Multi-Omics Maps Compartment-Specific Gut-Kidney Axis in Diabetic Kidney Disease, Prioritizing Caffeic Acid and Naringenin Chalcone
核心洞察
An integrative multi-omics study identified six glomerular and five tubulointerstitial core genes linked to gut microbiota metabolites in diabetic kidney disease (搜索).
Machine learning and Mendelian randomization prioritized MMP7 (搜索) as the only core gene shared by both kidney compartments and consistently tied to eGFR decline.
Caffeic acid and naringenin chalcone (搜索) were the only metabolites retained in both compartment networks after drug-likeness and toxicity screening.
An integrative multi-omics analysis has produced a compartment-resolved map of how gut microbiota metabolites may influence diabetic kidney disease (搜索) (DKD), identifying distinct molecular signatures in the glomerulus and tubulointerstitium and prioritizing two shared candidate metabolites for further study. The work, published in Frontiers in Immunology, combines machine learning, Mendelian randomization, single-cell profiling, molecular dynamics simulation and in vitro validation to construct a "Microbiota-Substrate-Metabolite-Target" (M-S-M-T) regulatory network for each kidney compartment.
The authors frame the study explicitly as computational and hypothesis-generating, noting that the kidney's cellular heterogeneity has limited prior target-identification efforts that relied on general databases such as GeneCards. "This is a computational, hypothesis-generating study whose mechanistic center was the compartment-resolved 'Microbiota-Substrate-Metabolite-Target' relationship," they write, adding that immune and therapeutic observations were "supportive and correlative rather than mechanistically established."
Compartment-Specific Core Genes
Investigators screened 248 gut microbiota metabolites from the gutMGene database, deriving 1,267 targets from the Similarity Ensemble Approach and 777 from SwissTargetPrediction, with 556 shared target genes after intersection. Ten transcriptomic datasets from the NCBI Gene Expression Omnibus were partitioned into glomerular and tubulointerstitial training, validation and test cohorts.
After differential expression analysis and weighted gene co-expression network analysis, machine learning models were built using twelve algorithms across 137 models, with selection based on a penalized score designed to limit overfitting. The analysis converged on six glomerular core genes — IGFBP6 (搜索), PLA2G4A (搜索), CTSK (搜索), HTR2B (搜索), PDGFRA (搜索) and MMP7 (搜索) — and five tubulointerstitial core genes — CA2 (搜索), HSD11B2 (搜索), NQO2 (搜索), MMP7 and CYP24A1 (搜索). All six glomerular genes were markedly upregulated; in the tubulointerstitium, CA2, HSD11B2 and NQO2 were downregulated while MMP7 and CYP24A1 were upregulated.
Diagnostic performance varied by gene. PLA2G4A (搜索) and IGFBP6 (搜索) showed robust glomerular diagnostic utility with area under the curve values above 0.82, while HSD11B2 (搜索) reached an AUC above 0.87 in the tubulointerstitial compartment. SHAP analysis identified IGFBP6 and CTSK (搜索) as the primary glomerular model contributors and MMP7 (搜索) as the leading tubulointerstitial contributor.
Genetic and Clinical Corroboration
Two-sample Mendelian randomization using cis-eQTL, cis-pQTL and cis-mQTL instruments provided supportive genetic evidence. Genetically predicted higher NQO2 (搜索) expression was associated with reduced eGFR in eQTL analysis (OR = 0.999, p = 0.032), supported by mQTL evidence against renal failure (OR = 0.953, 95% CI 0.918–0.990, p = 0.014). Elevated CTSK (搜索) expression and higher plasma PLA2G4A (搜索) levels were tied to reduced eGFR (CTSK: OR = 0.998, p < 0.001; PLA2G4A: OR = 0.992, p = 0.046). HTR2B (搜索) methylation linked to reduced eGFR (OR = 0.997, p < 0.001) alongside increased DKD risk (OR = 1.153, p = 0.038), while HSD11B2 (搜索) methylation preserved eGFR (OR = 1.038, p < 0.001) and higher CYP24A1 (搜索) expression was protective against DKD (OR = 0.607, p = 0.013).
In the Nephroseq v5 clinical cohorts, all six glomerular core genes correlated negatively with eGFR, with statistical significance after Bonferroni correction for IGFBP6 (搜索), PLA2G4A (搜索) and MMP7 (搜索). In the tubulointerstitial cohort, CA2 (搜索), HSD11B2 (搜索) and NQO2 (搜索) correlated positively with eGFR, whereas MMP7 and CYP24A1 (搜索) correlated negatively.
Single-cell profiling revealed cell-type-specific expression patterns. In the glomerulus, mesangial cells showed the highest activity, with PLA2G4A (搜索) upregulated and IGFBP6 (搜索) and HTR2B (搜索) downregulated; MMP7 (搜索) was upregulated in glomerular parietal epithelial cells. In the tubulointerstitium, MMP7, CYP24A1 (搜索) and NQO2 (搜索) were upregulated in the proximal convoluted tubule, while HSD11B2 (搜索) and NQO2 were decreased in the loop of Henle.
Immune Landscape Shifts
Immune infiltration analysis using CIBERSORT indicated a higher estimated proportion of M2 macrophages and a shift in estimated mast cell composition from activated to resting states in both compartments, correlating with core gene expression. Gene set variation analysis showed upregulated core genes positively associated with macrophage- and mast cell-related pathway enrichment scores. The authors caution that these findings derive from deconvolution and gene set scoring of bulk transcriptomes and "do not demonstrate that the core genes, or the candidate metabolites, modulate macrophage polarization or mast cell activity."
Metabolite Screening and In Silico Prioritization
Tracing the M-S-M-T networks back from the shared core gene MMP7 (搜索) yielded eight candidate metabolites under Lipinski's Rule of Five, all of which were excluded for predicted nephrotoxicity or other organ toxicities. Broadening the search identified 33 drug-like metabolites in the glomerular network and 69 in the tubulointerstitial network. After ADMETlab 3.0 toxicity filtering, caffeic acid and naringenin chalcone (搜索) emerged as the only metabolites retained in both compartment-specific networks.
Molecular docking positioned naringenin chalcone (搜索) as the superior candidate, with robust affinity for HTR2B (搜索) (−8.602 kcal/mol) surpassing the standard inhibitor SB-204741 (−8.027 kcal/mol). Unlike SB-204741, which relied primarily on hydrophobic contacts, naringenin chalcone established a π-stacking interaction with TYR237 and dual hydrogen bonds with THR240. Caffeic acid bound HTR2B with lower affinity (−6.897 kcal/mol). Molecular dynamics simulations showed binding free energies of −108.1949 ± 11.6700 kJ/mol for the HTR2B–naringenin chalcone complex and −115.3688 ± 7.7848 kJ/mol for HTR2B–caffeic acid, both more stable than the MMP7 (搜索)–naringenin chalcone complex (−40.5371 ± 10.9060 kJ/mol).
In Vitro Validation and Target Dependency
In high-glucose-treated HK-2 proximal tubular cells and podocytes, both metabolites reversed core gene dysregulation at working concentrations of 100 μM naringenin chalcone (搜索) and 50 μM caffeic acid. Western blot analysis showed that high glucose significantly downregulated MMP-7 protein in both cell types and HTR2B (搜索) in podocytes, with both metabolites reversing these changes. High glucose also downregulated the anti-apoptotic protein Bcl-2, which was restored by either metabolite.
To probe target dependency, investigators added the selective HTR2B (搜索) inhibitor RS-127445 or the MMP-7 inhibitor marimastat. These inhibitors exacerbated high-glucose-induced Bcl-2 downregulation and abolished the restorative effect of the drugs, suggesting both targets contribute to the observed effect.
Translational Distance and Limitations
The authors draw a clear line between their computational framework and clinical readiness. Among all core genes, only MMP7 (搜索) currently carries multi-dimensional clinical evidence as a measurable biomarker, with unbiased kidney tissue proteomics identifying plasma MMP7 as a predictor of renal function decline and urinary MMP7 associated with disease progression and mortality in type 2 diabetes. PDGFRA (搜索) is described as an intermediate case — useful as a tissue marker of fibrosis in biopsy sections rather than a liquid-biopsy analyte. The remaining core genes have not been evaluated in any clinical cohort.
No target-directed clinical trial exists for any of these genes. The authors note that PDGFRA (搜索) can be addressed by multi-target tyrosine kinase inhibitors such as imatinib and sunitinib with supporting animal evidence in renal fibrosis, MMP7 (搜索) by broad-spectrum and selective MMP inhibitors with preclinical antifibrotic activity, and CA2 (搜索) by the approved carbonic anhydrase inhibitor acetazolamide — which nevertheless reduced glomerular filtration in obese non-diabetic individuals, illustrating the risk of interfering with a physiologically essential transport enzyme.
Both candidate metabolites carry liabilities: limited oral bioavailability and chemical stability, with naringenin chalcone (搜索) studied predominantly in plant science and caffeic acid approved in China only for thrombocytopenia. Caffeic acid is further reduced by gut microbiota to dihydrocaffeic acid, itself proposed as a risk marker for type 2 diabetes.
The study's limitations are substantial and acknowledged. Mendelian randomization relied on blood-sourced genetic instruments from European populations, limiting tissue relevance and transferability to other ancestries, and single-SNP instruments could not undergo pleiotropy testing. The M-S-M-T networks are predictions from curated databases and target prediction algorithms; no microbiome sequencing or fecal, serum or renal metabolomics was performed, so the authors cannot demonstrate that predicted metabolites are altered in DKD or reach renal compartments at biologically relevant concentrations. The in vitro data are limited to two cell lines with acute high-glucose exposure, which cannot reproduce the chronic DKD microenvironment. Several core genes did not reach statistical significance in individual validation or test cohorts, most plausibly due to small sample sizes and clinical heterogeneity.
"These findings establish a hypothesis-generating framework for the gut-kidney axis in DKD and define specific priorities for validation in clinical microbiome and metabolomic studies," the authors conclude, calling for real-world clinical samples with gut microbiome sequencing and direct measurement of caffeic acid and naringenin chalcone (搜索) in serum, feces, urine and kidney tissue before intervention strategies targeting specific microbial components are pursued.