Genomic scan points to seven kidney genes, but only one earns diabetes-specific support

Researchers have assembled a large exploratory framework that sifts public genetic data for genes whose regulation in specific kidney compartments may be linked to kidney function, albuminuria, and disease progression in both diabetes and the general population. The study, published in Frontiers in Genetics, is deliberately conservative in its conclusions: it produces a hypothesis-generating panel of seven genes rather than a list of confirmed causal genes for diabetic kidney disease, and it explicitly labels which pieces of evidence come from diabetes-specific outcomes and which come from broader population kidney traits.

The motivation is a familiar bottleneck in kidney genetics. Genome-wide association studies have expanded the catalog of loci linked to estimated glomerular filtration rate (eGFR) and urinary albumin-to-creatinine ratio (UACR), but they have not fully resolved how diabetes-associated regulatory variation maps onto albuminuria, reduced filtration, and progressive decline. Meanwhile, transcriptome-wide association studies, kidney disease single-cell atlases, and diabetic kidney disease-focused single-cell projects have built rich maps of injured kidney cells, yet these resources are rarely used as formal components of genetic target prioritization. The authors identified four gaps: diabetes-specific and population-wide outcomes are usually studied separately; progression phenotypes are handled apart from eGFR and UACR; single-cell maps are seldom integrated into prioritization; and Mendelian randomization-ranked genes are often carried forward without locus-level colocalization to check whether the same variant drives both expression and the kidney phenotype.

To address these gaps, the team integrated a long list of public resources: CKDGen and KP4CD outcome summary statistics, GTEx version 10 expression quantitative trait locus (eQTL) data, the Susztak/NephQTL kidney-compartment resources, the Human Protein Atlas, the INTERVAL plasma proteome, Open Targets annotations, and kidney disease single-cell atlases. Eleven outcome files were standardized, spanning diabetic eGFR in European and African ancestries, overall eGFR in more than 1.2 million participants, overall UACR in trans-ethnic and European analyses, diabetic UACR, two CKDGen progression endpoints (rapid decline and a 25 percent eGFR drop), and three KP4CD diabetes-specific eGFR-decline analyses in European, mixed, and African ancestries.

The exposure side came from eQTL datasets. After quality control, the GTEx significant-pairs resources contained 209,031 kidney-cortex variant-gene pairs across 2,440 genes and 2,744,871 whole-blood pairs across 13,871 genes. Independent kidney-compartment reproducibility used Susztak/NephQTL tubule, glomerulus, and combined kidney-meta eQTL resources, with older legacy datasets retained as sensitivity checks. After outcome overlap, allele harmonization, and instrument filtering, between 31 and 2,232 screening pairs entered the analysis per displayed status outcome. Importantly, the authors note that these eQTL resources largely represent non-diabetic tissue and predominantly European-ancestry samples, so they cannot model diabetes-dependent regulatory effects or provide ancestry-matched instruments for African-ancestry outcomes.

The genetic screen itself used best-instrument Mendelian randomization as an exploratory layer. Because kidney eQTL resources often had sparse cis instruments after liftover and filtering, a single representative variant per gene-outcome-resource combination was retained, and single-SNP Wald ratios were treated as screening estimates rather than stand-alone causal effect estimates. Primary false discovery rate correction was applied within each resource-outcome family, with a study-wide Benjamini-Hochberg sensitivity analysis across all 597,099 screening rows as a check. The authors are candid about the limits: independence and exclusion-restriction assumptions could not be verified, Steiger filtering was not possible, winner’s-curse bias may inflate apparent instrument strength, and no formal power calculation was undertaken, so null results from smaller resources are not evidence of no effect.

The seven-gene panel that emerged divides into three descriptive axes. For albuminuria, PRKCI and WIPF3 are primary follow-up loci and A3GALT2 is secondary; for eGFR, ITIH4 and LINC02166 are secondary; and for progression, UMOD is primary and FGF5 secondary. Colocalization analysis, which tests whether the same variant likely drives both gene expression and the trait in a locus, used a single-signal Wakefield approximate Bayes factor model across 28 prioritized tissue-locus combinations. PRKCI and WIPF3 showed very high kidney-cortex shared-variant-compatible evidence for overall UACR in European ancestry, with posterior probabilities of 0.999947 and 0.999390 respectively, and remained robust across tested priors and window sizes. In contrast, LINC02166, ITIH4, and A3GALT2 had high default posterior probabilities that collapsed sharply under alternative priors, with diffuse local patterns, so all three were downgraded to sensitivity-limited secondary loci. Four other genes, PGAP3, CDK10, RPP25, and SMDT1, were excluded after showing multisignal or weak colocalization patterns.

UMOD stands out as the only final-panel gene with false-discovery-rate-level support from both diabetes-specific outcomes and general kidney traits. In kidney-compartment analyses, UMOD was significant in tubule, glomerulus, and kidney-meta resources for both general CKDGen progression endpoints, with the strongest kidney-meta result for the 25 percent eGFR decline endpoint (P = 2.08e-8, within-outcome FDR = 2.08e-4, study-wide FDR = 3.42e-5). UMOD also met within-outcome FDR for KP4CD European and mixed-ancestry diabetes-specific eGFR decline, although the smaller African-ancestry outcome was null. Because the exposure eQTLs were not ancestry matched, the authors treat these cross-ancestry comparisons as exploratory transportability checks rather than ancestry-specific replication. FGF5 met within-outcome FDR only for general CKDGen progression in tubule and glomerulus, with no FDR-level diabetes-specific support.

Biological context layers added texture without being treated as validation. Healthy kidney single-cell annotation placed UMOD strongly in the distal tubule (pTPM = 883.8), consistent with thick ascending limb biology, and PRKCI in collecting duct cells (pTPM = 73.0). Disease-state single-cell analysis showed UMOD differential-expression evidence across several tubule segments and injured proximal-tubule trajectory membership, while PRKCI appeared in multiple interstitial, immune, and fibroblast cell populations. External expression datasets from microdissected DKD glomeruli, tubulointerstitium, and diabetic nephropathy biopsies provided FDR-level differential-expression context for WIPF3, ITIH4, UMOD, and FGF5, with PRKCI showing only nominal context. Protein and tractability annotation showed PRKCI with the broadest coverage, including kidney immunohistochemistry detection, plasma target-assay coverage, and direct target-protein QTL evidence, though these are described strictly as annotation and feasibility signals, not evidence of therapeutic readiness.

The authors repeatedly caution against overreading the results. Kidney-compartment recurrence reuses the same outcome GWAS summary statistics, so it represents eQTL-resource reproducibility rather than independent external replication or proof of kidney specificity. The single-signal colocalization model cannot rule out multiple causal variants at a locus, and no SuSiE or LD-aware conditional analysis was performed. Most panel support derives from general kidney traits rather than diabetes-specific outcomes, and neither PRKCI nor WIPF3 has FDR-level diabetic UACR support. The framework’s value, the authors argue, lies in its auditable, stagewise selection path and explicit downgrade rules, which prioritize loci for mechanistic follow-up while avoiding claims of definitive causality, diabetic kidney disease specificity, or therapeutic readiness. Independent ancestry-matched replication, multisignal colocalization, and targeted perturbation experiments in diabetes-relevant kidney models remain the necessary next steps before any of these seven genes can be called a causal or actionable DKD target.

Subject of Research: An exploratory Mendelian randomization and colocalization framework prioritizing kidney-compartment genes across diabetic and population kidney traits

Article Title: Kidney-compartment genomic prioritization across diabetic and population kidney traits: an exploratory Mendelian randomization and colocalization study

Article References: Chen, R., & Bi, X. (2026). Kidney-compartment genomic prioritization across diabetic and population kidney traits: an exploratory Mendelian randomization and colocalization study. Frontiers in Genetics, 17, Article 1921914. https://doi.org/10.3389/fgene.2026.1921914

Image Credits: AI Generated

DOI: 10.3389/fgene.2026.1921914

Keywords: diabetic kidney disease, Mendelian randomization, colocalization, eQTL, kidney cortex, UMOD, PRKCI, WIPF3, albuminuria, eGFR, single-cell transcriptomics, GWAS