https://orcid.org/0009-0008-4708-3583
Figures
Abstract
Background
A reduction in kidney function cause metabolites that normally are cleared by the kidney to accumulate. These accumulated metabolites could potentially link kidney function to an increased risk of cardiovascular disease.
Aim
To assess the relationships between creatinine-based estimated glomerular filtration rate (eGFR) and the metabolome (791 endogenous metabolites) in the general population, as well as the cortisol to cortisone ratio.
Patients and methods
Three population-based cohorts, Epihealth, PIVUS and POEM (total n = 3,444), were used with a discovery/validation approach. Metabolomics was measured by mass spectroscopy. eGFR was calculated using the 2021 CKD-EPI creatinine-based equation. Multivariable linear regression was used to assess the correlations between each metabolite concentration and eGFR, adjusting for sex, % fat mass, cardiovascular risk factors and medications. The metabolites significant in the linear regression were analyzed using MultiVariable Mendelian Randomization (MVMR), adjusting for diabetes and BMI.
Results
Nighty-six metabolites were significantly associated to eGFR and had the same direction of association in both linear regression and MVMR. The vast majority of these associations were negative. Apart from creatinine, N-acetylalanine, N,N-dimethyl-pro-pro, N,N,N-trimethyl-alanylproline betaine (TMAP) were the top findings. Pathway enrichment analysis (Metaboloanalyst 6.0) found five significantly enriched pathways, two of which involved amino acid metabolism.
Conclusions
In a general population with only mild to moderately reduced kidney function, several metabolites were associated with kidney function, including the cortisol to cortisone ratio. This finding is of interest since several of these metabolites might link reduced kidney function to an increased risk of cardiovascular disease.
Citation: Ljunggren J, Delgado-Velandia M, Sundström J, Engström G, Smith JG, Ärnlöv J, et al. (2026) Impact of kidney function on the metabolome in the general population. PLoS One 21(6): e0347652. https://doi.org/10.1371/journal.pone.0347652
Editor: Anil Bhatia, University of California Riverside, UNITED STATES OF AMERICA
Received: July 13, 2025; Accepted: April 6, 2026; Published: June 30, 2026
Copyright: © 2026 Ljunggren et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Data Availability: Due to Swedish laws on personal integrity and health data, as well as the decision by the Ethics Committee, we are not allowed to make any data including health variables open to the public, even if made anonymous. The data could be shared with other researchers after a request to the Head of Department (josef.jarhult@medsci.uu.se).
Funding: The EpiHealth study is funded as a strategic research area by the Swedish government. PIVUS80 and POEM were funded by Uppsala University Hospital. In addition, the Uppsala University Hospital ALF grants and establishment funding (MKS), Swedish National Strategic Research Initiative EXODIAB (Excellence of Diabetes Research in Sweden), the Swedish Kidney Foundation (Njurfonden) and the Family Ernfors Foundation funded this study.
Competing interests: The authors have declared that no competing interests exist.
Background
The prevalence of chronic kidney disease (CKD) is increasing and currently afflicts about 10% of the general population [1,2]. This increase is in part explained by a growing prevalence of risk factors for CKD, the most important being older age, obesity, hypertension and diabetes. [1,2]
A reduction in glomerular filtration rate (GFR) causes metabolites that normally are cleared by the kidney to accumulate, via a reduction in plasma clearance. These changes are of interest since such metabolites might link CKD to an increased risk of cardiovascular disease (CVD) which is the leading cause of death in patients with CKD [3]. One potential mechanism could be endothelial damage [4–6]. Trimethylamine-N-oxide (TMAO) is a metabolite linked to both CKD and endothelial dysfunction where circulating levels have been shown to correlate with increased risk for myocardial infarction, stroke and death [7,8].
Previous studies utilizing metabolomics have found that several metabolites whose concentrations are associated with kidney function [1,5–7,9–14]. Examples of such metabolites include n-acetylated amino acids, which normally are metabolized in the kidney [12], metabolites involved in tryptophan metabolism [7], pseudouridine [1,6,9,11], erythronate [9], TMAO [7], and N,N,N-trimetyl-alanylproline betaine (TMAP) [13]. The majority of these metabolites increase when kidney function decline but not all studies are in agreement, with metabolites being significantly related to kidney function in some, but not all, studies [1,5–7,9–14].
Another metabolite of interest is cortisol and its ratio to cortisone. This ratio is skewed towards cortisol in CKD and an increased ratio has been shown to be a risk factor for CVD and mortality [15]. An increase in the ratio of cortisol to cortisone in CKD can be explained by several mechanisms, including an altered conversion between cortisol and cortisone and accumulation of cortisol metabolites [15].
Obesity is associated with the development of CKD [2] and has also been shown to induce profound changes in the metabolome [16]. Thus, fat mass affects kidney function, the metabolome, and also the estimation of kidney function (GFR) [17]. To our knowledge, previous studies on the association between kidney function and metabolites have not accounted for fat mass.
The aim of the present study was to assess the relationship between creatinine-based estimated glomerular filtration rate (eGFR) and the circulating metabolome (using 791 endogenous metabolites) in a general population with mild to moderately reduced kidney function. In addition, the relationship between creatinine-based eGFR and the cortisol to cortisone ratio was investigated. This was done in three population-based cohorts; Epihealth, POEM and PIVUS [18–21], employing a discovery/validation approach to minimize false positive findings. In addition, to obtain an estimate of the causal effect of kidney function on metabolite concentration in the blood, multivariable Mendelian randomization (MVMR) analysis was performed for relationships being significant in the observational setting.
Subjects and methods
Subjects
Three different population-based cohorts were used:
EpiHealth. The Epidemiology for Health study (EpiHealth) has been described elsewhere [18,19]. Starting in 2011, inhabitants in the cities of Uppsala and Malmö between the ages of 45 and 75 were randomly invited to participate in health survey. In 2018, a total of 25,000 patients had been recruited. The aim of the study was to investigate the connections between lifestyle factors, as well as genetic factors, and common diseases. Metabolomic data were analyzed from the first invited 2,342 subjects in the Uppsala sample. These individuals were used as the discovery cohort in this study.
PIVUS. The Prospective Study of the Vasculature in Uppsala Seniors (PIVUS) has also been described in detail elsewhere [18,21]. PIVUS was created with the goal to investigate if endothelial dysfunction implied increased cardiovascular risk. Starting in 2001, individuals in the city of Uppsala were recruited at the age of 70 and were then followed up at ages 75 and 80 years. Initially there were 1,016 individuals, out of which 50% were women. At the follow-up at age 80 (PIVUS80) 600 individuals attended. A DXA scan and an analysis of the metabolome were parts of the 80-year examination.
POEM. The Prospective investigation of Obesity, Energy and Metabolism (POEM) has also been described elsewhere [18,20]. In POEM, 502 individuals were recruited in the city of Uppsala between 2011 and 2016. All individuals were 50 years old and 50% were female. The aim of the study was to investigate how obesity affected cardiovascular disease.
In this study, the results from PIVUS80 and POEM were combined by meta-analysis and used as the validation cohort.
SCAPIS (Swedish CArdioPulmonary Imaging Study)
Between 2013 and 2018, a total of 30,154 men and women between the ages of 50 and 65 were recruited from six Swedish cites. The project was a collaboration between six universities in Sweden. [18] The participants underwent quantification of their cardiovascular risk factors, computed tomography coronary angiography, carotid artery ultrasound examination and further extensive imaging. For 4,985 participants in Uppsala and 3,978 participants in Malmö, metabolomic measurements on plasma were performed.
All studies were conducted with the informed consent from all participants, with approval from the responsible Ethics Committees in accordance with the principles of the Declaration of Helsinki (dnr 2010/402, 2011/045, 2009/057 and 2024-05807-01). The data were most recently accessed for research purposes between 2024-08-10 and 2025-06-27 and the authors had full access to information that could identify individual participants during that time period.
Metabolomics
In all three study samples, non-targeted metabolomics (Metabolon inc., USA) was performed on plasma samples being stored at −80℃. Samples were prepared using the automated MicroLab STAR® system from Hamilton Company. Several internal standards were added prior to the first step in the extraction process for QC purposes. To remove protein, dissociate small molecules bound to protein or trapped in the precipitated protein matrix, and to recover chemically diverse metabolites, proteins were precipitated with methanol under vigorous shaking for 2 min (Glen Mills GenoGrinder 2000) followed by centrifugation. The resulting extract was divided into five fractions: two for analysis by two separate reverse phases (RP)/UPLC-MS/MS methods with positive ion mode electrospray ionization (ESI), one for analysis by RP/UPLC-MS/MS with negative ion mode ESI, one for analysis by hydrophilic interaction (HILIC)/UPLC-MS/MS with negative ion mode ESI, and one sample was reserved for backup. Only annotated, non-xenobiotic metabolites with a detection rate >75% in all samples were used in the analyses (n = 791). The values were normalized and given in arbitrary units.
Compounds were identified by comparison to library entries of purified standards or recurrent unknown entities. Metabolon maintains a library based on >3,000 authenticated standards that contains the retention time/index (RI), mass to charge ratio (m/z), and chromatographic data (including MS/MS spectral data) on all molecules present in the library.
Several types of controls were analyzed in concert with the experimental samples: a pooled matrix sample generated by taking a small volume of each experimental sample (or alternatively, use of a pool of well-characterized human plasma) served as a technical replicate throughout the data set; extracted water samples served as process blanks; and a cocktail of QC standards that were carefully chosen not to interfere with the measurement of endogenous compounds were spiked into every analyzed sample, allowed instrument performance monitoring and aided chromatographic alignment.
Instrument variability was 5% as determined by calculating the median relative standard deviation (RSD) for the internal standards that were added to each sample prior to injection into the mass spectrometers. Overall process variability was 7% as determined by calculating the median RSD for all endogenous metabolites (i.e., non-instrument standards) present in 100% of the Client Matrix samples, which are technical replicates of pooled client samples. In a published comparison between the 4 MS platforms used, the average laboratory coefficient of variation (CV) on the 4 platforms was between 9.3 and 11.5%, average inter-assay CV ranged from 9.9 to 12.6% and average intra-assay CV ranged from 5.7 to 6.9% [22].
Kidney function
In both POEM and PIVUS80, plasma creatinine was measured by standard clinical laboratory methods. Creatinine was analyzed on Roche Cobas Pro instruments using an IDMS-calibrated enzymatic method for quantification. Results are then used to compute estimated glomerular filtration rates (eGFR). In Sweden, national IDMS calibration of creatinine assays was implemented in 2004 and all laboratories participate in external quality assurance programs for both creatinine and cystatin C (Equalis.se), ensuring consistent performance.
Such measurements were not performed in EpiHealth. However, creatinine was one of the metabolites measured in the metabolomic analysis, but the results are given on a relative scale. Since both PIVUS80 and POEM contained plasma creatinine measured by both metabolomics and standard methods, we fitted a linear regression model between the creatinine values obtained by the two different methods. The model obtained was creatinine by standard methods (µmol/L) = −28 + 107*metabolomics measured creatinine. This linear model was then used to estimate plasma creatinine in absolute values in EpiHealth. eGFR was thereafter calculated using the 2021 CKD-EPI creatinine-based equation [23] in all three cohorts.
Dietary measurements
MiniMealQ, a comprehensive food frequency questionnaire, was a part of the internet-based questionnaire given to the participants of the EpiHealth study [24]. Including common Swedish beverages, foods and dishes, the MiniMealQ questionnaire covers the diet over the past few months. Depending on the food, either a five-level or eight-level ordinary scale was used to quantify the intake of the dietary items in the questionnaire. The five-level scale spanned from once to thrice monthly to more than seven times per week (vegetables, fruits, baked goods, and different dishes were estimated on this scale). The eight-level scale spanned from once to twice weekly to five or more times per day (example of products reported on this scale was dairy- and bread products). For alcohol, a six-level scale was used: spanning from never to four or more times weekly. From these data, the daily energy intake and protein intake were calculated.
Fat mass and body weight
Fat mass and body weight was given from a weight scale that also calculates fat mass by mean of bioimpedance (Tanita, Tokyo, Japan) in all three cohorts used for the observational analysis. Fat mass percentage is fat mass divided by body weight.
Genotyping
The extraction of DNA from the samples in SCAPIS was performed at the Karolinska Institutet Biobank. Utilizing Illumina Infinium Global Screening Array Multiple Disease version 3 and the SNP&SEQ Technology Platform, the samples were genotyped in 10 batches. For the unrelated samples, Principal component analysis (PCA) was performed and onto these components were projected all the samples. PLINK was used to make the PCA. All samples and markers with >2% missing data, with failed sex checks, with heterozygosity measures >3 standard deviations, with non-European ancestries, or with minor allele frequency < 0.1% and Hardy-Weinberg equilibrium P < 1x10-8 were removed as part of Pre-imputation quality control. Then, utilizing the “Pre-phasing and imputation with EAGLE2 + PBWT” pipeline, the Sanger Imputation Service was used to impute the SCAPIS genotype data to the HRCr 1.1.
A previous publication has described the genotyping of EpiHealth in detail [25]. In brief, from 400 µL EDTA whole blood, DNA was extracted and then dissolved in 145 µL 10 mmol/L Tris-HCl buffer (pH 8.0). The examination for quantity and purity was done by measuring absorbance at 230, 260 and 280 nm. The genotyping was performed with Illumina HumanCoreExome-12 v.1.0 BeadChip (Illumina, San Diego, CA, USA), which included 522,731 autosomal markers. Quality control was then performed, and after that 2378 samples remained, and their genotyping data was imputed up to the 1000 Genomes phase 3 (v5) reference panel. In the final dataset, about 12 million markers, with a minor allele count >= 1 were included.
Genome-wide association of metabolites in SCAPIS and EpiHealth.
Using 777 metabolites measured in both EpiHealth (N = 2,276) and SCAPIS (N = 8,133), applying linear mixed models in REGENIE, genome wide association tests were performed [26]. The genetic additive model used in the analysis was adjusted for genetic principal components 1–10, sex, age, age^2, Metabolon instrument batch in SCAPIS, and for genetic principal components 1–10, sex, age in EpiHealth. METAL version 2011-03-25 was used to perform a fixed-effects meta-analysis, based on inverse-variance weighting, of the association statistics of EpiHealth and SCAPIS [27].
Genetic instruments of estimated glomerular filtration rate (eGFR), BMI and type 2 diabetes mellitus
We obtained summary-level GWAS statistics from a GWAS meta-analysis for serum creatinine eGFR (n = 1,004,040) including European-only data from the Chronic Kidney Disease Genetics Consortium and UK Biobank (PMID: 34272381). Likewise, European-only, summary-level GWAS statistics for BMI and T2DM were obtained from the UK Biobank (n = 405,010) [28], and the Diabetes Genetics Replication and Meta-analysis Consortium (DIAGRAM, n = 1,812,017) [29], respectively.
SNPs with a p-value < 5x10-8 for any of the three exposures were selected as instruments. In total, 173,848 instruments were found. Instruments were clumped remotely using the OpenGWAS server through the clump_data function (TwoSampleMR package, version 0.6.29), with default parameters: clumping p1 = 0.99, clumping R2 0.001, clumping window 10,000 kb, and European super-population as reference panel, leaving 860 instruments. Afterward, instruments with incomplete information for any of the exposures were removed, leaving 709 instruments. All these 709 SNPs were found in our in-house GWAS of plasma metabolites, but after data harmonization, removal of SNPs with incompatible alleles and palindromic SNPs with intermediate allele frequencies, 703 SNPs remained. The final number of instruments available for MR analyses was 190 for creatinine-based eGFR, 255 for body mass index (BMI) and 289 for type 2 diabetes mellitus (T2DM).
Statistical methods
Observational analysis.
In order to make the distributions of the metabolites and eGFR standardized to the Z-scale and given a normal distribution, rank-based inverse normal transformation was used.
A discovery/validation approach was used with the EpiHealth cohort used as the discovery sample and a meta-analysis of POEM and PIVUS80 as the validation sample. Both the discovery and the validation analysis utilized multivariable linear regression. Multivariable linear regression models were fitted for each metabolite vs eGFR. The dependent variable was always the metabolite. In the discovery stage, eGFR, age, sex, non-fat mass, protein intake and energy intake were the independent variables. The first validation model (model 1) used eGFR and sex as independent variables. Since the age was 50 in all individuals in POEM and 80 in PIVUS80, age was not used as an independent variable. Validation model 2 used eGFR, sex and percent fat mass as independent variables, and validation model 3 used eGFR, sex, percent fat mass, systolic blood pressure, diabetes, number of years smoked, physical activity, use of statins and antihypertensive drugs as independent variables. The regression analyses of the validation cohorts were done separately. The results from the two cohorts were thereafter combined by an inverse-variance weighted (IVW) random effects meta-analysis.
For each model, participants with missing data in any of the variables were removed from the analysis for that model.
The p-values from the discovery cohort were adjusted using Benjamin-Hochberg to get the false discovery rate (FDR). The metabolites were deemed to be significant if they showed FDR < 0.05 in both discovery and validation for each of the models, respectively.
Mendelian randomization.
We performed a Multivariable Mendelian randomization (MVMR) on the association of creatinine-based eGFR with plasma metabolites adjusting for BMI and T2DM using SNP level summary data obtained from European populations with no known participant overlap between exposures and outcomes (Description of GWAS consortiums used is found in the supplemental materials, table in S5 Table). BMI and T2DM were used as covariates in the MVMR model, since BMI and T2DM were regarded as the most powerful confounders in the relationship between eGFR and the metabolites according to observational findings. Effect estimates were obtained using inverse variance weighting (IVW) and represent the effect of one standard deviation change in natural log transformed eGFR levels (ml/min/1.73 m2). In the supplemental materials, table in S6 Table, the instrumental strength (given as the F-value) is given for each SNP used in the MVMR. Only metabolites that were significant in the observational setting (model 3) were investigated with MVMR.
The resulting associations were corrected using FDR < 0.05. Only metabolites with the same direction of association in the MVMR and in the observational analysis were regarded as valid.
Analyses for the observational section were performed in R. Multivariable Mendelian randomization analyses were performed in Stata (version 16.1) using the mrmvivw command from the mrrobust package (version 0.3.0) (doi.org/10.1093/ije/dyy195).
Pathway enrichment analysis
The metabolites that were significant and had the same direction in both validation model 3 and MVMR were used for a pathway enrichment analysis using MetaboAnalyst (https://www.metaboanalyst.ca/MetaboAnalyst/ModuleView.xhtml). Here, a Hypergeometric test with relative betweenness centrality was used. The selected library was the Homo sapiens KEGG library. The significance level was set to a nominal p < 0.05.
Results
Clinical and biochemical characteristics of study participants
The basic clinical and biochemical characteristics of study participants in EpiHealth, PIVUS80, and POEM are presented in Table 1. The major difference between the cohorts was age. In Epihealth mean age of the participants was 61 years, while in PIVUS80 all participants were 80 and in POEM 50 years. Study participants of higher age had higher systolic blood pressure, were more often treated with statins and antihypertensive medication, and had a lower eGFR. The average eGFR for each cohort was within two standard deviations of the expected value based on their average age [30,31].
Because of missing data, 53 participants were removed from the analysis of model 2 and 176 participants were removed from model 3. Thus, the number of participants were 1102 in model 1, 1049 in model 2, and 926 in model 3.
Linear regression between metabolites and kidney function (eGFR)
The workflow of this paper is described in Fig 1. Out of 791 metabolites analyzed, 254 had a significant regression coefficient for the eGFR variable at FDR < 0.05 in model 1with eGFR and sex as independent variables. Six of these coefficients were positive while the remaining were negative (see supplemental materials, table in S1 Table). In Table 2, the 15 metabolites with the largest negative coefficients are shown. These were creatinine, N,N-dimethyl-pro-pro, N,N,N-trimethyl-alanylproline betaine (TMAP), N-acetylalanine, N-acetylserine, N-formylmethionine, pseudouridine, 1-methylhistidine, 2,3-dihydroxy-5-methylthio-4-pentenoate (DMTPA)*, 3-(3-amino-3-carboxypropyl)uridine*, hydroxyasparagine**, N-acetylthreonine, N6-carbamoylthreonyladenosine, C-glycosyltryptophan, and erythronate. In addition, the cortisol to cortisone ratio had, for the eGFR variable, a regression coefficient of −0.25, FDR = 5.69*10−4. The metabolites with a positive regression coefficient were creatine, alpha-ketobutyrate, 1-palmitoyl-2-palmitoleoyl-GPC (16:0/16:1)*, 1-palmitoyl-2-oleoyl-GPC (16:0/18:1), 1-stearoyl-2-oleoyl-GPC (18:0/18:1), and 2-aminobutyrate (see Table 2). The metabolite with the largest negative coefficient was creatinine, and the one with the largest positive coefficient was creatine.
In model 2, further adjusting for fat mass (%), 270 metabolites had a significant regression coefficient for eGFR, at FDR < 0.05, seven were positive and the rest negative. 29 metabolites were significant in model 2 but not in model 1, while 13 metabolites were significant in model 1, but not in model 2. In S2 Table, all metabolites being significant in model 2 are shown. In Table 3, the metabolites that were significant in model 2 but not in model 1 are listed. The metabolites with a positive regression coefficient in model 2 were creatine, alpha-ketobutyrate, aspartate, 1-palmitoyl-2-palmitoleoyl-GPC (16:0/16:1)*, 1-stearoyl-2-oleoyl-GPC (18:0/18:1), 1-palmitoyl-2-oleoyl-GPC (16:0/18:1), and 1,2-dipalmitoyl-GPC (16:0/16:0), se Table 4. Also, in model 2, the cortisol to cortisone ratio was strongly associated to eGFR (regression coefficient −0.24, FDR = 3.85*10−4). In model 2, the largest positive coefficient was for creatine. A different regression model was fit with independent variables eGFR, sex and body fat in kg instead of percent fat mass. The results from this analysis were similar to those of model 2 (data not shown).
In model 3, also adjusting for cardiovascular risk factors and medications, 312 metabolites had a significant coefficient for eGFR. Six of these were positive while the rest, 306, were negative. Compared with model 1, there were 70 metabolites being significant in model 3 but not in model 1, and 12 significant metabolites in model 1 that were not significant in model 3. Compared with model 2, there are 52 metabolites that are significant in model 3 that are not significant in model 2 and 10 that are significant in model 2 but not in model 3. A Venn diagram over the significant metabolites in each model is shown in Fig 2. All significant metabolites in model 3 are presented in S3 Table. The 15 metabolites with the largest negative regression coefficients in model 3 were creatinine, N-acetylalanine, N,N-dimethyl-pro-pro, N,N,N-trimethyl-alanylproline betaine (TMAP), N-acetylserine, pseudouridine, 3-(3-amino-3-carboxypropyl)uridine*, N-formylmethionine, 2,3-dihydroxy-5-methylthio-4-pentenoate (DMTPA)*, hydroxyasparagine**, C-glycosyltryptophan, N-acetylthreonine, N6-carbamoylthreonyladenosine, 1-methylhistidine, and 5,6-dihydrouridine. The metabolites in model 3 with a positive regression coefficient for eGFR were creatine, glutamate, aspartate, alpha-ketobutyrate, mannose, and S-methylcysteine, see Table 5. The cortisol to cortisone ratio had a regression coefficient for eGFR of approximately −0.26 and FDR = 4.79*10−3 in model 3. The metabolite with the largest negative regression coefficient for eGFR in model 3 was creatinine, and the one with largest positive coefficient was creatine.
Multivariable Mendelian randomization (MVMR)
MVMR analysis was performed for 310 out of the 312 significant metabolites in model 3. Out of these 98 were significant at FDR < 0.05 and out of those, 96 had the same direction of association in MVMR as in the observational analysis. The only metabolite with a positive direction of association was for creatine. The 15 metabolites with the largest negative coefficients in the MVMR analysis were: creatinine, 1-methylhistidine, N-acetylalanine, hydroxyasparagine**, N,N,N-trimethyl-alanylproline betaine (TMAP), homocitrulline, N,N-dimethyl-pro-pro, 3-(3-amino-3-carboxypropyl)uridine*, methionine sulfone, 2,3-dihydroxy-5-methylthio-4-pentenoate (DMTPA)*, 3-hydroxy-3-methylglutarate, N-acetylserine, N6-carbamoylthreonyladenosine, pseudouridine, and deoxycarnitine, see Table 6. The complete results are found in the table in S4 Table.
Pathway enrichment analysis (KEGG)
The 96 metabolites that were significant and had the same direction of association in the MVMR and observational analysis were used in a pathway enrichment analysis. In total, there were five significant pathways at p < 0.05, as shown in table 7. Of these five significantly enriched pathways, two involved amino acid metabolism.
Discussion
This is to our knowledge the largest comprehensive study exploring the relationship between kidney function (creatinine-based eGFR) and the metabolome in a general population with only mild to moderately reduced kidney function. In this study, we applied a robust two-stage design using discovery and validation cohorts, coupled with stringent multiplicity testing, to identify metabolites and explore their association with creatinine-based eGFR. This approach not only minimizes the risk of false discoveries, but also provides insights into the underlying biological processes by focusing on individual metabolites and their pathways. The number of validated metabolites found to be related to eGFR were 254 when adjusting for age and sex; 270 if fat mass was added, and 312 when adjustment also was done for traditional cardiovascular risk factors and medications. Mendelian randomization analysis revealed that 96 of these metabolites were significant and had the same direction of association as in the observational analysis.
Most of these metabolites were inversely associated with eGFR. This is not unexpected since the kidneys play a major part in the plasma clearance of metabolites. Neither was it surprising that the metabolite with the largest regression coefficient to eGFR estimated using creatinine was creatinine in itself, in this case thus serving as a positive control.
Many significant metabolites in this study have been described previously in relation to kidney function. For example, c-mannosyltryptophan, erythronate, pseudouridine, and n-acetylated amino acids were found by several other groups [1,6,7,9,11,12]. Several amino n-acetylated amino acids were associated with eGFR. This is also expected since metabolism of these compounds occurs mainly in the kidney [33]. 1-palmitoyl-2-oleoyl-GPC (16:0/18:1) and 1-stearoyl-2-oleoyl-GPC (18:0/18:1) are significant and positive in model 1 and model 2 but not in model 3.The cortisol to cortisone ratio was negatively related to eGFR in all linear regression models. This suggests that the activity of 11β-HSD type 2, an enzyme that converts active 11β-hydroxyglucocorticoids (cortisol) to their inactive 11-keto forms (cortisone) and thereby fine-tuning the activation of mineralocorticoid and glucocorticoid receptors, is reduced and correlated with kidney function, which also has been shown in previous studies [34–35]. A recent publication showed that a decline in 11β-HSD2 activity was observed with progressively lower eGFR in individuals spanning a wide spectrum of kidney function, including those with apparently normal kidney function [36]. This is in line with the findings in this present study. Cortisol in itself was not significantly associated with eGFR in any validation model. However, the average eGFR in these cohorts were relatively high [30,31], and it is possible that changes in cortisol itself are only detectable at lower values of eGFR.
The population-based cohorts used in this study displayed a higher mean eGFR than in most previous studies, supporting that these changes are present already in a general population with only mild to moderately reduced kidney function.
A previous study investigating metabolites associated with an increased risk of incident CVD in EpiHealth showed that 37 metabolites were correlated to an increased risk of CVD, ten of which were significantly related to eGFR in the multiple-adjusted model in the present study. Some were strongly associated to eGFR with three of the metabolites being in the top 15 most negatively associated metabolites in the MVMR analysis of the metabolites significant in both observational and MVMR: N-acetylalanine, N-acetylserine, and N6-carbamoylthreonyladenosine. Out of these, all had a hazard ratio below above 1 for incident CVD [37]. In a study by Pietzner et.al [38], they investigate the correlation between the metabolome and 27 non-comunicable diseases and mortality. Comparing the significant metabolites in model 3 and MVMR with the significant metabolites found by Pietzner et.al. for the outcomes: kidney disease, CVD and mortality show that 42 of the significant metabolites in model 3 were related to these outcomes. Of these were 21 related to kidney disease, 31 to arterial CVD (atrial fibrillation, heart failure, cerebrovascular infarction, ischemic heart disease, peripheral arterial disease and abdominal aorta aneurysm) and 40 with mortality. The majority of these were associated with both increased hazard ratio of the conditions (kidney disease, CVD and mortality) and inversely related to eGFR.
Future studies have to disentangle if some of these metabolites are involved in the known relationship between CKD and future CVD [39].
The pathway enrichment analysis was exploratory in nature. It found 5 enriched pathways. 2 of those pathways involve amino acid synthesis or metabolism as would be expected, but also pentose and glucoronate intraconversions, as well as ascorbate and aldarate metabolism were amongthe enriched pathways. The ascorbate and aldarate metabolism pathway, and the lysine degradation pathway have previously been shown to be associated to CKD progression [39]. When considering the link to CVD, the lysine degradation pathway has been shown to be enriched in the urine of patients with unstable angina [40]. Other studies of the metabolome that included pathway enrichment analysisin patients with CVD did not find these pathways to be enriched [37,41,42].
The ascorbate and aldarate metabolism pathway is vitamin C metabolism. Patients with CKD have lower vitamin C, primarily due to decreased intake caused by decreased appetite and dietary restrictions, as well as losses from dialysis treatment [43]. These causes are unlikely to be the reasons that these pathways were significantly enriched in this study, which investigated the general population. One function of vitamin C is as an antioxidant. CKD is associated with oxidative stress and vitamin C might counteract part of this effect [43]. However, in patients with decreased GFR, oxalate formed from breakdown of vitamin C can accumulate in the blood, which may cause cardiovascular damage [43].
Previous research has shown that the gut microbiota plays a role in plasma metabolomics [44]. Koen F. Dekkers and colleagues have developed the GUTSY atlas, which for example, quantifies the proportion of variance in specific plasma metabolites attributable to microbiome variations, calculated using nested 10-fold cross-validated ridge regressions. Using those results and applying them to the 96 metabolites that were significant and had the same direction of association in both model 3 and the MVMR analysis in this study showed that the variances of 91 metabolites were influenced by the microbiome. However, for most of these metabolites, the effect of the microbiome was small, and only 12 had ≥ 5% of their variance explained by the microbiome.
Creatinine levels were measured by mass spectroscopy as a part of the metabolomic evaluation in the EpiHealth cohort and were reported on a relative scale only. Therefore, the mean value for eGFR in the Epihealth study was estimated by a linear regression model using data from the POEM and PIVUS studies, in which creatinine was measured by both clinical chemistry methods (providing absolute values) and by the same mass spectroscopy method as in the EpiHealth cohort. The correlation between the clinical chemistry and metabolomic methods in PIVUS and POEM was high (0.92–0.93). It is important to notethat clinical chemistry measured creatinine was used in the validation step, thereby mitigating any error introduced by the mass spectroscopy measured creatinine in the discovery step.
Creatinine based eGFR was used because it is the dominantly used measurement in the clinical setting. Furthermore, other compounds that are used to estimate eGFR, such as Cystatin C, were not measured in the cohorts used in this study.
The observed changes in metabolite concentrations seen in this study are likely a result of decreased clearance rather than dysregulation of any specific metabolic pathway. The reason for this is that the eGFR values of the participants in this study are normal or only slightly reduced. Dysregulation of more specific metabolic pathways might emerge later in the course of CKD progression.
The metabolomic measurements, namely UPLC-MS/MS, was performed by a commercial company (Metabolon Inc, U.S.A). Therefore, some details of the metabolic measurement are trade secrets of the company and thushus, a fully detailed description is not publicly available.
The major strength of the present study is the use of a metabolomic platform that made it possible to evaluate a wide range of non-xenobiotic metabolites in these large population-based samples. This ensures high-quality data with accuracy and reliability of the metabolomic measurements across cohorts. In addition, the large sample sizes used in both the discovery and validation cohorts enhance the statistical power of the study, allowing for the detection of robust associations between eGFR and metabolites of interest. The use of population-based cohorts adds to the study’s external validity, improving the representativeness of the findings across different populations. We should however acknowledge that we have studied almost exclusively individuals with European ancestry, so future studies in other ethnic groups are warranted. Another limitation is that in the discovery cohort, creatinine was not measured using standard clinical chemistry methods, which could potentially introduce a small variability or bias in the assessment of kidney function. Additionally, the lack of data on most medications used by the participants represents a limitation.
In conclusion, in this study, to our knowledge, the largest comprehensive study of the relationship between kidney function and the metabolome in a general population with mild to moderately reduced kidney function, identified several cardiovascular disease-related metabolites related to CVD, including the cortisol to cortisone ratio, as being associated with kidney function. For most of these metabolites, the effect of the microbiome was small.
Supporting information
S5 Table. Description of GWAS consortiums used in the MVMR analysis.
https://doi.org/10.1371/journal.pone.0347652.s005
(XLSX)
S6 Table. The instrumental strength for the SNPs used in the MVMR analysis.
https://doi.org/10.1371/journal.pone.0347652.s006
(XLSX)
References
- 1. Peng H, Liu X, Ieong CA, Tou T, Tsai T, Zhu H, et al. A Metabolomics study of metabolites associated with the glomerular filtration rate. BMC Nephrol. 2023;24(1):105. pmid:37085754
- 2.
Vaidya SR, Aeddula NR. Chronic Kidney Disease. In: StatPearls [Internet]. StatPearls Publishing; 2022 [cited 2024 Jul 2]. Available from: https://www.ncbi.nlm.nih.gov/books/NBK535404/
- 3. Go AS, Chertow GM, Fan D, McCulloch CE, Hsu C yuan. Chronic Kidney Disease and the Risks of Death, Cardiovascular Events, and Hospitalization. N Engl J Med. 2004;351(13):1296–305.
- 4. Kwiatkowska I, Hermanowicz JM, Mysliwiec M, Pawlak D. Oxidative Storm Induced by Tryptophan Metabolites: Missing Link between Atherosclerosis and Chronic Kidney Disease. Oxid Med Cell Longev. 2020;2020:6656033. pmid:33456671
- 5. Dubin RF, Rhee EP. Proteomics and metabolomics in kidney disease, including insights into etiology, treatment, and prevention. Clin J Am Soc Nephrol. 2020;15(3):404–11.
- 6. Coresh J, Inker LA, Sang Y, Chen J, Shafi T, Post WS. Metabolomic profiling to improve glomerular filtration rate estimation: a proof-of-concept study. Nephrol Dial Transplant. 2019;34(5):825–33.
- 7. Kalim S, Rhee EP. An overview of renal metabolomics. Kidney Int. 2017;91(1):61–9. pmid:27692817
- 8. Thomas MS, Fernandez ML. Trimethylamine N-Oxide (TMAO), Diet and Cardiovascular Disease. Curr Atheroscler Rep. 2021;23(4):12.
- 9. Sekula P, Goek O-N, Quaye L, Barrios C, Levey AS, Römisch-Margl W, et al. A Metabolome-Wide Association Study of Kidney Function and Disease in the General Population. J Am Soc Nephrol. 2016;27(4):1175–88. pmid:26449609
- 10. Fino NF, Adingwupu OM, Coresh J, Greene T, Haaland B, Shlipak MG, et al. Evaluation of novel candidate filtration markers from a global metabolomic discovery for glomerular filtration rate estimation. Kidney Int. 2024;105(3):582–92. pmid:38006943
- 11. Peng H, Liu X, Aoieong C, Tou T, Tsai T, Ngai K, et al. Identification of Metabolite Markers Associated with Kidney Function. J Immunol Res. 2022;2022:6190333. pmid:35928631
- 12. Titan SM, Venturini G, Padilha K, Tavares G, Zatz R, Bensenor I, et al. Metabolites related to eGFR: Evaluation of candidate molecules for GFR estimation using untargeted metabolomics. Clin Chim Acta. 2019;489:242–8.
- 13. Velenosi TJ, Thomson BKA, Tonial NC, Rao Peters AAE, Mio MA, Lajoie GA, et al. Untargeted metabolomics reveals N, N, N-trimethyl-L-alanyl-L-proline betaine (TMAP) as a novel biomarker of kidney function. Sci Rep. 2019;9(1):6831.
- 14. Nkuipou-Kenfack E, Duranton F, Gayrard N, Argilés À, Lundin U, Weinberger KM, et al. Assessment of metabolomic and proteomic biomarkers in detection and prognosis of progression of renal function in chronic kidney disease. PLoS One. 2014;9(5):e96955. pmid:24817014
- 15. Sagmeister MS, Harper L, Hardy RS. Cortisol excess in chronic kidney disease - A review of changes and impact on mortality. Front Endocrinol (Lausanne). 2023;13:1075809. pmid:36733794
- 16. Ahmad S, Hammar U, Kennedy B, Salihovic S, Ganna A, Lind L, et al. Effect of General Adiposity and Central Body Fat Distribution on the Circulating Metabolome: A Multicohort Nontargeted Metabolomics Observational and Mendelian Randomization Study. Diabetes. 2022;71(2):329–39. pmid:34785567
- 17. Chang AR, Zafar W, Grams ME. Kidney Function in Obesity-Challenges in Indexing and Estimation. Adv Chronic Kidney Dis. 2018;25(1):31–40. pmid:29499884
- 18. Lind L. Population-based cardiovascular cohort studies in Uppsala. Ups J Med Sci. 2019;124(1):16–20. pmid:30278808
- 19. Lind L, Elmståhl S, Bergman E, Englund M, Lindberg E, Michaelsson K, et al. EpiHealth: a large population-based cohort study for investigation of gene-lifestyle interactions in the pathogenesis of common diseases. Eur J Epidemiol. 2013;28(2):189–97. pmid:23435790
- 20. Lind L, Strand R, Michaëlsson K, Ahlström H, Kullberg J. Voxel-wise Study of Cohort Associations in Whole-Body MRI: Application in Metabolic Syndrome and Its Components. Radiology. 2020;294(3):559–67. pmid:31891319
- 21. Lind L, Fors N, Hall J, Marttala K, Stenborg A. A comparison of three different methods to evaluate endothelium-dependent vasodilation in the elderly. Arterioscler Thromb Vasc Biol. 2005;25(11):2368–75.
- 22. Ford L, Kennedy AD, Goodman KD, Pappan KL, Evans AM, Miller LAD, et al. Precision of a clinical metabolomics profiling platform for use in the identification of inborn errors of metabolism. J Appl Lab Med. 2020;5(2):342–56. pmid:32445384
- 23. Inker LA, Eneanya ND, Coresh J, Tighiouart H, Wang D, Sang Y. New creatinine- and cystatin C–based equations to estimate GFR without race. N Engl J Med. 2021;385(19):1737–49.
- 24. Almqvist C, Adami H-O, Franks PW, Groop L, Ingelsson E, Kere J, et al. LifeGene--a large prospective population-based study of global relevance. Eur J Epidemiol. 2011;26(1):67–77. pmid:21104112
- 25. Beijer K, Nowak C, Sundström J, Ärnlöv J, Fall T, Lind L. In search of causal pathways in diabetes: a study using proteomics and genotyping data from a cross-sectional study. Diabetologia. 2019;62(11):1998–2006. pmid:31446444
- 26. Mbatchou J, Barnard L, Backman J, Marcketta A, Kosmicki JA, Ziyatdinov A, et al. Computationally efficient whole-genome regression for quantitative and binary traits. Nat Genet. 2021;53(7):1097–103. pmid:34017140
- 27. Willer CJ, Li Y, Abecasis GR. METAL: fast and efficient meta-analysis of genomewide association scans. Bioinformatics. 2010;26(17):2190–1. pmid:20616382
- 28. Loya H, Kalantzis G, Cooper F, Palamara PF. A scalable variational inference approach for increased mixed-model association power. Nat Genet. 2025;57(2):461–8. pmid:39789286
- 29. Suzuki K, Hatzikotoulas K, Southam L, Taylor HJ, Yin X, Lorenz KM, et al. Genetic drivers of heterogeneity in type 2 diabetes pathophysiology. Nature. 2024;627(8003):347–57. pmid:38374256
- 30. Fenton A, Montgomery E, Nightingale P, Peters AM, Sheerin N, Wroe AC, et al. Glomerular filtration rate: new age- and gender- specific reference ranges and thresholds for living kidney donation. BMC Nephrol. 2018;19(1):336. pmid:30466393
- 31. Wetzels JFM, Kiemeney LA, Swinkels DW, Willems HL, den Heijer M. Age- and gender-specific reference values of estimated GFR in Caucasians: The Nijmegen Biomedical Study. Kidney Int. 2007;72(5):632–7.
- 32. Xia J, Wishart DS. MetPA: a web-based metabolomics tool for pathway analysis and visualization. Bioinformatics. 2010;26(18):2342–4. pmid:20628077
- 33. Wen D, Zheng Z, Surapaneni A, Yu B, Zhou L, Zhou W, et al. Metabolite profiling of CKD progression in the chronic renal insufficiency cohort study. JCI Insight. 2022;7(20):e161696. pmid:36048534
- 34. Quinkler M, Zehnder D, Lepenies J, Petrelli MD, Moore JS, Hughes SV, et al. Expression of renal 11beta-hydroxysteroid dehydrogenase type 2 is decreased in patients with impaired renal function. Eur J Endocrinol. 2005;153(2):291–9. pmid:16061836
- 35. Gant CM, Minovic I, Binnenmars H, de Vries L, Kema I, van Beek A, et al. Lower Renal Function Is Associated With Derangement of 11-β Hydroxysteroid Dehydrogenase in Type 2 Diabetes. J Endocr Soc. 2018;2(7):609–20. pmid:29942925
- 36. Uslar T, Newman AJ, Tapia-Castillo A, Carvajal CA, Fardella CE, Allende F, et al. Progressive 11β-hydroxysteroid dehydrogenase type 2 insufficiency as kidney function declines. J Clin Endocrinol Metab. 2024.
- 37. Lind L, Fall T, Ärnlöv J, Elmståhl S, Sundström J. Large-Scale Metabolomics and the Incidence of Cardiovascular Disease. J Am Heart Assoc. 2023;12(2):e026885. pmid:36645074
- 38. Pietzner M, Stewart ID, Raffler J, Khaw K-T, Michelotti GA, Kastenmüller G, et al. Plasma metabolites to profile pathways in noncommunicable disease multimorbidity. Nat Med. 2021;27(3):471–9. pmid:33707775
- 39. Kang J, Guo X, Peng H, Deng Y, Lai J, Tang L. Metabolic implications of amino acid metabolites in chronic kidney disease progression: a metabolomics analysis using OPLS-DA and MBRole2.0 database. Int Urol Nephrol. 2024;56(3):1173–84.
- 40. Li Z, Liu X, Wang J, Gao J, Guo S, Gao K, et al. Analysis of urinary metabolomic profiling for unstable angina pectoris disease based on nuclear magnetic resonance spectroscopy. Mol Biosyst. 2015;11(12):3387–96. pmid:26471262
- 41. Mehta A, Liu C, Nayak A, Tahhan AS, Ko Y-A, Dhindsa DS, et al. Untargeted high-resolution plasma metabolomic profiling predicts outcomes in patients with coronary artery disease. PLoS One. 2020;15(8):e0237579. pmid:32810196
- 42. Song L, Zhang Z, Qiu Z, Jiang T. Serum Metabonomic Study of Patients With Acute Coronary Syndrome Using Ultra-Performance Liquid Chromatography Orbitrap Mass Spectrometer. Front Cardiovasc Med. 2021;8:637621. pmid:33718457
- 43. Clase CM, Ki V, Holden RM. Water-soluble vitamins in people with low glomerular filtration rate or on dialysis: a review. Semin Dial. 2013;26(5):546–67. pmid:23859229
- 44. Dekkers KF, Sayols-Baixeras S, Baldanzi G, Nowak C, Hammar U, Nguyen D, et al. An online atlas of human plasma metabolite signatures of gut microbiome composition. Nat Commun. 2022;13(1):5370. pmid:36151114