpgm ([176]Fig. S6C), and Eno1 ([177]Fig. S6D),
   as well as the higher expression of Gpx3 ([178]Fig. S6E) in fibroblasts
   compared to other cell types. Of these, the specific marker Gpx3 was
   further supported from our targeted proteomics data (n = 29), as
   evidenced by comparing the protein expression level for DN patients to
   the HCs. Based on the AUC values from receiver operating characteristic
   curve of protein level, Gpx3 (AUC = 0.995) demonstrated the high
   sensitivity and specificity for DN patients ([179]Fig. 7(G)).

Fig. 7.

   [180]Fig. 7
   [181]Open in a new tab

   Single-cell RNA sequencing atlas verify specific markers in diabetic
   mouse kidney with ASIV treatment.

   (A) tSNE visualization of single cell transcriptome atlas from db/m
   mice (left), db/db mice (middle) and diabetic mice kidney after ASIV
   treatment (right). (B) Uniform manifold approximation and projection
   (UMAP) plot representation of single-cell transcriptomes from db/m mice
   (left), db/db mice (middle) and diabetic mice kidney after ASIV
   treatment (right), with different colors representing different
   clusters. (C) UMAP plot revealing the distinct cell types from diabetic
   mice kidney detected by single-cell RNA sequencing, different colors
   represent different cell types; UMAP visualization of all clusters,
   colored by all cell types. Five cell types were identified in the
   dataset. (D) Heatmap showing the expression of key marker genes
   specific to biosamples from each group. (E) Comparative expression of
   marker genes in different groups. (F) Spatial feature plots showing the
   expression pattern of selected markers for db/m mice (upper), db/db
   mice (middle) and ASIV treatment (lower). (G) Based on the receiver
   operating characteristic curve of protein expression level, the AUC
   value demonstrating that Gpx3 exhibited the high sensitivity and
   specificity for DN patients.

3. Discussion

   DN, a prevalent diabetic complication, has drawn significant attention
   due to its rising role in end-stage renal disease etiology and
   incidence ([182]Guo, Luo, & Zuo, 2024; [183]Smeijer et al., 2025).
   Emerging evidence highlights region-specific spatial features in renal
   metabolic pathways and protein functions, with distinct physiological
   roles of the cortex and medulla ([184]Li & Humphreys, 2024b). While
   multi-omics technologies enhance biological data characterization, they
   often sacrifice spatial context, limiting comprehensive analysis of
   metabolic activity and regional inconsistencies between protein and
   metabolic profiles. Integrating 3D spatially resolved multi-omics data
   (metabolome, proteome) is crucial for deciphering DN progression
   mechanisms.

   This study systematically identified DN-associated metabolic signatures
   through metabolomic screening, revealing 10 differentially abundant
   metabolites enriched in glutathione metabolism, pentose phosphate
   pathway, and glycolysis pathways. It was observed that these metabolite
   combination panels demonstrated the high diagnostic capability for DN
   patients. Incorporation of spatially resolved metabonomic data further
   demonstrated the unique spatial metabonomic characteristics of diabetic
   kidney tissue segmentation. Initially, targeted metabolomics revealed
   the significantly changes of the metabolite signals in kidney tissue
   samples from db/m mice and transgenic db/db mice. Furthermore, spatial
   metabolomics elucidate metabolic regionalization of kidney tissue
   segmentation.

   Mass spectrometry imaging (MSI)-driven spatial metabolomics uncovered
   compartmentalized metabolic signatures across anatomically defined
   kidney regions. Feature extraction and clustering algorithms identified
   region-specific metabolites: cysteinylglycine, pyroglutamic acid, and
   stearic acid in Cor zone, functionally linked to glutathione
   metabolism; dihydroxyacetone phosphate, dodecanoic acid, myristic acid,
   and homocitric acid in OM zone, spatially aligned with pentose
   phosphate pathway activity; and citric acid, gluconic acid, and
   deoxyribose 1-phosphate in IM zone, enriched in
   glycolysis/gluconeogenesis pathways. These zonal metabolic patterns
   demonstrated tight coupling between spatial localization and pathway
   functionality. Our study establishes the multimodal spatial atlas
   linking metabolic dysregulation to DN progression, revealing endogenous
   metabolic compartmentalization as a key driver of pathogenesis. By
   integrating spatially resolved metabolomics with structural analysis,
   we provide a micro-scale visualization framework for mapping metabolic
   biomarkers to tissue morphology, advancing mechanistic insights into
   how regional metabolic abnormalities underpin kidney disease.

   Spatial proteomics delineated region-specific molecular signatures
   across diabetic kidney compartments (cortex/Cor, outer medulla/OM,
   inner medulla/IM), identifying 64 differentially expressed proteins
   linked to microenvironmental features. Pathway analysis revealed
   compartmentalized functional specialization: Cor-associated proteins
   (e.g., glutathione metabolism, xenobiotic detoxification, fatty acid
   biosynthesis) aligned with redox regulation; OM-enriched networks
   (carbon metabolism, glucagon signaling, pentose phosphate pathway)
   reflected metabolic adaptation; IM-focused pathways (glycolysis, amino
   acid biosynthesis) highlighted energy reprogramming. Crucially, GPX3,
   BPGM, and ENO1 exhibited strong correlations with clinical biochemical
   markers, bridging molecular dysregulation to disease phenotype. This
   multiregional proteomic atlas not only maps compartment-specific
   metabolic derangements in DN but establishes a translational framework
   for linking spatial protein networks to dysfunction, offering
   mechanistic insights into diabetic kidney pathology.

   Further investigation is warranted to assess the therapeutic potential
   of dysregulated metabolites critical to metabolic homeostasis in DN.
   ASIV is the main active ingredient of Astragalus membranaceus, which
   has multiple pharmacological effects such as anti-inflammatory,
   antioxidant, antifibrosis and immune regulation ([185]Ding et al.,
   2025; [186]Qiu et al., 2025; [187]Zhang et al., 2025). Spatial
   metabolomics elucidated the compartment-specific metabolic remodeling
   induced by ASIV and MET treatments in diabetic mouse kidneys.
   High-resolution MALDI-MSI and PCA-based 3D visualization mapped
   treatment-modulated metabolite re-distribution, revealing distinct
   spatial regulation: ASIV preferentially restored glutathione metabolism
   in the Cor region, activated pentose phosphate pathways in OM region
   via the enriched markers, and enhanced glycolysis/glycerophospholipid
   metabolism in the IM region. MET similarly demonstrated
   region-selective metabolic modulation. These findings delineate a
   spatially resolved metabolic atlas of diabetic kidneys under
   pharmacological intervention, with ASIV exhibiting multi-compartment
   regulatory effects. Crucially, this work advances understanding of
   antidiabetic drug actions through spatially resolved metabolic mapping,
   offering a framework for targeting compartmentalized metabolic
   dysregulation in kidney disease therapeutics.

   Single-cell resolution studies are pivotal for identifying
   disease-relevant cellular phenotypes ([188]Cappuyns et al., 2024;
   [189]Fandrey et al., 2024; [190]Lambert & Jørgensen, 2024). Single-cell
   transcriptomes from db/m, db/db, and ASIV-intervention groups were
   analyzed via tSNE/UMAP clustering, defining five major renal cell
   populations: endothelial, epithelial, fibroblasts, macrophages, and
   neutrophils. ASIV treatment significantly downregulated metabolic
   markers (Gpx3, Bpgm, Eno1) in diabetic kidneys, with
   fibroblast-specific Gpx3 overexpression emerging as a hallmark.
   Clinical validation confirmed elevated GPX3 protein levels in DN
   patients versus HCs, demonstrating high diagnostic accuracy.
   Cell-type-specific expression patterns revealed Gpx3 may be a
   compartmentalized metabolic regulator. This multimodal approach not
   only deciphers ASIV's therapeutic mechanism through single-cell
   metabolic reprogramming, and establishes Gpx3 as a functional
   target-bridging spatial metabolic dysregulation with cellular
   pathophysiology.

   Integrating spatial metabolomics, proteomics, and scRNA-seq, this study
   delineates compartmentalized molecular landscapes across murine kidney
   regions in DN. Ten region-specific metabolites were identified,
   reflecting renal structural-functional specialization, with spatially
   resolved multi-omics integration exposing pathway dysregulation in DN
   progression. Cross-modal analysis linked clinical parameters to
   compartment-enriched targets, revealing therapeutic candidates.
   Single-cell resolution mapping uncovered cell-type-specific metabolic
   rewiring, while spatial proteomics highlighted GPX3 as a
   fibroblast-associated biomarker. Key findings include metabolic
   zonation patterns-glutathione metabolism in cortex, pentose phosphate
   activity in outer medulla, and glycolytic remodeling in inner
   medulla-and ASIV-induced restoration of metabolic homeostasis. This
   multimodal framework bridges micron-level metabolic redistribution to
   tissue architecture, identifying previously unrecognized
   partition-specific metabolites and actionable targets.

   This study has limitations, including a limited cohort size affecting
   spatial omics integration robustness and exclusion of renal
   substructures. While focusing on cortex/medulla zonation, molecular
   mechanisms of GPX3, BPGM and ENO1 as DN biomarkers require deeper
   functional validation. Nevertheless, our multiregional multi-omics
   atlas provides a foundational resource for decoding renal
   microenvironment complexity. The identified metabolic-protein networks
   highlight candidate diagnostic markers and therapeutic targets for DN.
   Future studies should expand anatomical coverage, validate findings in
   larger cohorts, and mechanistically interrogate pathway dysregulation.
   This spatially resolved framework advances precision strategies for
   kidney repair, bridging molecular mapping to clinical translation.

4. Conclusion

   This study developed a spatially resolved molecular atlas of diabetic
   kidneys through convergent analysis of high-throughput spatial
   multi-omics (metabolomics/proteomics) and single-cell RNA sequencing.
   Our integrative approach delineated compartment-specific metabolic
   dysfunction, identifying cortex-enriched glutathione metabolism,
   medullary pentose phosphate activity, and glycolytic remodeling as
   hallmark features of renal zonation. In addition, pathway enrichment
   and highly connected protein networks of spatial proteomics showed that
   differentially expressed proteins were primarily involved in the
   glutathione metabolism, pentose phosphate pathway and glycolysis in
   kidney tissue multiregion. Specifically, targeted proteome and clinical
   data integration identified GPX3, BPGM and ENO1 were closely related to
   clinical indicators. Our results emphasized the importance of spatial
   analysis in understanding treatment and suggested that Gpx3 is a
   promising biomarker for DN patients. This study constructed a spatial
   map of renal tissue metabolism, analyzed renal zoning metabolic
   disorders and how they are regulated, demonstrated functional metabolic
   pathways, and deepened our understanding of spatial metabolism in DN
   development. In summary, our research findings not only provide a
   comprehensive perspective for multimodal and regional renal profiling,
   but also contribute to the development of a potential metabolic related
   therapeutic strategy.

5. Experiment and method

5.1. Ethics statement

   The experiments involving human samples in this study were conducted in
   accordance with institutional ethical regulations and the Declaration
   of Helsinki and under the approval of the ethics review committee of
   First Affiliated Hospital of Hainan Medical University (Trial
   registration number: ChiCTR2300079160). Human samples from healthy
   controls (HC, n = 29) and diabetic nephropathy (DN, n = 29) patients
   were obtained from the First Affiliated Hospital of Hainan Medical
   University, collected and then stored at −80 °C. In addition, these
   patient characteristics can be presented in supporting materials
   ([191]Table S1).

5.2. Sample preparation for metabolomics

   For comprehensive metabolite profiling, we employed a methanol-based
   extraction protocol optimized for maximal metabolite coverage in
   untargeted metabolomics. Briefly, 100 μL of urine sample was mixed with
   400 μL of ice-cold methanol (final concentration: 80% methanol) and
   vortexed thoroughly. After vigorous shaking for 20 min at 4 °C,
   proteins were precipitated by centrifugation at 12,000 rpm for 10 min
   (4 °C). All procedures were performed under cold conditions (4 °C) to
   minimize metabolite degradation. The supernatant was lyophilized under
   vacuum and stored at −80 °C until LC-MS analysis.

5.3. Untargeted metabolomics analysis for human participants

   For untargeted metabolomics analysis, LC-MS (liquid chromatography-mass
   spectrometry) system comprised an Thermo Vanquish (Thermo Fisher
   Scientific, USA) system coupled to a Orbitrap Exploris 120 with ESI ion
   source. The urine sample was centrifuged at 12,000 rpm and 4 °C for 10
   min, and the supernatant was filtered through a 0.22 μm membrane and
   transferred to a bottle. Inject urine sample (2 μL) into Vanquish UHPLC
   at a volume of 0.3 mL/min. For (+) MS, solvent A consists of a 0.1​%
   formic acid aqueous solution (v/v), and solvent B consists of a 0.1%
   formic acid acetonitrile solution (v/v). Separation gradient was
   carried out: 0–1 min, 8% B2; 1–8 min, 8%–98% B2; 8–10 min, 98% B2;
   10–10.1 min, 98%–8% B2; 10.1–12 min, 8% B2. The parameters of the
   Orbitrap MS with ESI parameters were configured as follows: capillary
   temperature of 325 °C; sheath gas pressure of 40 arb; auxiliary gas
   flow rate of10 arb; ESI (−) spray voltage of −2.50 kV; ESI (+) spray
   voltage of 3.50 kV; MS1 resolution of 60000 FWHM.

5.4. Proteomics analysis of human participants

   We added an appropriate amount of protein to the final concentration of
   5 mM DTT, incubated at 37 °C for 1 h, and then return to room
   temperature. C[18] desalination column was used to desalinate the
   sample, activate the desalination column with 100% acetonitrile,
   equilibrate the column with 0.1% formic acid, wash the column with 0.1%
   formic acid to remove impurities, and finally wash with 70%
   acetonitrile. We prepared mobile phase A (100% water, 0.1% formic acid)
   and mobile phase B (80% acetonitrile, 0.1% formic acid), centrifuged at
   14,000g for 20 min at 4 °C, and injected 1 μg supernatant. ORBITRAP
   ECLIPSE mass spectrometer and optional FAIMS Pro ™ The interface was
   both used together, and the application parameters are set as follows:
   the ion spray voltage is 2.0 kV, the compensation voltage CV switches
   between −45 and −65 every 1 s, the full scan range is m/z 350–1500, the
   resolution of the secondary mass spectrum is 15,000 (200m/z), the
   resolution of the primary mass spectrum is 120,000 (m/z).

5.5. Animal experiments

   All animal procedures were approved by the Institutional Animal Care
   and Use Committee of Hainan Medical University (Protocol
   #HYLL-2023-457) and conducted in accordance with institutional
   guidelines. Seven-week-old male C57BL/KsJ mice (db/m and db/db
   genotypes) were obtained from Jiangsu Jicui Yaokang Biotechnology Co.,
   Ltd. (China; License SCXK-2023-0009) and acclimatized for one week
   prior to experimentation. Animals were randomly allocated into four
   experimental groups (n = 6/group): db/m controls (saline-treated),
   db/db model (saline-treated), Metformin (MET)-treated db/db (65
   mg/(kg·day)), ASIV-treated db/db (1 g/(kg· day)); Chengdu Lemaitian
   Pharmaceutical Technology Co., Ltd, DST230706-015). Following a 4-week
   treatment period, kidney tissues were immediately harvested, and
   immediately frozen, and stored at −80 °C for subsequent analysis.

5.6. Targeted metabolome for kidney tissue

   Targeted metabolomics analyses were performed using by ultra-high
   performance liquid chromatography system (Thermo Fisher Scientific,
   USA) and ACQUITY UPLC ® HSS T3 column, (100 mm, 1.8 μm, Milford,
   Massachusetts, USA), temperature of 40 °C, flow rate of 0.3 mL/min,
   injection volume of 2 μL. Briefly, the parameter settings in cationic
   mode are as follows: the mobile phase is 0.1% formic acid water (A2)
   and 0.1% formic acid acetonitrile (B2), and the gradient elution
   program is: 0–1 min, 8% B2; 1–8 min, 8%–98 % B2; 8–10 min, 98% B2;
   10–10.1 min, 98%–8% B2; 10.1–12 min, 8% B2. The parameters in negative
   ion mode are set as follows: the mobile phase consists of ammonium
   formate water (A3) and acetonitrile (B3), gradient elution program: 0–1
   min, 8% B3; 1–8 min, 8%–98% B3; 8–10 min, 98% B3; 10–10.1 min, 98%–8%
   B3; 10.1–12 min, 8% B3. The mass spectrometry conditions were carried
   out by Orbitrap mass spectrometer (Thermo Fisher Scientific Co.,
   America) with positive and negative ion mode electric spray ion source
   (ESI). The parameters were set as follows: sheath gas 40 arb, capillary
   temperature 325 °C, negative ion spray voltage −2.50 kV, positive ion
   spray voltage 3.50 kV, auxiliary gas 10 arb, and full scan resolution
   60,000.

5.7. Sample preparation and tissue sectioning for spatial metabolome

   Kidney tissues were equilibrated from −80 °C storage to −20 °C in a
   pre-chilled temperature-controlled chamber for 30 min prior to
   sectioning. Using a cryostat microtome, we prepared 10 μm-thick
   sections which were then thaw-mounted onto pre-chilled indium tin oxide
   (ITO)-coated glass slides. The mounted sections were immediately
   flash-frozen on dry ice and stored at −80 °C until analysis. For matrix
   application, we employed an HTX TM sprayer (Bruker Daltonics) to
   uniformly coat the tissue sections with 10 mg/mL 9-aminoacridine (9-AA)
   matrix solution prepared in ethanol:water (7:3, v/v). The matrix-coated
   slides were dried under nitrogen gas before MALDI analysis.

5.8. Mass spectrometry imaging for spatial metabolomics

   MALD-timsTOF-MSI experiments experiment was conducted on a prototype
   Bruker timsTOF flex MS system (Bruker) equipped with a 10 kHz
   intelligent beam 3D laser. Mass spectrometry was obtained in negative
   mode, and data was collected within the mass range of m/z 150–1200 Da.
   The imaging spatial resolution of the organization is set to 30 μm, and
   each spectrum consists of 400 laser exposures. Normalize each sample
   using root mean square method. Based on organizational structure,
   spatial segmentation regions are divided through feature extraction,
   similarity evaluation, and clustering algorithms. Metabolite ions were
   annotated using Metaspace, HMDB, LipidMaps, CoreMetabolome and KEGG
   databases.

5.9. Spatial proteomics for kidney tissue sections

   Initially, kidney tissue sections were stained with H&E. We carefully
   scraped the mouse kidney tissue from the circled area, cut it into
   small pieces, wash it with PBS, and then transfer it to OCT (YEASEN,
   China) and quickly frozen. H&E stained slide was set on an LCM
   microscope and marked the regions of interest in the cut sections
   collected using microtubes and segmentation system. Peptide segment was
   dissolved in the mobile phase and separate it using the Vanquish Neo
   ultra-high performance liquid chromatography. Additionally, the peptide
   segments were then separated by the ultra-high performance liquid
   chromatography, and injected the supernatant into Orbitrap MS. The ion
   source voltage operating in 1900 V, and the peptide parent ion was
   analyzed by Orbitrap detector with the scanning range of 380–980 m/z.
   The functional annotation on identified proteins were performed using
   GO, KEGG, and STRING databases. Cluster analysis of proteins with
   similar expression patterns using mFuzz method. By analyzing the fold
   change (FC) and p-value between multiple groups, protein differences
   with fold enrichment > 1.5 and p-value < 0.05 were identified.

5.10. Weighted gene co-expression network analysis (WGCNA) for spatial
proteome

   WGCNA was used to identify co-expressed gene modules and explore genes
   closely related to phenotypes. It converts the weighted neighbor
   connection matrix into a topological overlap matrix (TOM) to estimate
   network connectivity, and applies hierarchical clustering methods to
   the clustering tree structure of the TOM matrix. Based on the weighted
   correlation coefficient of genes, classify them according to their
   expression patterns, and put genes with similar patterns into a module.
   Hierarchical clustering was used to identify modules of co-expressed
   genes, and then integrate modules with characteristic gene connectivity
   to identify the modules most relevant to DN traits. Each module of the
   gene tree contains at least 50 genes, and similar modules are merged
   with a height cutoff of 0.5. Important module genes were selected for
   further analysis.

5.11. Single-cell transcriptome

   Single cells were isolated from mouse kidneys and scRNA seq libraries
   were prepared using the Chromium Single Cell 3' kit (PN-1000075)
   according to the manufacturer's protocol. Using the 10× Genomics
   platform and microfluidic technology, beads and cells with cell
   barcodes were wrapped in droplets, and the droplets containing cells
   were collected. Then, the cells in the droplets were lysed, and the
   mRNA in the cells was connected to the cell barcode on the beads to
   form single-cell GEM. Perform reverse transcription reaction in
   droplets to construct a cDNA library, and distinguish the sample source
   of the target sequence based on the sample index on the library
   sequence. Single cell transcriptome sequencing uses measured
   transcription sequences combined with UMI and cell barcoding to
   determine the absolute number of each transcription molecule within a
   single cell. Use MNN (Mutual Nearest Neighbor) dimensionality reduction
   algorithm to eliminate batch processing effects. Then, based on the MNN
   dimensionality reduction results, the UMAP (Unified Manifold
   Approximation and Projection) algorithm is used to visualize the
   clustering of single-cell populations. The clustering algorithm uses
   SNN to ultimately obtain the optimal cell clustering.

5.12. Multivariate pattern recognition analysis

   The differential molecular substances of interest were extracted
   through volcano maps. By multivariate statistical analyses,
   principal-component analysis (PCA), partial least squares discriminant
   analysis (PLS-DA) and the separation states of different samples can be
   distinguished and visualized on the MetaboAnalyst 6.0 online service
   ([192]https://www.metaboanalyst.ca/). Obtain Variable Importance
   Projection (VIP) scores from PLS-DA to identify differential molecules
   that significantly contribute to group separation. Further annotation
   of metabolites was performed using HMDB, LipidMaps, Metlin, and
   MassBank databases. Metabolites with VIP values greater than 1 and
   p-values less than 0.05 are considered differential metabolites.

5.13. Functional pathway enrichment analyses

   KEGG pathway analysis was performed metabolites with spatial metabolome
   and proteins with spatial proteomics for biological processes (BP),
   molecular functions (MF), and cellular components (CC), respectively.
   GO analysis promotes understanding of cellular functional roles, while
   the KEGG pathway provides insights into involvement in biological
   processes. Based on the metabolic expression profile dataset,
   MetaboAnalyst 6.0 was used to explore differences in biological
   processes and functional annotation and enrichment analysis. STRING
   constructing protein-protein interaction molecules.

5.14. Statistical analysis

   When appropriate, statistical significance was assessed using paired
   Student's t-tests, and p-value <0.05 were considered statistically
   significant. The logistic regression-based metabolite panel
   demonstrated superior diagnostic performance, validated by Monte-Carlo
   cross validation (MCCV)-generated ROC curves with high AUC values
   ([193]Xu & Liang, 2001), utilizing PLS for feature ranking and
   classification ([194]Wold et al., 2021).

CRediT authorship contribution statement

   Shi Qiu: Writing – review & editing, Writing – original draft,
   Resources, Methodology, Investigation, Data curation. Zhibo Wang:
   Methodology, Investigation. Sifan Guo: Methodology, Investigation.
   Dandan Xie: Methodology, Investigation. Ying Cai: Methodology,
   Investigation. Xian Wang: Methodology, Investigation. Chunsheng Lin:
   Methodology, Investigation. Songqi Tang: Validation, Resources, Project
   administration, Methodology, Investigation. Yiqiang Xie: Validation,
   Resources, Methodology, Investigation. Aihua Zhang: Writing – review &
   editing, Writing – original draft, Supervision, Resources, Project
   administration, Methodology, Investigation, Funding acquisition,
   Conceptualization.

Declaration of interests

   No conflicts to declare.

Acknowledgments