Integrated Transcriptomic and Secretome Proteomic Analysis of Hyperglycemia-Stimulated Endothelial Cells and Screening of Target Compounds

Diabetes mellitus, as a global health challenge, affected 529 million people worldwide in 2021, with projections indicating it will impact 1.31 billion people by 2050¹. Beyond glycemic control, long-term complications are the primary source of morbidity, mortality, and reduced quality of life. These include microvascular complications (e.g., diabetic kidney disease (DKD), retinopathy, neuropathy) and macrovascular complications (e.g., accelerated atherosclerosis). Critically, DKD affects 30%-40% of diabetic patients, representing the leading cause of end-stage renal disease (ESRD) worldwide². The profound burden of these complications underscores an urgent unmet need for novel therapies targeting their underlying mechanisms.

Endothelial cells (ECs), forming the inner lining of all blood vessels, are pivotal gatekeepers of vascular health. Hyperglycemia-induced endothelial dysfunction acts as a central driver of diabetic vasculopathy and organ damage^3,4. High glucose levels trigger a cascade of maladaptive responses within ECs, encompassing dysregulated nitric oxide (NO) bioavailability, increased endothelin-1 (ET-1) secretion, upregulated adhesion molecule expression (e.g., VCAM-1, ICAM-1), heightened oxidative stress through excessive reactive oxygen species (ROS) generation (e.g., via NADPH oxidases), and persistent low-grade inflammation^3,4. Collectively, these alterations drive impaired vasodilation, increased vascular permeability, leukocyte adhesion and infiltration, pro-thrombotic states, and ultimately, vascular remodeling⁵. Within the renal microvasculature, glomerular endothelial damage disrupts the filtration barrier, promotes albuminuria, and contributes directly to extracellular matrix (ECM) accumulation, glomerulosclerosis, and tubulointerstitial fibrosis^5,6. The subsequent activation of resident renal cells, coupled with the influx of inflammatory cells, establishes a vicious cycle amplifying renal injury^5,7. Crucially, clinical studies demonstrate that markers of systemic endothelial dysfunction correlate strongly with the onset and progression of albuminuria and declining glomerular filtration rate (GFR) in diabetic patients^6,7,8, underscoring its direct relevance to human disease outcomes. Despite its recognized centrality, a comprehensive, systems-level understanding of the precise transcriptional and secretory reprogramming induced by chronic hyperglycemia in ECs-especially alterations driving ECM remodeling, key to renal fibrosis-remains incompletely characterized, limiting the identification of novel, mechanistic therapeutic targets for DKD and other diabetic vascular complications^7,8.

The profound molecular complexity underpinning endothelial dysfunction in diabetes presents a formidable challenge to traditional single-target therapeutic approaches. This is where computational biology and bioinformatics emerge as indispensable tools, enabling the integration and mining of diverse, high-dimensional omics datasets to prioritize causal hubs within pathological networks and identify multi-targeting therapeutic strategies⁹. Critically, conventional single-modality approaches (e.g., transcriptome/proteome-only or pharmacological screening) often miss critical interplay between molecular layers (e.g., gene expression, protein dynamics, drug response), failing to capture synergistic drivers or off-target effects^10,11. The integrated multi-omics pipeline overcomes these limitations by concurrently profiling transcriptomic, secretomic, and pharmacologic data within the same system, yielding a holistic disease mechanism view. Network pharmacology, ligand/target prediction algorithms, and in silico molecular docking allow for the systematic exploration of approved drug libraries, accelerating the repositioning or discovery of compounds capable of counteracting complex disease signatures³. This paradigm shift holds particular promise for identifying therapies targeting endothelial dysregulation and its devastating microvascular consequences, such as DKD, offering the potential for improved efficacy and reduced side effect profiles. Integration of transcriptomic/secretomic analyses of hyperglycemia-exposed endothelial cells identified 534 commonly upregulated molecules. Functional enrichment implicated extracellular matrix remodeling, oxidative stress, and inflammation in early DKD pathogenesis. Cross-database analysis prioritized 278 high-confidence targets, refined via PPI networks of hub genes. Network pharmacology screening of approved drugs predicted 85 candidate compounds (including brefeldin-A, bruceantin, paracetamol, and idelalisib) targeting these hubs. In silico transcription factor prediction and docking explored mechanisms. This work delineates endothelial drivers of DKD, establishes a computational discovery pipeline, and provides therapeutic candidates for validation. Ultimately, this pipeline is anticipated to promote therapeutic discovery programs for diabetic complications.

The animal experiments were conducted with the approval of the Committee on Animal Research and Ethics of the Hebei Yiling Medical Research Institute (approval number: N2024082). Check the Table of Materials for the experimental materials and instruments used in this study. Personnel must strictly wear appropriate personal protective equipment (PPE) when handling hazardous reagents such as TRIzol (skin/eye irritant, contains phenol) and protease inhibitors (potential respiratory/skin sensitizers). Segregate biological/liquid waste in autoclavable bags, and hazardous chemical waste in designated containers for institutional hazardous disposal. Decontaminate consumables before disposal.

Diabetic nephropathy mice modeling

The male db/db and db/m mice (12 weeks old) were maintained under specific pathogen-free (SPF) conditions at 22 ± 2 °C and 56% ± 5% humidity with 12 h light/dark cycles and ad libitum access to food/water. After a 1 week acclimatization period, the mice were randomly assigned numerical identifiers and allocated to experimental groups. The db/db mice were fed a high-protein diet for 6 weeks to establish DKD. Successful modeling was confirmed by a significantly elevated urinary albumin-to-creatinine ratio (UACR) versus db/m controls, with UACR >30 mg/g serving as the primary functional biomarker for early DKD¹².

High-glucose endothelial cell modeling

Human Renal Glomerular Endothelial Cells (HRGECs) were prepared under standardized conditions. HRGECs were cultured in either normal glucose (NG, 5.5 mM D-glucose) or high glucose (HG, 30 mM D-glucose) medium and maintained in a humidified incubator at 37 °C with 5% CO₂. Cells were either exposed uninterrupted to HG or grown in control conditions for 24 h prior to downstream analyses. Osmotic control (HM, containing 5.5 mM D-glucose with 24.5 mM mannitol) should be included. All conditions should meet strict cell fitness criteria: viability remained ≥85%, apoptosis levels stayed below 15%, ROS increases in HG compared to HM did not exceed 20%, and LDH release remained under 10%, validating cellular integrity prior to downstream analyses.

Conditioned media collection

Conditioned media (CM) were collected using a standardized protocol for secretome analysis¹³ with modifications. After 24 h of high glucose (30 mM D-glucose), normal glucose (5.5 mM D-glucose), or osmotic control (5.5 mM D-glucose + 24.5 mM mannitol) stimulation, HRGECs were washed with sterile phosphate-buffered saline (PBS) to remove residual serum proteins. Cells were replenished with serum-free, phenol red-free endothelial basal medium and incubated for 12 h at 37 °C with 5% CO₂ to accumulate secreted factors.

CM was harvested using pre-chilled centrifugation tubes. Sequentially, sample processing was done as described below.

Primary clarification: CM was transferred to RNase/DNase-free conical tubes and centrifuged at 3,000 x g for 10 min at 4 °C. The pelleted debris was discarded.

Protease inhibition: Within 2 min of primary clarification, EDTA-free protease inhibitor cocktail was added to a 1x final concentration (1:50) containing 1x phosphatase inhibitors.

Concentration: Supernatant was immediately loaded into pre-rinsed PBS centrifugal filters (3 kDa MWCO) and centrifuged at 4,000 x g for 90 min at 4 °C to achieve a 10x concentration.

Secondary clarification: Concentrate was transferred to pre-cooled microcentrifuge tubes and centrifuged at 12,000 x g for 15 min at 4 °C. Supernatant was collected, avoiding pelleted aggregates.

Storage: Aliquots of 50 µL volumes were made in sterile low-protein-binding cryovials. The vials were flash-frozen in liquid nitrogen within 30 min of harvest completion. Vials were stored at -80 °C (no freeze-thaw cycles).

CM batches underwent endotoxin testing (<0.25 EU/mL). Serum-free incubation protocol showed no stress induction (validated by HSP70/CHOP immunoblots). CM protein yield normalized to viable cell count/DNA content at harvest. Processing rigor maintained: sterile technique throughout; ≤30 min harvest-to-freeze window; rotor temperature equilibration confirmed.

CCK-8 assay of conditioned media-treated renal tubular cells

To evaluate the functional impact of the endothelial secretome on renal tubular cell viability, a CCK-8 assay was performed on the Human renal proximal tubular epithelial cell line HK-2 treated with CM derived from human renal glomerular endothelial cells (HRGECs), in accordance with established protocols. HK-2 cells were cultured in DMEM medium supplemented with 10% fetal bovine serum (FBS). For the assay, cells were seeded into 96-well plates at a density of 5 x 10³ cells per well. Serum-starvation of cells was done for 12 h in medium containing 0.5% FBS immediately prior to CM treatment.

CM was applied with equal total normalized protein mass. Frozen CM aliquots were thawed at 37 °C in a dry bath (<5 min) and clarified by centrifugation at 10,000 x g for 5 min at 4 °C. HK-2 cells (100 µL/well) were treated with HG-CM, NG-CM, or HM-CM, normalized to matched protein/DNA content. A total of 3 biological replicates per condition (with 3 technical replicates each) were prepared. Samples were incubated at 37 °C with 5% CO₂ for 24 h. Cell viability was assessed using the Cell Counting Kit-8. To do this, 10 µL of CCK-8 solution was added to each well and incubated for 1-4 h at 37 °C. Absorbance was measured at 450 nm using a microplate reader. Viability was calculated as:

Viability (%) = [(A_sample - A_blank) / (A_baseline_control - A_blank)] x 100

Oxidative stress detection

Cellular malondialdehyde (MDA) levels were quantified, one of the main end products of membrane lipid peroxidation, via the TBA assay. Cell lysates (1 x 10⁷ cells) were clarified by centrifuging at 10,000 x g for 15 min, and then 0.02 mL supernatant was aliquoted into sample tubes alongside 0.02 mL of MDA standards. To the aliquot, 0.02 mL of clarificant and 0.6 mL of acid reagent were added. The samples/standards were treated with 0.2 mL of the chromogenic agent (control tubes: 0.2 mL of 50% acetic acid was added). After sealing, mixing, and incubating at 100 °C for 40 min, the samples were cooled and then centrifuged at 9569 x g for 10 min. The supernatant was transferred to microplates at 250 µL for OD₅₃₂ measurement.

Transcriptomic profiling

Total RNA was isolated using TRIzol reagent following the manufacturer's protocol. RNA quantity and purity were assessed with a spectrophotometer and fragment analyzer, respectively. Only high-quality RNA samples were used with RNA Integrity Number (RIN) > 7.0 for sequencing library construction.

Polyadenylated (poly(A)+) mRNA was enriched by two rounds of oligo(dT) magnetic bead-based selection. The purified mRNA was fragmented to 200-300 nucleotides using a magnesium-based fragmentation kit at 94 °C for 5-7 min. First-strand cDNA synthesis was done with reverse transcriptase, followed by second-strand cDNA synthesis using E. coli DNA polymerase I and RNase H in the presence of dUTP (to enable strand specificity). Subsequent steps included end repair, A-tailing, adapter ligation, and size selection using magnetic beads. Prior to PCR amplification, the dUTP-incorporated second strand was digested with UDG enzyme.

A strand-specific library was constructed with an average insert size of 400 ± 50 bp using PCR under the following conditions: initial denaturation at 98 °C for 1 min; 14 cycles of denaturation at 98 °C for 10 s, annealing at 60 °C for 30 s, and extension at 72 °C for 30 s; and a final extension at 72 °C for 5 min. Paired-end 150 bp sequencing was done on an Illumina platform, with a sequencing depth of ≥40 million reads per sample (Q30 > 85%).

Bioinformatic analysis of transcriptomic data

Quality check of the raw reads was done using FastQC (v0.11.8), and alignment to the GRCh38.p13 reference genome with HISAT2 (v2.2.1) was done. Transcript quantification and per-sample transcript assembly were performed using StringTie (v2.1.6) with default parameters. All assembled transcriptomes were merged from individual samples to reconstruct a comprehensive transcriptome using gffcompare software (v0.12.6; gffcompare is a standalone tool, not part of StringTie, and its version is corrected to the common stable release). Differential expression analysis was done with DESeq2 (v1.38.3) using thresholds of |log2 fold change (FC)| > 1 and false discovery rate (FDR) < 0.05. Batch effects were corrected via the variancePartition package.

Secretome proteomic analysis

CM was collected from HRGECs cultured under NG, HG, or HM control conditions using a previously described method¹³ with modifications for secretome analysis. Cells were washed with PBS after stimulation and incubated in serum-free medium for 12 h. CM was collected and centrifuged at 3,000 x g for 10 min (4 °C).

Equal amounts of peptide from each sample were mixed and then diluted with solvent A (5% ACN, pH 9.8) and injected into the column. Fractionation of the peptide mixture was done using a 3.5 µm 4.6 x 150 300Extend- C18 column on the Binary Rapid Separation System. Gradient elution was performed at a flow rate of 0.3 mL/min: 5% to 21% solvent B (97% ACN, pH 9.8) in 38 min, 21.5% to 40% solvent B in 20 min, 40% to 90% solvent B in 2 min, 90% solvent B for 3 min, and 5% solvent B equilibrated for 10 min. Elution peaks were monitored at 214 nm, and fractions were collected every minute. Fractions were combined according to chromatograms of the elution peaks. A total of 10 fractions were collected, which were then freeze-dried.

Mobile phases A (100% water, 0.1% formic acid) and B (80% acetonitrile) were prepared, and the dried peptide samples were reconstituted with 0.1% formic acid and centrifuged at 20,000 x g for 10 min. Separation was done using a UHPLC liquid phase system. The samples were fed onto an ES906 HPLC column (150 mm) with an 8-min gradient at 2.5 µL/min, and separated with the following effective gradient: 0-4 min, 4% mobile phase B linearly ramping up to 25%; 4-6.9 min, mobile phase B linearly ramping up from 25% to 35%; 6.9-7.3 min, mobile phase B linearly ramping up from 35% to 99%; 7.3-8.0 min, maintained mobile phase B at 99%. The liquid-phase separated peptides were transferred to a mass spectrometer for DIA mode acquisition. The main parameters were set as follows: normalized collision energy 25%, default charge state 2, resolution 240,000, scanning every 0.6 s, 380-980 m/z scanning range, mass spectrometry AGC 500%. Fragmentation ion scans were recorded with a maximum scan time of 3 ms and utilized 300 2-Th scanning windows, ranging from 380 to 980 m/z.

The DIA-NN (https://www.nature.com/articles/s41592-019-0638-x, version 1.9.1) was used to analyze the DIA data with a library-free method. MS/MS data were searched against protein sequences, which were downloaded from Uniprot database with the following settings: enzyme: Trypsin/P; maximum missed cleavages: 2; fixed modification: carbamidomethyl (C); variable modifications: oxidation (M) and acetyl (protein N-term); precursor mass tolerance: 20 ppm; fragment mass tolerance: 0.05Da. Results were filtered by 1% FDR, and only those protein groups were used that passed this filter criterion in downstream analysis.

Omics functional enrichment analysis (GO, KEGG, and GSEA)

Functional enrichment analyses were conducted on filtered sets of differentially expressed genes (transcriptomics) and proteins (secretome proteomics), following established bioinformatics pipelines. Identifiers were mapped using Org.Hs.eg.db. GO enrichment (biological processes, molecular functions, cellular components) was executed via clusterProfiler. Benjamini-Hochberg adjusted p-value < 0.05, and semantic similarity reduction (cutoff=0.7) was applied to condense terms. KEGG pathway analysis was performed using KEGG REST API (hsa, 2023 release). Hypergeometric testing was utilized with Yekutieli FDR correction. Pathway topology visualization was done with Pathview. GSEA was applied using a rank-based approach with signal-to-noise gene ranking. Test enrichment against MSigDB collections (HALLMARK, C2, C5; v2023.2) and curated diabetic endothelial signatures was done. Significance was assessed after 1000 phenotype permutations (FDR < 25%).

Protein-protein interaction network analysis

To identify core molecular regulators within the pathogenic candidate pool, PPI network construction and topological analysis were performed using an integrated computational workflow. The 278 candidate genes were submitted to the STRING database (v12.0; Homo sapiens; evidence sources included experimental, co-expression, and curated databases; an interaction confidence (combined score) threshold >0.7 for high-confidence edges; and no additional interactor inflation. The results were imported into Cytoscape (v3.9.1), and network topology analysis was performed using MCODE (v1.5.2) and CytoHubba (v0.4.1) plugins. Subsequently, the network was visually optimized according to the degree of connection of nodes, so that nodes with high connectivity are darker in color, larger in size, and more prominent in label.

Development of a diabetic kidney disease target gene set

The criteria for co-upregulated mRNAs/proteins were defined as follows: transcriptome (FDR < 0.05 and |log2FC| > 1) and secretome (FDR < 0.01 and |log2FC| > 1.5). Protein and gene names were normalized using UniProt and converted to a unified format (Gene Symbol). A target gene set was constructed for diabetic kidney disease by integrating disease-associated genes from multiple public databases, including GeneCards (Version 5.25, query term: Diabetic Nephropathies, accessed date July 2025), the Comparative Toxicogenomics Database (CTD; https://ctdbase.org/, query term: Diabetic Nephropathies, accessed date July 2025), and Open Targets Platform (https://platform.opentargets.org/, Version 0.13.8, query term: Diabetic Nephropathies, accessed date July 2025)^13,14,15. This was designated as the comprehensive diabetic kidney disease reference set. The intersection between co-upregulated mRNAs/proteins and this gene set was identified for downstream analyses.

CCL2 protein detection

To quantify CCL2, a 96-well plate was coated with capture antibody (100 µL/well) and incubated at 4 °C overnight. The next day, the plate was washed with 2x with buffer and blocked with ELISA diluent (200 µL/well) for 1 h at RT. A 7-point standard curve was prepared by 2-fold serial dilutions of the S1 standard, then the samples (100 µL/well) were added and incubated for 2 h at 400 rpm. After washing 5x, detection antibody (100 µL/well) was added and incubated for 1 h at 400 rpm, followed by enzyme solution (100 µL/well) addition and incubation for 30 min at 400 rpm. The samples were developed with TMB substrate (15-30 min, RT). The reaction stopped with acid (100 µL/well), and the absorbance was read at 450 nm (620 nm reference optional). Key steps include stringent washing, timed incubations, and standardized dilutions to ensure reproducibility.

Drug repositioning

Therapeutic compounds were predicted using the following platform: LINCS L1000 Characteristic Direction Signatures Search Engine (L1000CDS2; https://maayanlab.cloud/L1000CDS2). This platform employs the MODZ method and characteristic direction analysis to identify small molecules or compound combinations capable of reversing or mimicking input gene expression signatures¹⁶. Candidate drugs were identified as those that reverse the dysregulation of differentially expressed renal genes using this platform. The compounds were ranked based on their reversal scores in renal cell lines by select kidney in the tissue column, and the top-ranked candidates were subjected to molecular docking to evaluate their binding affinities to the key protein targets.

Co-regulatory transcription factor screen

The hTFtarget database was queried for co-regulatory TFsbinding to the gene set. hTFtarget database was analyzed to identify human transcription factor (TF)-target gene relationships by computationally analyzing large-scale ChIP-Seq datasets across diverse cells, tissues, and conditions. The integrative pipeline combined ChIP-Seq peak signals with epigenetic context to pinpoint TF binding in promoters/enhancers, explicitly accounting for spatiotemporal regulatory dynamics. As shown by Zhang et al., it served as a multidimensional platform enabling exploration of TF targets or gene regulators, visualization of ChIP-Seq data, investigation of TF cooperativity, and prediction of binding sites¹⁷.

Computational pipeline and analysis for small molecule-protein docking

AutoDock Vina 1.2.0¹⁸ was used to conduct molecular docking analysis of binding interactions between active ingredients and the transcription factors BRD4, CTCF, EP300, and SPI1. The crystal structures of BRD4 (PDB ID: 7REK), CTCF (PDB ID: 5K5I), EP300 (PDB ID: 5NU5), and SPI1 (PDB ID: 8E3K) were obtained from the Protein Data Bank (https://www.rcsb.org/)^19,20. Small-molecule structures in SDF format were downloaded from the PubChem database (https://pubchem.ncbi.nlm.nih.gov/²¹, converted to MOL2 format using Open Babel 2.4.1, and energy-minimized with PyRx 0.8. Protein preprocessing was done by removing water molecules, adding hydrogen atoms, calculating Gasteiger charges, and converting them to PDBQT format (required for AutoDock Vina). A semi-flexible docking strategy was employed, defining the docking grid position and dimensions based on the binding sites of co-crystallized ligands in each protein. MA9-086 serves as the positive-control ligand for BRD4, and XDM-CBP for EP300. For CTCF and SPI1 (lacking known positive ligands), potential binding pockets were identified via structural analysis using the Proteins Plus online platform (https://proteins.plus/). Docking grid parameters were set as follows (coordinates and dimensions in Å along the x, y, and z axes): BRD4 center at (-11.907, -8.617, -3.973) with a box size of (18.387, 18.387, 18.387); CTCF center at (15.165, 4.854, 6.388) and box size (17.25, 20.25, 9.75); EP300 center at (39.781, 5.326, 9.998) and box size (16.516, 16.516, 16.516); SPI1 center at (3.155, -3.853, 0.668) and box size (17.25, 25.5, 10.5). During docking, the energy range was set to 3, the exhaustiveness parameter to 8, and the grid spacing to 0.375 Å. Binding stability was assessed from binding energy, where lower values indicate more stable conformations and higher interaction probability, with a threshold of <-5 kcal/mol defined as strong binding activity^22,23. Finally, interaction modes between compounds and receptors were identified using PyMOL.

Statistical analysis

Data is presented as mean ± SEM. Statistical tests were selected based on normality (Shapiro-Wilk) and variance homogeneity (Levene's test). Unpaired t-tests (two groups) or one-way ANOVA with LSD post hoc tests (≥3 groups) were used for between-group comparisons. A p < 0.05 was defined as a significance threshold. Graphs were generated with GraphPad Prism 9.0.

Demonstration of the pathogenic impact of the hyperglycemic endothelial secretome on renal tubules

This study commenced with the validation of the pathogenic effect of the hyperglycemic endothelial secretome. Conditioned medium from hyperglycemic human glomerular endothelial cells (HG-CM) significantly impaired the viability of human proximal tubule epithelial cells (HK-2) compared to the control groups.

Building upon this functional evidence, integrated diabetic kidney transcriptomic and high glucose-treated endothelial secretome analyses (LC-MS/MS of HG-CM) were performed. This identified a core set of concordantly upregulated pathogenic genes and their corresponding secreted proteins. Leveraging these common pathogenic targets, L1000CDS2 compound libraries were computationally screened for agents predicted to functionally downregulate their expression or activity. Concurrently, utilizing transcription factors (TFs) databases identified key TFs regulating these pathogenic targets. Computer-aided screening via molecular docking was performed to evaluate the binding potential of candidate repurposed compounds against the pathogenic TFs. Finally, validation of the expressions of TFs' downstream pathogenic genes should be performed by qPCR (for mRNA levels) and ELISA (for secreted protein products), confirming the regulatory effects of both prioritized compounds and identified transcription factors on the pathogenic cascade (Figure 1).

Integrated transcriptomic and secretome proteomic analysis reveals consistently upregulated pathways in hyperglycemia-stimulated endothelial cells

The detrimental impact of hyperglycemia-stimulated endothelial cell secretome on renal tubular epithelial cells was first validated (Figure 2A). CCK-8 assays and MDA detection demonstrated that HK-2 cells exposed to HG-CM exhibited significantly reduced viability and increased MDA level compared to normal glucose medium and osmotic control medium, confirming the cytotoxicity of hyperglycemic endothelial secretome (Figure 2B,C).

To elucidate the molecular mechanisms underlying this effect, multi-omics analyses were performed. Principal component analysis (PCA) of transcriptomic data demonstrated clear separation between hyperglycemic and control endothelial cells, confirming the robustness of the model (Figure 3A). Differential gene expression analysis revealed extensive transcriptomic alterations in hyperglycemic endothelial cells, with a volcano plot displaying numerous significantly upregulated and downregulated genes (Figure 3B) and a heatmap further visualizing clustered expression patterns of these dysregulated genes (Figure 3C and Supplementary Table 1). Parallel proteomic analysis of endothelial secretomes identified dysregulated secreted proteins in HG-CM (Figure 3D). Integrative analysis uncovered 534 consistently upregulated genes/proteins at both transcript and secreted protein levels (Figure 3E).

Functional enrichment analysis of the core molecules established their convergent involvement in pivotal disease-promoting pathways. Gene Ontology (GO) analysis demonstrated significant enrichment in biological processes critical to diabetic nephropathy pathogenesis: Collagen-containing extracellular matrix, reflecting pathological ECM deposition driving renal fibrosis, inflammatory response, characterized by macrophage infiltration and TNF-α-mediated renal injury amplification, positive regulation of cell migration, enabling monocyte infiltration into renal interstitium; and extracellular exosome, implicated in pathological signal transduction via TGF-β and other profibrotic cargoes (Figure 3F). KEGG pathway analysis further corroborated involvement in core pathways: the AGE-RAGE signaling pathway in diabetic complications, TNF signaling pathway, and HIF-1 signaling pathway, all converging on HIF-1α-mediated amplification of fibrosis, aberrant angiogenesis, and inflammatory renal damage (Figure 3G). GO enrichment and Gene Set Enrichment Analysis (GSEA) substantiated these findings through enrichment in the regulation of actin cytoskeleton, where podocyte cytoskeletal disorganization impairs glomerular filtration barrier integrity, leading to proteinuria, TNF signaling pathway, propagating oxidative stress and inflammatory tissue destruction (Figure 3H-J).

Identification and validation of hub pathogenic mediators in diabetic nephropathy

To identify endothelial-derived mediators of hyperglycemic renal injury, transcriptomic and proteomic datasets were integrated via intersection of 534 commonly upregulated endothelial genes defined by modality-specific thresholds (transcriptome: FDR<0.05 and |log2FC|>1; secretome: FDR<0.01 and |log2FC|>1.5) with established diabetic nephropathy-associated genes from public databasesi ncluding GeneCards, the Comparative Toxicogenomics Database, and Open Targets Platform, which yielded 278 high-confidence pathogenic candidates (Figure 4A). Gene-protein mapping used UniProt identifiers with maximal fold-change resolution for one-to-many relationships.Protein-protein interaction (PPI) network analysis identified critical molecular hubs, including GAPDH, FN1, CD44, ITGB1, ENO1, ICAM1, SERPINE1, CCL2, CAV1, VWF, APOE, ANXA2, THBS1, VCAM1, CDC42, FLNA, and NFKB1 (Figure 4B), all exhibiting extensive connectivity in endothelial-to-renal signaling. Previous studies suggested that VWF (von Willebrand Factor) drives thrombotic microangiopathy and endothelial inflammation in diabetic glomeruli24,25; CCL2 (C-C motif chemokine ligand 2) recruits monocytes/macrophages to propagate renal inflammation26; SERPINE1 (plasminogen activator inhibitor-1) stimulates fibrotic pathways through ECM accumulation27; and THBS1 (thrombospondin-1) activates TGF-β-dependent matrix remodeling during injury repair28. Transcriptomic and proteomic validation quantified their significant hyperglycemia-induced dysregulation: VWF mRNA increased 2.2-fold with 6.06-fold protein upregulation (Figure 4C,D); CCL2 mRNA rose 1.23-fold alongside 2630-fold secreted protein elevation (Figure 4E,F and Supplementary Table 2); SERPINE1 exhibited2.52-fold transcriptional induction with concordant 3.32-fold protein upregulation (Figure 4G,H); and THBS1 showed 1.15-fold mRNA enhancement with 1.54-fold protein induction (Figure 4I,J). ELISA validation demonstrated that hyperglycemia markedly elevated CCL2 protein levels in human glomerular endothelial cells (Figure 4K), and treatment of HK2 cells with this conditioned medium impaired cell viability-an effect rescued by 20 nM of CCR2 inhibitor (Figure 4L). These data collectively confirm that these effectors may function as central mediators of hyperglycemic renal injury.

Drug repositioning screen identifies compounds reversing pathogenic signatures

A systematic query of the transcriptomic profiles of compounds in the L1000CDS2 database of mammalian kidney cells was performed to identify candidates capable of reversing the hyperglycemia-induced pathogenic gene signatures. Using the 278 pathogenic mediators identified in diabetic nephropathy (Figure 4B) as input, 33621 clinically used compounds were screened through a multi-step bioinformatic pipeline (Figure 5A). This analysis revealed 85 compounds that effectively downregulated the collective pathogenic signature (Figure 5B). Among these, paracetamol, idelalisib (both are FDA-approved drugs)29,30, brefeldin-A, and bruceantin (both are experimental molecules)31,32 demonstrated exceptional efficacy, reversing expression of pathogenic targets in kidney cells (Table 1, Table 2). This computational repositioning framework prioritizes clinically translatable compounds capable of simultaneously targeting multiple convergent pathways in diabetic nephropathy.

Transcriptional regulators and computational docking reveal candidate compound mechanisms

To delineate mechanisms underlying pathogenic gene dysregulation, interrogation of transcriptional control networks orchestrating the 278 hyperglycemia-upregulated targets led to the identification of several master transcription factors potentially regulating these pathogenic genes, including CTCF, SPI1, EP300, and BRD4 (Figure 6).

Computational docking was performed to characterize potential interactions between the top repurposed compounds and identified transcription factors. Virtual screening suggested computationally identified affinity binding between the molecules and the TFs (Figure 7A). Idelalisib showed predicted binding to CTCF and BRD4 with docking scores of -5.9 kcal/mol and -5.2 kcal/mol, respectively; Bruceantin showed predicted binding to SPI1 with a docking score of -7.4 kcal/mol; Paracetamol potentially binds to EP300 with a docking score of -4.9 kcal/mol (Figure 7A). Structural analyses revealed hypothesized binding modes between the predicted compounds and their respective transcription factors. Idelalisib was modeled to form potential H-bonds with hydrogen bonds with CTCF at Thr417, Arg448, and Thr421, along with van der Waals interactions with Pro375, Met442, Trp374, Val439, Glu438, and Leu387 of BRD4 (Figure 7B,C). MA9-086, a BRD4 inhibitor, exhibited a docking score of -8.7 kcal/mol with BRD4. It formed conventional hydrogen bonds with Pro82 and Asn140, a carbon-hydrogen bond with Tyr97, and engaged in van der Waals interactions with Trp81, Met149, and Asp145, as well as alkyl interactions with Leu92, Phe83, Ile146, Val87, and Cys136 (Figure 7D). Bruceantin established hydrogen bonds with SPI1 at Asn219, Lys217, Trp213, Arg220, Lys221, and Lys227, and van der Waals contacts with Ile170, Asn234, Met223, and Arg230 (Figure 7E). Paracetamol interacted with EP300 via a hydrogen bond with Ile178 and van der Waals forces with Lys182, Leu120, and Leu124 (Figure 7F). XDM-CBP, an EP300 inhibitor, had a docking score of -7.3 kcal/mol with EP300. Its binding was stabilized by conventional hydrogen bonds with Asn1132, a carbon-hydrogen bond with Ser1136, van der Waals interactions with Arg1133, Tyr1131, Pro1074, Ile1086, and Tyr1089. Additionally, it formed a pi-sigma interaction with Val1079, alkyl interactions with Leu1084, Val1138, and Ala1128, and a pi-pi T-shaped interaction with Phe1075 (Figure 7G). These modeling results mechanistically explain compound efficacy in reversing pathogenic signatures through transcriptional hub disruption.

Integrative computational analytical pipeline

The representative results presented here collectively demonstrate the application and validation of the multi-layered computational pipeline for deciphering hyperglycemia-induced endothelial pathobiology and identifying repurposed therapeutic candidates. The workflow is functionally anchored by initial phenotypic validation, confirming the cytotoxic impact of the hyperglycemic endothelial secretome on renal tubules (Figure 2). This established a direct link between endothelial stimulation and tubular injury, providing the biological rationale for subsequent omics interrogation.

The core of the technique lies in the integrated transcriptomic and secretomic analysis (Figure 3),which moves beyond single-modality approaches to identify a high-confidence set of 534 concordantly upregulated genes/proteins. This integrated dataset serves as the critical input for downstream computational mining. The subsequent steps, functional enrichment analysis, cross-database validation to derive 278 high-confidence mediators, and PPI network topology analysis to pinpoint central hubs like CCL2 and SERPINE1 (Figure 4), exemplify how the pipeline transforms raw multi-omics data into a refined, disease-relevant molecular network. The strong correlation between transcriptional and secretory levels for key hubs (e.g., CCL2) validates the biological coherence of the identified targets.

The utility of this network for therapeutic discovery is then showcased. The systematic screening of compound libraries against the pathogenic signature (Figure 5) transitions the pipeline from mechanistic mapping to translational hypothesis generation, identifying candidates like idelalisib and paracetamol. Finally, the integration of transcription factor analysis and computational docking (Figure 6, Figure 7) provides a mechanistic hypothesis layer, suggesting how these compounds might exert their effects by targeting key regulatory nodes (e.g., CTCF, BRD4). The docking scores and predicted binding modes offer a structural rationale for the prioritized compounds, which should be interpreted as in silico evidence supporting their potential as network disruptors. To analyze the outcome, the docking scores and interaction profiles should be compared against known active inhibitors (e.g., MA9-086 for BRD4) as internal positive controls, while the consistency between the compound's known mechanism (e.g., idelalisib as a PI3Kδ inhibitor) and its newly predicted TF targets may reveal novel polypharmacology.

In summary, these sequential results validate the pipeline's capacity to connect an initial pathological stimulus (hyperglycemia) to a dysfunctional secretory phenotype, map the underlying molecular network, prioritize key mediators, and computationally repurpose drugs with hypothesized mechanisms of action, thereby establishing a reusable framework for endothelial-centric disease investigation and drug discovery.

Data availability:

The mass spectrometry proteomics data have been deposited to the ProteomeXchange Consortium (https://proteomecentral.proteomexchange.org) with the dataset identifier PXD066579 (upload date 25.07.2025, Verification status: Submit). The sequencing data is stored in the Sequence Read Archive (SRA, https://submit.ncbi.nlm.nih.gov/subs/sra/) with accession number PRJNA1303041 (upload date 08.08.2025, Verification status: Submit).

Figure 1: Schematic overview of the integrated multi-omics and drug screening workflow. Experimental pipeline: 1) Parallel transcriptomic (RNA-seq of hyperglycemic glomerular endothelial cells) and secretome proteomic (LC-MS/MS of HG-CM) profiling; 2) Identification of concordantly upregulated pathogenic mediators via transcript-protein integration; 3) Computational screening of FDA-approved compounds predicted to reverse the 278-gene pathogenic signature; 4) Bioinformatic identification of master transcription regulators (CTCF/SPI1/EP300/BRD4); 5) Structure-based screening via molecular docking of prioritized compounds against functionally validated TFs; 6) Validation of the expressions of TFs' downstream genes by qPCR/ELISA. Please click here to view a larger version of this figure.

Figure 2: Cytotoxic effect of hyperglycemic endothelial secretome on renal tubules. (A) Experimental design for modeling hyperglycemic secretome-induced renal injury. (B) CCK-8 assay demonstrates significantly impaired HK-2 cell viability following HG-CM exposure vs. NG-CM controls and osmotic controls. n=3. Data analyzed by one-way ANOVA. (C) MDA detection demonstrates significantly increased cellular oxidative stress levels following HG-CM exposure vs. NG-CM controls and osmotic controls. n=3. Data analyzed by one-way ANOVA. Error bars represent the mean ± standard error of the mean (SEM).All data presented as mean ± SEM. *p<0.05, **p<0.01, ***p<0.001. Please click here to view a larger version of this figure.

Figure 3: Multi-omics characterization of hyperglycemia-dysregulated endothelial pathways. (A) Principal component analysis (PCA) of transcriptomic data shows a clear separation between the two groups. (B) Volcano plot of differential gene expression in DN kidney. (C) Heatmap of significantly regulated genes in DN kidney with the color bar indicating the relative expression level (log2 fold change).(D) Heatmap of dysregulated secreted proteins in HG-CM with the color bar indicating the relative expression level (log2 fold change). (E) Venn diagram identifying 534 consistently upregulated entities at transcript/protein levels. (F) GO enrichment of pathogenic targets showing top pathways involved in DN. (G) KEGG pathway enrichment highlighting Key KEGG pathways in DN progression. (H) GO enrichment confirming coordinated activation of diabetic nephropathy-related pathways. (I-J) GSEA confirming coordinated activation of diabetic nephropathy pathways in regulation of actin cytoskeleton (|NES|=0.49439877, NOM p-val=0.024564994, FDR q-val=0.598172) and TNF signaling (|NES|=0.55800134, NOM p-val=0.028017242, FDR q-val=0.6151668). Please click here to view a larger version of this figure.

Figure 4: Identification and validation of hyperglycemia-driven renal pathogenic hubs. (A) Integration pipeline: 278 high-confidence mediators derived by intersecting endothelial multi-omics data (534 targets) with DN-associated genes from GeneCards, the Comparative Toxicogenomics Database, and Open Targets Platform. (B) PPI network (STRING DB) showing key hub genes. (C-J) mRNA and protein levels of four top hubs: (C, D)VWF transcript and protein, (E, F) CCL2 transcript and protein; (G, H) SERPINE1 transcript and protein, (I, J) THBS1 transcript and protein. (K) ELISA validation shows hyperglycemia-induced CCL2 protein elevation in human glomerular endothelial cells. n=3. Data analyzed by unpaired t-tests. (L) HK2 cell viability assay demonstrates functional impact of HG-CM (containing elevated CCL2), with CCR2 inhibition rescuing the cytotoxic effect. n=3. Data analyzed by one-way ANOVA. Error bars represent the mean ± standard error of the mean (SEM). **p<0.01, ***p<0.001. Please click here to view a larger version of this figure.

Figure 5: Drug repositioning screen identifying compounds reversing pathogenic signatures. (A) Bioinformatic pipeline screening 33621 compounds against 278 pathogenic targets. (B) Heatmap of reversal efficiency of pathogenic targets. Please click here to view a larger version of this figure.

Figure 6: Transcriptomic regulator network analysis. Bioinformatic pipeline screening transcriptional factors regulating the expression of 278 pathogenic targets. Please click here to view a larger version of this figure.

Figure 7: Computational simulations of compound-transcription factor interactions. (A) Docking scores (kcal/mol) reflect the predicted binding affinities between compounds and transcription factors. (B) Molecular interactions between Idelalisib and CTCF, including three hydrogen bonds, five van der Waals contacts, and one Pi-Alkyl interaction. (C) Binding pose of Idelalisib with BRD4, featuring six van der Waals interactions, one Pi-Pi T-shaped interaction, and two Pi-Alkyl interactions.(D) Binding pose of MA9-086 with BRD4. (E) Bruceantin is bound to SPI1 through six hydrogen bonds and four van der Waals interactions. (F) Paracetamol formed one hydrogen bond and three van der Waals interactions with EP300. All structures were visualized using PyMOL. (G) Binding pose of XDM-CBP with EP300. Please click here to view a larger version of this figure.

Table 1: Functional efficacy of top FDA approval candidates.

Table 2: Experimental molecules according to the kidney cell line transcriptome.

Supplementary Table 1: Top 30 differentially expressed genes. This table lists the top 30 differentially expressed genes identified through transcriptomic analysis. It includes the gene names, normalized expression values for each replicate in the hyperglycemia model group (Model) and the normal control group (Control), the log2-transformed fold-change values [log2(FC)], p-values (pval), and adjusted false discovery rate q-values (qval). This list highlights the genes exhibiting the most significant expression changes under hyperglycemic conditions. Please click here to download this file.

Supplementary Table 2: Proteomics raw data of VWF, CCL2, SERPINE1, and THBS1. This table provides the raw quantitative data for four key hub proteins (VWF, CCL2, SERPINE1, THBS1) obtained via liquid chromatography-tandem mass spectrometry (LC-MS/MS) analysis. The data show the protein abundance from three biological replicates each in the high glucose-treated group (High_Glucose) and the normal glucose control group (Normal). Please click here to download this file.

This study establishes hyperglycemia-induced endothelial secretome dysfunction as a critical driver of diabetic kidney injury through integrated multi-omics and computational approaches. Demonstration that secretome-mediated cytotoxicity toward renal tubules coincides with systematic dysregulation of 534 endothelial-derived factors converging on matrix remodeling, oxidative stress, and inflammation-core pathways in DKD pathogenesis. Identification of 278 high-confidence mediators, including mechanistically validated hubs CCL2, VWF, SERPINE1, and THBS1 (exhibiting 1.15 to 2630-fold co-upregulation), resolves a key knowledge gap by mapping how endothelial transcriptional reprogramming translates to secreted pro-fibrotic, pro-inflammatory effectors^26,33. Of particular note is the CCL2 (MCP-1), a potent chemokine that orchestrates monocyte/macrophage recruitment into renal tissue. Clinical and experimental evidence consistently implicates CCL2 as a central mediator of diabetic kidney damage progression, and its overexpression worsens albuminuria, interstitial inflammation, and tubulointerstitial fibrosis^34,35. The exceptional magnitude of CCL2 induction observed here likely reflects its pivotal role in establishing a pro-inflammatory microenvironment that accelerates renal deterioration in diabetes. The concordance across transcriptomic and proteomic layers underscores the biological robustness of these targets, while PPI network analysis reveals their dense functional interconnectivity (SERPINE1-THBS1-FN1 fibrosis axis; CCL2-ICAM1 inflammation cluster), suggesting coordinated amplification of renal damage cascades. Crucially, these findings bridge in vitro hyperglycemic modeling with human DKD pathobiology through cross-database validation, providing a molecular roadmap for endothelial-directed therapeutics.

The computational drug repositioning pipeline revealed clinically actionable compounds capable of reversing this pathogenic signature. Significantly, idelalisib, bruceantin (a natural product mechanistic probe), and paracetamol demonstrated exceptional multi-target efficacy (top candidates among 85 identified inhibitors), serving as functional validations of the screening strategy. However, interpretation requires caution regarding translational readiness. The repositioned compound idelalisib-a clinically approved PI3Kδ inhibitor for hematologic malignancies-carries known immunotoxicity risks, including fatal infections and autoimmune complications³⁶; Bruceantin remains investigational and was included primarily as a molecular probe of SPI1-related signaling, not as a clinical candidate³⁷; Paracetamol (acetaminophen), though generally renal-safe in therapeutic doses, poses well-characterized hepatotoxicity at supratherapeutic levels^38,39. These profiles necessitate rigorous risk-benefit evaluation that extends beyond the scope of this computational study. Notably, mechanistic predictions were explored to illustrate network connectivity: Idelalisib's established PI3Kδ inhibition was computationally extended with conjectural docking to CTCF/BRD4 hubs (scores: -5.9/-5.2 kcal/mol), potentially linking to fibrosis/hypoxia responses^40,41; Bruceantin was computationally mapped to SPI1-mediated inflammation signaling⁴² ; while paracetamol potentially inhibited EP300-dependent matrix remodeling^42,43. This observed mechanistic coverage addresses DKD's multifactorial nature more effectively than single-target approaches, potentially overcoming limitations of current RAS blockade-based therapies^40,41. Importantly, the multi-TF targeting strategy (Figure 6, Figure 7) thus generates the mechanistic hypothesis that compounds could coordinate pathogenic networks as network stabilizers.

While the integrated computational and experimental approach provides substantial mechanistic insights, it is imperative to acknowledge specific methodological considerations that contextualize these findings and guide future validation. Practical constraints include potential serum-free incubation-induced stress during secretome collection. Proteomic depth was limited by incomplete low-abundance factor detection (e.g., cytokines below 3-kDa cutoff); exosomal component loss during ultrafiltration; and logistical challenges from time-sensitive processing. Cellular models used 24 h glucose exposure, shorter than human DKD progression timelines. Tubular cytotoxicity was assessed via static endpoint measurement (CCK-8), potentially missing dynamic responses. While the accelerated db/db mouse model offers efficiency, its transcriptional responses may not fully recapitulate human diabetic nephropathy. These protocol-specific technical limitations require consideration in interpretation.

Although the computational models predict compound-TF binding (e.g., idelalisib forming dual H-bonds with CTCF/BRD4; bruceantin occupying SPI1's functional pocket), in vitro and in vivo confirmation remains essential. Moreover, the strategy of targeting transcription factors (TFs) warrants consideration of selectivity challenges. Given TFs regulate both pathogenic networks (e.g., fibrosis/hypoxia) and essential cellular functions, off-target effects-such as disruption of homeostasis-related genes-represent a valid concern. On the other hand, future studies should also validate whether these compounds suppress target TF activity (e.g., CTCF chromatin looping, BRD4 super-enhancer formation), reduce pathogenic gene expression (SERPINE1/CCL2/THBS1), and mitigate secretome toxicity in advanced models, including patient-derived organoids and diabetic rodents. The prominent enrichment of fibrosis-regulating CTCF aligns with DKD's progressive fibrotic phenotype but warrants exploration in later disease stages. Critically, the automated docking and containerized bioinformatics require wet-lab validation to realize translational potential. Finally, idelalisib's dual inhibition of CTCF/BRD4 effectors, coupled with established safety profiles, positions it as a high-priority candidate for clinical evaluation in diabetic microvasculopathy. These acknowledged technical caveats not only inform the interpretation of the current data but crucially delineate essential parameters for the critical next phase: experimental validation.