Introduction

Acute kidney injury (AKI) is a common clinical syndrome characterized by a rapid decline or loss of kidney function within a short period, often accompanied by elevated serum creatinine levels and reduced urine output.1 The occurrence of AKI is closely associated with higher mortality rates, prolonged hospitalization, and an increased risk of progression to chronic kidney disease (CKD), representing a major healthcare burden for hospitalized and critically ill patients worldwide.2 Epidemiological data indicate that the incidence of AKI is approximately 10–15% among general hospitalized populations, while it can exceed 50% in intensive care units (ICUs).1,3 Ischemia-reperfusion (I/R)-induced AKI, which ranks as the first leading cause of AKI in the intensive care unit, is associated with high perioperative mortality and morbidity as well as increased medical expense.4 Despite continuous improvements in supportive care, there is currently no clinically validated pharmacological therapy available to prevent or reverse the onset of AKI. This challenge primarily arises from the complex etiology and heterogeneous clinical phenotypes of AKI, significant pathophysiological differences between animal models and humans, and persistent limitations in the design and patient selection of existing clinical trials.5 Consequently, current AKI management still largely relies on supportive measures and treatment of underlying causes, lacking effective targeted interventions. Therefore, identifying novel biomarkers and potential pathogenic mechanisms remains a critical strategy to improve the prognosis of AKI.6

In recent years, the rapid advancement of high-throughput omics technologies has enabled the integration of proteomics with genome-wide association studies (GWAS), offering new opportunities to systematically elucidate the molecular mechanisms underlying complex diseases.7 Plasma proteins, as key bioactive molecules, are extensively involved in pathological processes highly relevant to AKI, including inflammation, oxidative stress, and immune regulation.8,9 Due to their accessibility in body fluids, plasma proteins have become one of the most common targets in drug development, with the majority of U.S. Food and Drug Administration (FDA)-approved drugs acting on circulating proteins. Observational studies have preliminarily suggested that certain plasma proteins may play a role in the onset and progression of AKI. For instance, CCL-2 and CXCL-8 are markedly upregulated in patients with snakebite-associated AKI; Src family kinases (eg., Src, Fyn, Lyn) participate in the regulation of oxidative stress and apoptosis; and KIM-1, NGAL, and IL-18 have demonstrated predictive value in contrast-induced AKI.10,11 However, most of these studies are association-based and are susceptible to confounding bias and reverse causality, leaving causal relationships insufficiently established. Therefore, more rigorous study designs and methodological approaches are urgently needed.

Mendelian randomization (MR) is an analytical approach that uses genetic variants as instrumental variables to infer causal relationships between an exposure and an outcome.12 Because genetic variants are fixed at conception, MR inherently mimics randomization, thereby minimizing confounding and reverse causality.13 At the proteomic level, the use of protein quantitative trait loci (pQTL) in proteome-wide Mendelian randomization (PWMR) enables the systematic identification of protein targets associated with disease, and has been successfully applied in therapeutic target discovery for disorders such as multiple sclerosis and inflammatory bowel disease.14,15

Although proteome-wide MR studies have yielded promising findings in several disease areas, research related to AKI remains scarce. Current protein-related investigations in AKI have predominantly focused on identifying diagnostic biomarkers, with limited exploration of their causal roles in pathogenesis.16,17 Therefore, this study aimed to identify potential candidate proteins associated with the risk of AKI through a systematic PWMR analysis. We integrated genetic instruments for plasma proteins from the UK Biobank Pharma Proteomics Project (UKB-PPP) and performed PWMR using AKI GWAS data from the FinnGen study. To enhance the robustness of causal inference, we further conducted Bayesian colocalization analysis to confirm shared genetic signals between proteins and AKI, and replicated the findings in independent pQTL datasets. In addition, a phenome-wide association study (PheWAS) was performed to assess the phenotypic spectrum of candidate targets, and two-step MR was used to explore potential mediation pathways involving inflammatory and metabolic factors. Through this multidimensional integrative approach, our study aims to provide a systematic foundation and theoretical basis for the discovery and translation of candidate biomarkers for AKI.

Materials and Methods Study Design and Ethics

This study aimed to identify candidate biomarkers for acute kidney injury (AKI), and the overall workflow is illustrated in Figure 1. Plasma proteomic data from the UK Biobank Pharma Proteomics Project (UKB-PPP) were integrated with AKI genome-wide association study (GWAS) data from the Finnish database to perform a proteome-wide summary-data–based Mendelian randomization (SMR) analysis, aimed at preliminarily screening candidate proteins significantly associated with AKI. To enhance the robustness of causal inference, significant associations were further subjected to heterogeneity in dependent instruments (HEIDI) testing and Bayesian colocalization analysis to exclude spurious associations driven by linkage disequilibrium. Subsequently, an independent protein quantitative trait loci (pQTL) dataset was introduced to validate the robustness of the initially identified proteins. Genetic information for proteins was obtained from deCODE Genetics, while AKI GWAS data were derived from the IEU Open GWAS platform, and the causal associations between candidate proteins and AKI were validated within a two-sample Mendelian randomization (MR) framework. To further clarify whether these proteins contribute to AKI through immune mechanisms, 486 serum metabolites and 91 circulating inflammatory factors were incorporated to construct a two-stage mediation model, exploring the mediating role of immune pathways in the protein–AKI relationship. Finally, to assess other potential clinical effects of these candidate proteins, a phenome-wide association study (PheWAS) was performed on the candidate proteins to systematically identify possible off-target effects. Some data used in this study were obtained from publicly available databases, and the GWAS and pQTL studies involved had received ethical approval in the original publications;Clinical data were collected with ethical approval from the First Affiliated Hospital of Zhengzhou University (2022-KY-0270).Our article includes all essential reporting elements for Mendelian randomization, complies with the reporting guidelines, and the completed STROBE-MR checklist is provided.

Figure 1 Schematic of the study design in this Mendelian randomization (MR) analysis.

Data Sources Plasma Proteomics

The plasma protein quantitative trait loci (pQTL) data used in this study were obtained from the UK Biobank Pharma Proteomics Project (UKB-PPP), which comprises large-scale genetic and proteomic data from 54,219 European participants.A total of 2,940 plasma proteins were quantified and analyzed using a high-throughput proteomics platform, generating genome-wide pQTL association results.18 The replication plasma proteomic dataset was derived from deCODE Genetics, which utilized large-scale plasma proteomics data in GWAS, encompassing genetic and phenotypic information from 35,559 Icelandic individuals. Protein levels were measured for 4,907 proteins using the SomaScan v4 aptamer-based multiplex assay.19

Acute Kidney Injury

GWAS summary statistics for AKI in the discovery cohort were obtained from the FinnGen database (version R12, https://r12.finngen.fi/, GWAS ID: finngen_R12_N14_ACUTERENFAIL), comprising 8,383 cases (3,194 females and 5,450 males) and 480,448 controls, with a total of 20,112,606 SNPs. In the Finnish database, acute renal failure is defined as “An acute condition that is characterized by the inability of the kidneys to adequately filter the blood.”Diagnosis was based on ICD-10 code N17, with endpoint definitions determined by the FinnGen expert panel.20 The mean age at first diagnosis was 70.55 years. Replication GWAS data for AKI were obtained from the IEU Open GWAS database (ID: ebi-a-GCST90018790), including 482,266 individuals and 24,187,658 SNPs.21

Mediator Data Sources

GWAS data for 91 circulating inflammatory proteins analyzed in this study were derived from the latest SCALLOP Consortium study and the publicly available GWAS Catalog dataset (https://www.ebi.ac.uk/gwas/publications/37563310), based on Olink Target platform measurements in 14,824 participants.22 The dataset included key cytokines such as IL-4, IL-6, IL-7, IL-10, IL-12, IL-15, IL-17, IL-22, IL-33, and CXCL10. Laboratory assays were performed at Olink’s Uppsala laboratory, with SNP genotyping conducted via array-based methods and imputation performed using the 1000 Genomes/HRC reference panels. This study also utilized GWAS results from Shin et al, which analyzed 7,824 European adults (from the Twins UK and KORA cohorts) using untargeted metabolomics (HPLC/GC-MS) to identify 486 serum metabolites. Among these, 309 were validated via the KEGG database and classified into eight categories: amino acids, carbohydrates, cofactors and vitamins, energy metabolites, lipids, nucleotides, peptides, and exogenous compounds. The remaining 177 structurally unidentified metabolites (denoted as “X-”) were excluded from analysis.23 The original dataset is publicly available from the Helmholtz Metabolomics GWAS Server (https://metabolomics.helmholtz-muenchen.de/gwas/), containing 2.16 million SNP-metabolite association records.Through the above filtering process, the final set of instrumental variables demonstrated strong independence and explanatory power, providing a solid foundation for subsequent causal inference between proteins and disease outcomes.

Summary-Data-Based Mendelian Randomization (SMR) and HEIDI Test Analysis

This study employed summary-data-based Mendelian randomization (SMR) to integrate GWAS and plasma protein quantitative trait loci (pQTL) datasets, in order to assess the causal relationship between circulating plasma protein levels and disease risk.24 Analyses were conducted using the SMR software (v1.3.1/V1.03), with a false discovery rate-adjusted p-value threshold of less than 0.2 for statistical significance.25 Considering the exploratory nature of the proteome-wide screening analysis, an FDR threshold of 0.2 was adopted during the discovery stage to retain potentially informative signals for downstream validation analyses.The identified associations should therefore be interpreted as exploratory findings requiring further validation.This threshold was intended to maximize sensitivity for candidate identification, while subsequent colocalization analyses, external MR validation, and clinical validation were used to improve the robustness and credibility of the findings.To differentiate pleiotropy from linkage effects, the heterogeneity in dependent instruments (HEIDI) test was further applied: a p-value > 0.2 suggests the observed association is likely driven by shared genetic variants (pleiotropy), whereas p < 0.2 indicates linkage effects, ie., distinct but closely linked genetic variants influencing plasma protein levels and disease risk separately. Multiple testing errors were controlled via FDR correction, with a HEIDI test p-value > 0.05 set as the criterion for excluding linkage disequilibrium confounding. Novel plasma protein biomarkers were identified based on the following criteria: (1) achieving significance in SMR analysis; (2) supporting pleiotropy mechanism by HEIDI test (p > 0.05); and (3) lacking prior evidence of association with the target disease at genetic polymorphism, expression, or protein levels. This approach provides robust evidence to elucidate the causal associations between circulating proteins and diseases.24

Colocalization Analysis

To exclude false positives caused by linkage disequilibrium and validate the colocalization of genetic variants, Bayesian colocalization analysis was performed using the R package “coloc”. The analysis was based on five hypotheses: (i) H0: no causal variant affecting either exposure or outcome in the genomic region; (ii) H1: a causal variant affects only the exposure (eg., protein expression); (iii) H2: a causal variant affects only the outcome (eg., disease risk); (iv) H3: two distinct causal variants independently regulate exposure and outcome; (v) H4: a shared causal variant drives both exposure and outcome. Prior probabilities were set as follows: p1 (SNP associated with exposure) = 1 × 10−4, p2 (SNP associated with outcome) = 1 × 10−4, and p12 (SNP associated with both) = 1 × 10−5.26,27 SNPs within ±1000 kb of the pQTL locus were included for testing. A posterior probability of PP.H4 > 80% was considered strong evidence of colocalization, indicating that exposure and outcome may be regulated by the same causal variant.28 This method further validated the biological relevance of SMR findings and provided a reliable basis for drug target screening.

External MR Validation

A two-sample Mendelian randomization approach was applied to evaluate the causal associations between plasma proteins and disease. Instrumental variables were first selected based on genome-wide significance (p < 5 × 10−8), with F-statistics calculated for each SNP (F = β2/SE2); weak instruments with F < 10 were excluded. Independence of instruments was ensured by linkage disequilibrium pruning (r2 < 0.001, within 10,000 kb). All analyses were performed using the “TwoSampleMR” package (version 0.5.8) in R software (version 4.3.1). The inverse variance weighted (IVW) method was primarily used to estimate causal effects, with p-values corrected for false discovery rate (FDR). Sensitivity analyses were conducted using weighted median and MR-Egger methods to verify robustness. Reverse causation was excluded via Steiger filtering. Heterogeneity was assessed using Cochran’s Q test, and horizontal pleiotropy was examined with MR-Egger intercept and MR-PRESSO methods, ensuring result reliability.

Phenome-Wide Association Study

This study employed a phenome-wide Mendelian randomization (PheWAS) approach to systematically investigate potential adverse effects of positive proteins identified as candidate biomarkers through prior SMR and colocalization analyses.29 Using 2,469 clinical phenotypes from the Finnish FinnGen R12 database (N=412,181), cis-pQTLs were used as genetic instruments and evaluated via a two-stage analytical strategy: the Wald ratio method was applied for single instrumental variables, whereas the inverse variance weighted (IVW) method was used for multiple instruments. All analyses were subject to stringent multiple testing correction (FDR < 0.05 using the Benjamini-Hochberg method), and instrument strength was ensured with F-statistics exceeding 10. To control potential biases, Steiger directionality testing was performed to exclude reverse causation, and horizontal pleiotropy was assessed using the MR-Egger intercept test (p > 0.05). This analysis not only validated the specificity of the association between target proteins and diseases but also systematically screened for a broad spectrum of potential clinical effects, providing critical safety information for subsequent drug development. All statistical analyses were conducted using the TwoSampleMR package (version 0.5.8) in R software (version 4.3.1).

Two-Step Mendelian Randomization Mediation Analysis

In causal inference studies, the focus extends beyond assessing the direct effects of exposures on outcomes to elucidating the underlying biological mechanisms and pathways. In this study, we applied a two-step Mendelian randomization (MR) framework to systematically evaluate the mediating roles of plasma proteins in the effects of inflammatory factors (n = 91) and plasma metabolites (n = 486) on acute kidney injury (AKI), thereby delineating causal pathways across multiple omics layers and quantifying their relative importance. Specifically, two-sample MR analyses were conducted to estimate (1) the causal associations between exposures (inflammatory factors/metabolites) and potential mediators (plasma proteins) (β1), and (2) the associations between mediators and AKI (β2). Mediation effects were calculated using the product-of-coefficients method (β1 × β2), with 95% confidence intervals (CIs) estimated via the delta method. The proportion mediated was computed as (mediation effect / total effect) × 100%, where the total effect was derived from standard MR analyses and the direct effect was defined as the total effect minus the mediation effect.

All analyses met quality control criteria including instrument strength with F-statistics > 10, use of IVW as the primary analytical method, and assessment of horizontal pleiotropy via MR-Egger intercept test (p > 0.05). Results were categorized into two levels based on evidence strength: a potential mediation pathway was identified when significant causal associations existed among exposure, mediator, and outcome; strong evidence of mediation was confirmed if the 95% confidence interval of the mediation effect did not include zero. This analysis provided important insights into the multi-omics regulatory mechanisms underlying AKI pathogenesis. All calculations were conducted using the “TwoSampleMR” package in R (version 4.3.1).

Clinical Data and Specimens

With ethical approval from the First Affiliated Hospital of Zhengzhou University (2022-KY-0270), blood samples were collected from patients undergoing cardiac surgery with cardiopulmonary bypass and healthy controls were recruited from individuals undergoing routine physical examinations at the hospital physical examination center during the same study period. The study enrolled patients between September 1, 2022, and April 30, 2025. Inclusion criteria for the patient cohort were: (1) age between 18 and 80 years; (2) undergoing elective cardiac surgery with cardiopulmonary bypass; (3) availability of complete clinical and laboratory data; and (4) provision of written informed consent.Exclusion criteria included: (1) pre-existing chronic kidney disease; (2) history of renal replacement therapy or kidney transplantation; (3) active infection or sepsis; (4) malignancy; (5) severe hepatic dysfunction; (6) autoimmune or inflammatory diseases; (7) perioperative use of nephrotoxic medications; (8) pregnancy; and (9) incomplete clinical data.

Healthy controls were recruited from individuals undergoing routine health examinations at the same institution during the study period. Inclusion criteria for healthy controls included age between 18 and 80 years, normal renal function, and absence of major systemic disease. Individuals with kidney disease, active infection, autoimmune disease, malignancy were excluded. Clinical data recorded encompassed demographics, comorbidities, vital signs, surgical type, operation duration, cost, and admission levels of serum creatinine and blood urea nitrogen. Based on acute kidney injury status, patients were divided into an AKI group (n=50) and a non-AKI group (n=50), with an additional healthy control group (n=20). Serum levels of CX3CL1 (Cloud-Clone Corp, Wuhan, China, Catalog No.USEA040Hu) and uromodulin (UMOD)(Cloud-Clone Corp, Wuhan, China, Catalog No.USEG918Hu) were measured using commercially available enzyme-linked immunosorbent assay (ELISA) kits according to the manufacturers’ instructions. All samples were analyzed in duplicate, and the mean values were used for statistical analysis.

Statistical Analysis

Data analysis was conducted with SPSS 25 and GraphPad Prism 9.5. Inter-group comparisons utilized the Student’s t-test (two groups) or one-way ANOVA (multiple groups). Non-normally distributed data and categorical variables were analyzed via the Kruskal–Wallis test and Chi-squared test, respectively. Relationships between variables were examined using Spearman’s rank correlation. Continuous data are reported as mean ± standard deviation (SD), and the Mann–Whitney U-test was used for comparing means. A threshold of P < 0.05 defined statistical significance.

Results Plasma Protein Expression and AKI

In this study, 2017 genes containing cis-pQTLs were extracted from the UKB-PPP database (Supplementary 1). Under the threshold of FDR-adjusted p-value < 0.2 for P_SMR, three genes were identified via summary-data-based Mendelian randomization (SMR) analysis to have significant causal associations with AKI susceptibility. These genes included HLA-DRA (OR = 1.129, 95% CI: 1.066–1.197, p_SMR = 3.57 × 10−5, P_adj = 0.059, P_HEIDI = 0.100), UMOD (OR = 1.082, 95% CI: 1.041–1.125, p_SMR = 5.92 × 10−5, P_adj = 0.059, P_HEIDI = 0.056), and CX3CL1 (OR = 1.323, 95% CI: 1.139–1.536, p_SMR = 2.46 × 10−4, P_adj = 0.166, P_HEIDI = 0.242) (Figure 2a and b). All loci passed the HEIDI test.

Figure 2 (a). Volcano plot of candidate plasma proteins associated with AKI identified by SMR analysis. (b) Forest plot of three proteins showing causal relationships with AKI. (c) Colocalization analysis of three AKI-associated causal proteins.

Moreover, colocalization analysis was performed on the protein loci that passed both SMR and HEIDI tests, providing evidence of colocalization between AKI risk and UMOD (PP.H4.abf = 90.0%, PP.H3.abf = 4.82%) and CX3CL1 (PP.H4.abf = 96.2%), suggesting that these loci may represent true causal effects on AKI risk. However, HLA-DRA (PP.H4.abf = 37.1%, PP.H3.abf = 53.0%) showed no evidence of colocalization with AKI. This discrepancy highlights the necessity of employing multiple analytical approaches to fully understand gene-phenotype associations. Colocalization plots for various protein loci are presented in Figure 2c.

To further validate these genes and emphasize their consistency across different populations and genetic backgrounds, Mendelian randomization analyses were conducted using external datasets. Under stringent selection criteria, 5 and 30 SNPs were chosen as instrumental variables (IVs) for CX3CL1 and UMOD, respectively, with F-statistics ranging from 30.98 to 3693.84, indicating no weak instrument bias. Causal relationships with acute renal failure were evaluated using five MR methods. The inverse variance weighted (IVW) method, which has the highest statistical power, indicated significant positive associations for CX3CL1 (OR = 1.111, 95% CI: 1.002–1.232, Pfdr = 0.046) and UMOD (OR = 1.065, 95% CI: 1.012–1.122, Pfdr = 0.032) with AKI, suggesting that elevated plasma CX3CL1 and UMOD may increase AKI risk (Figure 3a). No significant heterogeneity was detected by Cochran’s Q test (P > 0.05), and no evidence of horizontal pleiotropy was found from the MR-Egger intercept, indicating robustness of the results. Leave-one-out sensitivity analysis demonstrated consistent results after excluding any single SNP (Figure 3b–e), further confirming the stability of the analysis. The scatter plot (Figure 3c–f) showed consistent positive slope directions across the five MR methods; the forest plot (Figure 3d–g) further confirmed IVW and MR-Egger results; the funnel plot (Figure 3e–3i) indicated symmetrical SNP distribution, suggesting absence of publication bias.

Figure 3 (a). Summary forest plot of external MR validation for colocalization-positive plasma proteins (b). Leave-one-out sensitivity analysis of CX3CL1 protein(c). MR scatter plot of CX3CL1 protein (d). Effect size forest plot of CX3CL1 protein(e). Funnel plot of CX3CL1 protein (f). Leave-one-out sensitivity analysis of UMOD protein(g). MR scatter plot of UMOD protein (h). Effect size forest plot of UMOD protein(i). Funnel plot of UMOD protein.

Figure 3 continued.

MR-PheWAS Analysis of CX3CL1 and UMOD as Candidate Biomarkers for AKI

Finally, we conducted Mendelian Randomization Phenome-wide Association Study (MR-PheWAS) analysis to investigate potential beneficial or adverse effects of CX3CL1 and UMOD protein loci as candidate biomarkers for acute kidney injury (AKI). Single nucleotide polymorphisms (SNPs) were selected from previous protein loci and migraine MR analyses. We screened 2,469 phenotypes from the Finnish FinnGen database (version R12), comprising 500,348 individuals. After correction for false discovery rate (FDR), no obvious beneficial or harmful effects were observed. To further explore the results, the top three most likely phenotypes associated with each protein were selected for detailed analysis. UMOD showed significant associations with disorders of carbohydrate metabolism, other and unspecified tonsillitis, and pain and other conditions associated with female genital organs and the menstrual cycle. CX3CL1 was significantly associated with hypertension, antihypertensive medication (noting other indications), and meningococcal infection. More detailed results are shown in Figure 4. PheWAS results for CX3CL1 and UMOD associations with other disease outcomes are provided in Supplementary 2.

Figure 4 Manhattan plot of MR-PheWAS analysis for associations between CX3CL1/UMOD plasma proteins and disease outcomes in FinnGen R12.The x-axis represents different phenotype categories included in the Mendelian randomization phenome-wide association study (MR-PheWAS), and the y-axis represents the statistical significance of the associations expressed as −log10(P value). Each point corresponds to a specific phenotype, with red and blue points representing associations for CX3CL1 and UMOD, respectively. Higher values on the y-axis indicate stronger statistical evidence for association. Selected phenotypes with relatively significant associations are labeled in the plot. This analysis was performed to evaluate potential pleiotropic associations of CX3CL1 and UMOD across diverse disease phenotypes.

Mediation Analysis

To investigate the indirect effects of proteins on AKI outcomes via inflammatory factors and plasma metabolites, we performed mediation analysis using two-step Mendelian randomization effect estimates and primary MR total effects. This analysis was limited to two proteins, CX3CL1 and UMOD, which demonstrated evidence of causal effects on inflammatory factors, plasma metabolites, and stroke outcomes in MR analyses.We used the product method to estimate indirect effects and the delta method to calculate standard errors (SE) and confidence intervals (CI). Initially, 91 inflammatory factors and 486 metabolites were considered exposures. After excluding 5 metabolites without clear identification, three inflammatory factors—Interleukin-24, C-C motif chemokine 4, and Eukaryotic translation initiation factor 4E-binding protein 1—and seven plasma metabolites (N-acetylalanine, Laurylcarnitine, ADpSGEGDFXAEGGGVR*4-androsten-3beta,17beta-diol disulfate 2, Gamma-glutamylisoleucine*Alpha-ketoglutarate, Mannitol) were found to reduce AKI risk.Conversely, four plasma metabolites—Chiro-inositol, Glycodeoxycholate, 1,6-anhydroglucose, and 1-oleoylglycerophosphoethanolamine—were identified as risk factors for AKI (Figure 5). Subsequent MR analyses revealed that CX3CL1 influences levels of 1,6-anhydroglucose, ADpSGEGDFXAEGGGVR*and Mannitol.

Figure 5 Heatmap of causal metabolites and inflammatory factors associated with AKI.Heatmap showing the P values of Mendelian randomization (MR) analyses for inflammatory factors and plasma metabolites associated with acute kidney injury (AKI) across five MR methods, including IVW, MR-Egger, simple mode, weighted median, and weighted mode. Color intensity represents the significance level of the corresponding P values.

Based on these results, mediation analysis was performed to evaluate the mediating role of specific metabolites in the causal relationship between CX3C motif chemokine ligand 1 (CX3CL1) and acute kidney injury (AKI), thereby exploring potential biological pathways. The results are summarized in Table 1.Three potential “CX3CL1–metabolite–AKI” mediation pathways were identified. Among them, 1,6-anhydroglucose exhibited a partial mediation effect in the CX3CL1–AKI causal pathway (β = −0.020, 95% CI: −0.054 to 0.005, P = 0.089), with a mediation proportion of 23.39% (95% CI: −4.66% to 51.44%), suggesting this carbohydrate metabolite may partially mediate CX3CL1’s effect on AKI risk. Additionally, the peptide ADpSGEGDFXAEGGGVR showed a significant positive mediation effect (β = 0.034, 95% CI: 0.003 to 0.072, P = 0.067) in the CX3CL1–AKI causal pathway, despite the total effect being negative. This resulted in a negative mediation proportion (−32.50%, 95% CI: −68.30% to 3.30%), indicating possible compensatory effects or competitive mediation mechanisms, warranting further mechanistic investigation. The third mediation pathway involved Mannitol, which showed a certain negative mediation effect (β = −0.017, 95% CI: −0.040 to 0.007, P = 0.147), with a mediation proportion of 15.74% (95% CI: −6.73% to 38.20%). Although the P-value did not reach statistical significance, the mediation trend is noteworthy. In summary, this study preliminarily identified multiple metabolites potentially involved in CX3CL1-mediated pathogenesis of AKI, highlighting their possible mediating roles in immune-inflammatory mechanisms underlying AKI.

Table 1 Mediation Analysis of Plasma Metabolites in the CX3CL1-AKI Pathway

CX3CL1 and UMOD are Candidate Biomarkers of I/R-induced AKI

Previous studies have identified CX3CL1 and UMOD as risk factors for AKI.We next performed clinical validation in patients undergoing cardiac surgery with cardiopulmonary bypass.It is well known that cardiac surgery can cause ischemic acute kidney injury (AKI). In this study, 100 post-cardiac surgery patients were enrolled, divided into the non-AKI group (n = 50) and the AKI group (n = 50), with a healthy control group (n = 20) serving as controls. Blood samples were collected within 24 hours post-surgery. The baseline characteristics of all patients are presented in Table 2. Notably, the average ages of AKI and non-AKI patients were 56.12 ± 11.23 years and 58.60 ± 10.27 years, respectively. In AKI group, 26 (52.0%), 13 (26.0%) and 11 patients (22.0%) belonged to AKI stages 1, 2, and 3, respectively. The duration of cardiac surgery,the length of hospitalization, and the cost of hospitalization were higher in the AKI group compared to the non-AKI group (P< 0.05). To assess whether CX3CL1 and UMOD were candidate biomarkers for cardiac surgery patients with ischemic AKI, ELISA results indicated that the plasma levels of CX3CL1 and UMOD were slightly elevated in non-AKI patients compared with the healthy group, and those in AKI patients were noticeably increased compared to non-AKI patients (Figures 6a and b). Furthermore, the plasma concentrations of CX3CL1 and UMOD were markedly increased in patients with Kidney Disease: Improving Global Outcomes (KDIGO) AKI stage III compared to those with KDIGO AKI stage I(P<0.05). In patients with KDIGO AKI stage III, these concentrations were not significantly increased compared to those with KDIGO AKI stage II (P>0.05) (Figures 6c and d). Additionally, the plasma levels of CX3CL1 and UMOD were remarkably correlated with sCr in AKI group (r = 0.848, P < 0.001; r = 0.794, P< 0.001, respectively) (Figure 6e and f). At the same time, we found that the elevated level of plasma CX3CL1 was consistent with the changing trend of plasma UMOD in AKI group (r = 0.787, P < 0.001) (Figure 6g). Our analysis confirmed that plasma CX3CL1 and UMOD were remarkably had a strong correlation with urinary [TIMP2]*[IGFBP7] in AKI group (r = 0.78, P < 0.001;r = 0.76, P< 0.001, respectively)(Figure 6h and i). To further evaluate the diagnostic value of plasma CX3CL1and UMOD, receiver operating characteristic (ROC) curves were established. As displayed in Figure 6j, the area under the ROC curve (AUC) of plasma CX3CL1 was 0.89 (95% CI, 0.808–0.949) with the sensitivity and specificity of 68.0% and 92.0%, respectively, at a cut-off value of 206.5 pg/mL. The AUC of plasma UMOD was 0.81 (95% CI, 0.730–0.892) with the lower sensitivity of 60.0% and specificity of 88.0% compared to CX3CL1 level (Figures 6k). The data analysis suggested that CX3CL1 and UMOD may have potential utility in I/R-induced AKI risk assessment.

Table 2 The Baseline Clinical Characteristics and Laboratory Values of the Patients

Figure 6 CX3CL1 and UMOD are candidate biomarkers of I/R-induced AKI. (a) The level of plasma CX3CL1 in different group of people (Healthy group n=20;non-AKI group n=50;AKI group n=50). (b) The level of plasma UMOD in different group of people (Healthy group n=20;non-AKI group n=50;AKI group n=50).(c) The level of plasma CX3CL1 in different stages of AKI (Stage I n=26;Stage II n=13;Stage III n=11;). (d) The level of plasma UMOD in different stages of AKI. (e) Spearman correlation between serum creatinine and plasma CX3CL1 levels in patients with AKI (n=50). (f)Spearman correlation between serum creatinine and plasma UMOD levels in patients with AKI (n=50). (g)Spearman correlation between plasma CX3CL1 and plasma UMOD levels in patients with AKI (n=50). (h)Spearman correlation between plasma CX3CL1 and urinary [TIMP2] × [IGFBP7] levels in patients with AKI (n=50). (i)Spearman correlation between plasma UMOD and urinary [TIMP2] × [IGFBP7] levels in patients with AKI (n=50).(j)Receiver operating characteristic (ROC) curve analysis of plasma CX3CL1 for identifying AKI among patients undergoing cardiac surgery (n=100).(k)Receiver operating characteristic (ROC) curve analysis of plasma UMOD for identifying AKI among patients undergoing cardiac surgery (n=100).Data are presented as Mean±SD, Correlation analyses were performed using Spearman’s rank correlation test. Diagnostic performance was evaluated using ROC curve analysis, and the area under the curve (AUC) with corresponding 95% confidence intervals is presented.#P< 0.05, non-AKI or AKI vs. Healthy group; *P < 0.05, AKI vs. non-AKI group; stage III of AKI group vs. stage I.

Discussion

Acute kidney injury (AKI) is a common and severe clinical syndrome characterized by high incidence and mortality rates. Currently, therapeutic options for AKI are extremely limited, and clinical management primarily relies on supportive care.1 Despite extensive research exploring potential molecular mechanisms and intervention targets, most findings have failed to translate into effective drugs. One major reason is the lack of systematic evidence supporting genetically inferred associations between protein biomarkers and disease outcomes. To address this gap, the present study systematically investigated candidate proteins and genetically supported pathways associated with AKI from a proteomic perspective by integrating large-scale multi-omics data. We applied multiple advanced analytical methods, including summary-data-based Mendelian randomization (SMR), Bayesian colocalization analysis, two-sample Mendelian randomization (MR), mediation analysis, and phenome-wide association study (PheWAS), to identify and validate candidate plasma proteins involved in AKI pathogenesis. Ultimately, we identified two plasma proteins, CX3CL1 and UMOD, with significant causal associations with AKI, and further validated their robustness and explored underlying mechanisms using independent datasets. These findings offer strong biological rationale and translational potential for future studies on AKI biomarkers and pathways.

CX3CL1, also known as Fractalkine, is a chemokine that interacts with its specific receptor CX3CR1 and plays an important role in renal inflammation and injury.30,31 The CX3CL1/CX3CR1 axis has recently gained attention as a potential therapeutic target because it participates in the individual development, homeostatic migration, and colonization of kidney phagocytes.32 Previous animal and clinical studies have demonstrated significant upregulation of the CX3CL1/CX3CR1 axis in various kidney diseases such as ischemia-reperfusion injury, drug-induced nephrotoxicity, and diabetic nephropathy. This axis promotes recruitment and activation of inflammatory cells including monocytes and T cells, ultimately exacerbating renal parenchymal damage.33,34 Moreover, Gong et al published a study in Molecular Medicine showing that in a cisplatin-induced AKI mouse model, CX3CL1 levels were significantly elevated. Knockout of CX3CL1 effectively alleviated podocyte ferroptosis, improved mitochondrial function, and suppressed inflammatory responses, indicating a clear pathogenic role of CX3CL1 in cisplatin-induced AKI (Cis-AKI).35 This research not only elucidated the mechanistic link between CX3CL1 and ferroptosis, oxidative stress, and endoplasmic reticulum stress, but also provided an experimental basis for targeting CX3CL1 in drug-induced kidney injury.In our study, CX3CL1 showed a significant causal association with AKI in SMR analysis (OR = 1.323), and Bayesian colocalization analysis yielded a high posterior probability (PP.H4 = 96.2%), indicating strong overlap between the genetic regulatory locus of CX3CL1 and AKI risk. Furthermore, two-sample MR analysis in external datasets replicated the positive causal relationship, with no significant heterogeneity or pleiotropy detected, demonstrating high reliability of the results.

Uromodulin (UMOD) is the most abundant glycoprotein in the kidney, predominantly secreted by the epithelial cells of the distal convoluted tubules and the thick ascending limb of the loop of Henle. It mainly exists in urine in a secreted form, with a small fraction present in circulation as soluble UMOD (sUMOD).36 Previous studies generally recognize that urinary UMOD exerts multiple renoprotective effects, including regulation of ion channel function in the loop of Henle, reduction of inflammatory responses, and attenuation of oxidative stress, thereby partially mitigating the occurrence and progression of acute kidney injury (AKI).36,37 However, the role of circulating UMOD in AKI remains controversial. Clinical studies evaluating the predictive value of circulating UMOD for AKI risk are limited and have yielded inconsistent results. Vonbrunn et al reported that sUMOD levels transiently increase in AKI animal models, suggesting its potential as an early biomarker for ischemic AKI.38 In contrast, Kuśnierz-Cabala et al found that sUMOD levels were decreased in patients with acute pancreatitis complicated by AKI.39 These conflicting findings imply that sUMOD may have stage-dependent and dynamic biological functions in vivo, with potentially different mechanisms and sources under varying disease contexts or stages of kidney injury. In this study, we systematically evaluated the causal relationship between plasma UMOD and AKI from a genetic perspective for the first time. By integrating protein quantitative trait loci (pQTL) data with AKI genome-wide association study (GWAS) data, we found that elevated plasma UMOD levels were significantly positively associated with AKI risk, supported by clear genetic colocalization evidence. This result provides strong causal evidence for the involvement of UMOD in the pathological process of AKI, suggesting that UMOD may not only serve as a biomarker but also play a pathogenic or regulatory role in AKI development. Our findings contribute to a better understanding of the bidirectional regulatory mechanisms of UMOD in different body fluid compartments related to AKI and offer potential targets and theoretical bases for future development of UMOD-related intervention strategies.

In further mediation analyses, we found that CX3CL1 may partially mediate its effect on AKI through multiple metabolites, such as 1,6-anhydroglucose and mannitol. 1,6-Anhydroglucose, a glucose derivative, has recently been implicated in disorders of glucose metabolism and oxidative stress.40,41 In our mediation model, this metabolite exhibited a modest mediating effect in the “CX3CL1–AKI” pathway (β = –0.020), accounting for 23.39% of the total effect. Additionally, mannitol also showed a potential mediating trend. Although the statistical significance was slightly lower, its dual role in AKI—as an osmotic diuretic and, under certain conditions, a potential exacerbator of renal injury—warrants further investigation. To assess potential pleiotropic effects of these target proteins on other diseases, we performed a phenome-wide association study (PheWAS). Among 2,469 phenotypes, CX3CL1 showed significant associations with hypertension, meningococcal infection, and other conditions, while UMOD was linked to carbohydrate metabolism disorders and tonsillitis. These findings indicate that although targeting these proteins may effectively intervene in AKI progression, cautious evaluation of their broad phenotypic effects is necessary during drug development to avoid off-target effects or immune-related adverse reactions.The strengths of this study include: (1) the use of multi-stage, mutually validating genetic epidemiological methods to enhance the reliability of causal inference; (2) integration of large-scale, multi-omics datasets to increase statistical power and generalizability; (3) incorporation of mediation MR analyses to further elucidate potential immune-metabolic pathways, broadening the mechanistic understanding of AKI; (4) combination with PheWAS analyses to provide prospective insights into the clinical safety of candidate biomarkers.

This study provides preliminary clinical evidence supporting the potential association of plasma CX3CL1 and UMOD with I/R-induced AKI.Research indicates that during ischemic AKI, CX3CL1 protein expression is substantially elevated within the kidney’s vascular system, reaching its maximum level 24 hours following ischemia-reperfusion injury, a pattern similar to that seen in our study.42 In ischemic AKI, serum UMOD shows a brief increase within 24 hours post ischemia-reperfusion, which may then be followed by a declining trend. This apparently contradictory phenomenon may occur because renal tubular cell injury causes UMOD to be released from cells into the blood, while the same injury leads to decreased UMOD synthesis.38,43 However,the biological role of UMOD in AKI appears to be complex and compartment-dependent. Previous studies have suggested that urinary UMOD is primarily secreted into the tubular lumen and is generally regarded as a marker of tubular integrity, nephron mass, and preserved renal function. Reduced urinary UMOD levels have frequently been associated with tubular injury and worse kidney outcomes. In contrast, circulating UMOD originates mainly from basolateral secretion and may reflect more complex biological processes, including systemic inflammation, renal injury responses, and kidney functional reserve. Consequently, circulating and urinary UMOD may exhibit distinct, and sometimes seemingly contradictory, associations with AKI severity and prognosis. In addition, discrepancies among previous studies may also be related to differences in patient populations, disease stages, sample types, and analytical methods.Furthermore, because the present study focused on circulating UMOD rather than urinary UMOD, our findings should not be directly extrapolated to urinary UMOD-associated mechanisms. Future studies simultaneously evaluating circulating and urinary UMOD are needed to clarify their respective biological and clinical roles in AKI.The diagnostic performance, as quantified by the area und