Metabolomic Signatures of Inflammation in Chronic Kidney Disease Authors: Teresa K. Chen, Aditya L. Surapaneni, Michelle M. Estrella, Lawrence J. Appel, Wassim Obeid, Chirag R. Parikh, Mary R. Rooney, Bing Yu, Eugene P. Rhee, Insa M. Schmidt, Sushrut S. Waikar, Morgan E. Grams Published: 2026 ABSTRACT Rationale & ObjectiveInflammation is associated with adverse kidney, cardiovascular, and mortality outcomes. Investigation of the metabolic milieu as it relates to inflammation may provide important insights into these disease processes.Study DesignProspective cohort.Setting & ParticipantsAfrican American Study of Kidney Disease and Hypertension (AASK), Atherosclerosis Risk in Communities (ARIC) study, and Boston Kidney Biopsy Cohort (BKBC) participants with available metabolomics and inflammatory protein data.PredictorsBaseline blood levels of 718 metabolites.OutcomesBaseline and longitudinal changes in blood levels of tumor necrosis factor receptors 1 and 2 (TNFR1, TNFR2), tumor necrosis factor-alpha (TNF-α), interferon-gamma (IFN-γ), interleukins 6, 8, and 10 (IL-6, IL-8, IL-10), uromodulin (UMOD), and epidermal growth factor (EGF).Analytical ApproachMultivariable linear regression and linear mixed-effects models.ResultsAmong 491 AASK participants (mean age 54 years; 37% women; mean glomerular filtration rate, 45 mL/min/1.73 m2), 367 cross-sectional associations between metabolites and inflammatory proteins were significant after correction for multiple comparisons. The direction of association was mostly positive for TNFR1 (97%), TNFR2 (97%), IL-8 (77%), and IL-10 (100%); negative for UMOD (80%) and EGF (97%); and variable for TNF-⍺, IFN-γ, and IL-6. Pathways were distinct for several inflammatory proteins (eg, tryptophan metabolism for TNFR2). Forty-five associations between metabolites and longitudinal change in inflammatory proteins were identified. Notable metabolites included tigylcarnitine and N2,N5-diacetylornithine, which were associated with 2-year increases in TNFR1 and/or TNFR2, and 1,5-anhydroglucitol, where lower levels were associated with decreases in UMOD. In ARIC (n = 3,773) and BKBC (n = 413), replication of cross-sectional associations was excellent for TNFR1 (ARIC 83%; BKBC 85%) and TNFR2 (ARIC 64%; BKBC 79%) but poor for IL-8 (ARIC 3%; BKBC 3%).LimitationsMetabolite data limited to baseline visit; potential for residual confounding.ConclusionsUsing an untargeted approach, multiple metabolites were cross-sectionally and longitudinally associated with inflammatory proteins in persons with chronic kidney disease. Inflammation is associated with risk of numerous adverse outcomes, including chronic kidney disease (CKD) progression, cardiovascular disease, and mortality.1, 2, 3, 4, 5, 6, 7, 8 We previously reported in the African American Study of Kidney Disease and Hypertension (AASK) that several serum biomarkers related to inflammation (soluble tumor necrosis factor receptors 1 and 2 [TNFR1, TNFR2], tumor necrosis factor-alpha [TNF-α], interleukins 8 and 10 [IL-8, IL-10], and uromodulin [UMOD]) were associated with risks of kidney failure with replacement therapy (KFRT) or mortality.9, 10, 11, 12, 13 Additionally, in the Atherosclerosis Risk in Communities (ARIC) study, we found positive associations of TNFR1 and TNFR2 with risk of estimated glomerular filtration rate (eGFR) decline or KFRT and of UMOD polymorphisms with risk of incident CKD.14,15 In the Boston Kidney Biopsy Cohort (BKBC), TNFR1 and TNFR2 were associated with tubulointerstitial and glomerular lesions as well as increased risk of kidney disease progression.16 Investigation of the metabolomic milieu associated with inflammation could provide important insights on cardiovascular and kidney disease processes. The overall objective of this study was to use an untargeted metabolomics approach to identify serum metabolites associated with 9 proteins related to inflammation (TNFR1, TNFR2, TNF-α, interferon-gamma [IFN-γ], interleukin-6 [IL-6], IL-8, IL-10, UMOD, and epidermal growth factor [EGF]) at baseline and over time among AASK participants. For baseline metabolite–inflammatory protein associations that were identified, we explored enriched metabolic pathways and replicated findings in ARIC and BKBC. Methods Study Populations AASK was a 3 × 2 factorial trial that randomized 1,094 self-identified Black adults with hypertension-attributed CKD (glomerular filtration rate [GFR] 20-65 mL/min/1.73 m2) to treatment with: (1) metoprolol, ramipril, or amlodipine; and (2) intensive (mean arterial pressure 102-107 mm Hg) or standard (≤92 mm Hg) blood pressure (BP) control.17 Enrollment occurred from February 1995 to September 1998. At the trial’s end, participants without KFRT were invited to continue with the cohort phase, during which they received ramipril and a common BP goal (<140/90 mm Hg and then <130/80 mm Hg in 2004).18 Additional details regarding the AASK design have previously been published.17,18 Our study population comprised 491 AASK participants with available metabolomics and inflammatory protein data at trial baseline. ARIC is an ongoing community-based cohort of 15,792 adults who were recruited from Forsyth County, North Carolina; Jackson, Mississippi; Minneapolis, Minnesota; and Washington County, Maryland.14,19 Enrollment for visit 1 occurred between 1987 and 1989; visit 5 took place from 2011 to 2013. We included 3,773 visit 5 participants without KFRT who had available metabolite, inflammatory protein, and covariate data. BKBC was a prospective cohort of adults who received a clinically indicated native kidney biopsy from September 2006 to July 2019 at a tertiary care hospital in Boston, Massachusetts (Brigham and Women’s Hospital, Massachusetts General Hospital, Beth Israel Deaconess Medical Center).16,20 Blood and urine samples were collected on the date of biopsy. We included 413 participants with available metabolite, inflammatory protein, and covariate data. Further details on the derivation of each study population are presented Fig S1. For each cohort, informed consent was obtained from participants, and institutional review boards at each participating center approved study protocols.17, 18, 19, 20 Metabolite Profiling Baseline serum samples from AASK, visit 5 serum samples from ARIC, and baseline plasma samples from BKBC underwent processing at Metabolon, Inc in 2017, 2019, and 2023, respectively.21,22 Samples were analyzed using a reverse-phase ultraperformance liquid chromatography tandem mass-spectrometry (RP/UPLC-MS/MS) with a positive ion mode electrospray ionization (ESI), a RP/UPLC-MS/MS with a negative ion mode ESI, and a hydrophilic interaction ultra-performance liquid chromatography with negative ion mode ESI. Identified metabolites were then matched to an in-house library based on their retention time/index, mass-to-charge ratio, and chromatographic data. For quality control, we excluded metabolites with >80% missingness.21, 22, 23 For all 3 cohorts, each metabolite was rescaled to a median of 1, then log-transformed to reduce skewness. Within each study, samples were removed if the first principal component was >5 standard deviations from the mean. Metabolites with a variance ≤0.01 were excluded, and the remaining metabolites were winsorized at 5 standard deviations. In AASK, 652 known nondrug metabolites were identified, and missing values were imputed with the minimum value. For known drug metabolites, we excluded those with ≥25% missingness (n = 94) and did not impute missing values in the remaining 66. In total, we evaluated 718 metabolites. In ARIC and BKBC, we considered only those metabolites identified in our primary analyses in AASK. Outcomes Outcomes of interest were baseline inflammatory protein levels and their relative annual changes. Inflammatory proteins in AASK were quantified using the Meso Scale Discovery platform (Meso Scale Diagnostics) from serum samples collected at the baseline, 12-, or 24-month visits of the trial.9, 10, 11, 12 These immunoassay-based measurements were performed in 2019-2020 (for TNFR1, TNFR2, TNF-⍺, IFN-γ, IL-6, IL-8, IL-10) and 2021 (for UMOD and EGF), with interassay coefficients of variation <7.5%. In ARIC, proteins were measured in 2018 using the SomaScan V4 assay (SomaLogic) from visit 5 plasma samples.14,24 In BKBC, proteins were measured in 2023 using the SomaScan V4.1 (SomaLogic) platform from plasma samples collected on the day of kidney biopsy.23 These measurements were based on a Slow Off-rate Modified Aptamer-based capture array, and relative concentrations were quantified as relative fluorescence units. Further details regarding these measurements have previously been reported.14,23 For replication, we focused only on those proteins previously shown to have strong correlations (ie, correlation coefficient >0.6) between immunoassay-based and aptamer-based measurements (ie, TNFR1, TNFR2, and IL-8).13,25 Covariates Covariates were assessed at the baseline visit of AASK and visit 5 of ARIC and included age, sex, self-reported race, smoking status, history of heart disease, history of diabetes, randomized treatment groups (AASK), body mass index, systolic BP, GFR (AASK) or eGFR based on creatinine and cystatin C (eGFRcr-cys; ARIC), and urinary protein-creatinine ratio (UPCR; AASK) or urinary albumin-creatinine ratio (UACR; ARIC). The average of the second and third BP measurements of three was used.14,17,18 Covariates for BKBC were collected on the date of biopsy and included age, sex, self-reported race, history of diabetes, eGFR based on creatinine (eGFRcr), and UACR. In AASK, GFR was measured by iodine I-125 iothalamate clearance.17,18 Urinary protein and creatinine were measured at a central laboratory from 24-hour urine collections. In ARIC, GFR was estimated using the 2021 Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) eGFRcr-cys equation.26 Serum creatinine was measured by the Roche enzymatic method (Roche Diagnostics); serum cystatin C was measured using the Gentian immunoassay (Gentian AS). Urinary albumin was measured using an immunoturbidometric method on the ProSpec nephelometric analyzer (Dade Behring GMBH); urinary creatinine was measured using the Roche enzymatic method (Roche-Hitachi Modular P chemistry analyzer with Roche Creatininase Plus Assay). In BKBC, serum creatinine on the date of kidney biopsy and the most proximal UACR up to 3 months prior to the kidney biopsy were abstracted from the electronic health record.23,27 When not available, these were then measured from collected blood or urine samples from date of kidney biopsy. GFR was estimated using the 2021 CKD-EPI eGFRcr equation.26 Statistical Analyses Baseline characteristics were summarized as mean ± standard deviation, number (percentage), or median (quartile 1-quartile 3). Correlations of inflammatory proteins with each other and with GFR and UPCR were assessed by Spearman’s rank correlation coefficients. Both serum metabolites and inflammatory proteins were log2-transformed to achieve more normal distributions. Using multivariable linear regression, we evaluated cross-sectional associations of serum metabolites with serum inflammatory proteins (dependent variable) at baseline in AASK, adjusting for age, sex, randomized treatment groups, history of heart disease, smoking status, systolic BP, body mass index, GFR, and log2-transformed UPCR. We used Bonferroni correction to account for multiple testing, with a P value <0.05/(718 metabolites × 9 inflammatory proteins), or <7.74 × 10−6, considered statistically significant. For metabolites found to be significantly associated with TNFR1, TNFR2, or IL-8, we performed replication analyses in ARIC and BKBC. Similar linear regression models were used with the exception of: (1) adjusting for eGFRcr-cys instead of GFR, log2-transformed UACR instead of UPCR, combinations of race and study center instead of randomized treatment groups, and additionally adjusting for history of diabetes in ARIC; and (2) adjusting for eGFRcr instead of GFR, log2-transformed harmonized UACR instead of UPCR, and not being able to adjust for history of heart disease, smoking status, systolic BP, and body mass index in BKBC. For these analyses, Bonferroni-corrected P values of 3.27 × 10−4 (<0.05/153 significant associations in AASK available for comparison in ARIC) and 3.31 × 10−4 (<0.05/151 significant associations in AASK available for comparison in BKBC) were considered statistically significant. To assess whether inconsistent replication efforts might be due to differences in the platform used to measure proteins (ie, immunoassay vs aptamer-based) or to differences in cohorts, we performed a sensitivity analysis that evaluated associations between metabolites and TNFR1, TNFR2, and IL-8 measured using the SomaScan V4.1 platform in AASK.13 Next, we used MetaboAnalyst (https://www.metaboanalyst.ca) to identify overrepresented pathways associated with each inflammatory protein. The Pathway Analysis module, which is designed specifically for metabolomic studies, utilizes Kyoto Encyclopedia of Genes and Genomes (KEGG) metabolic pathways as its backend knowledgebase. We inputted known nondrug metabolites with available KEGG identifiers that had been associated with each inflammatory protein in AASK (defined by a Hochberg P value <0.05 for these analyses). We then used the hypergeometric test as the enrichment method, relative-betweenness centrality as the topology measure, and Homo sapiens (KEGG) as the pathway library, comparing against our reference metabolome, which included all known nondrug metabolites with available KEGG identifiers (n = 309, 17 of which were not matched to KEGG compound identifiers in MetaboAnalyst). Given that these analyses were exploratory, pathways with a raw P value <0.05 were considered significant. In longitudinal analyses, we evaluated associations of baseline serum metabolites with annual change in inflammatory proteins from baseline to 24 months in AASK. These analyses were agnostic to the results of our cross-sectional analyses. Linear mixed-effects models included follow-up time, interaction terms between each metabolite and follow-up time, and the aforementioned baseline covariates using the protein level at each visit as the dependent variable. The same Bonferroni corrected P value threshold of <7.74 × 10−6 was used. Data were analyzed using Stata 17.0 SE, and the circos plot was constructed using the circlize package in R. Results Baseline Characteristics Among AASK participants, the mean age was 54 ± 11 years, 100% self-identified as Black, 37% were women, 0% had diabetes, mean GFR was 45 ± 13 mL/min/1.73 m2, and median UPCR was 91 (30-402) mg/g (Table 1). Characteristics were similar for participants included versus those excluded from the study population (Table S1).Table 1Baseline Characteristics of AASK Participants Included in the Present StudyCharacteristicAASK (n = 491)Age, y54.1 ± 10.6Black491 (100%)Female184 (37%)History of heart disease247 (50%)History of diabetes0 (0%)Current smoking129 (26%)Former smoking148 (30%)Body mass index, kg/m230.9 ± 6.6Systolic blood pressure, mm Hg151 ± 25Glomerular filtration rate, mL/min/1.73 m244.8 ± 12.7Glomerular filtration rate categories, mL/min/1.73 m2 >900 (0%) 60-9069 (14%) 45-59193 (39%) 30-44147 (30%) <3082 (17%)Urinary protein-creatinine ratio, mg/g90.8 (30.4-401.6)Randomized treatment groups Ramipril; MAP ≤92 mm Hg91 (19%) Ramipril; MAP 102-107 mm Hg99 (20%) Metoprolol; MAP ≤92 mm Hg105 (21%) Metoprolol; MAP 102-107 mm Hg99 (20%) Amlodipine; MAP ≤92 mm Hg48 (10%) Amlodipine; MAP 102-107 mm Hg49 (10%)Note: Data presented as mean ± standard deviation, number (percentage), or median (quartile 1-quartile 3).Abbreviations: AASK, African American Study of Kidney Disease and Hypertension; MAP, mean arterial pressure. Table 1Baseline Characteristics of AASK Participants Included in the Present StudyCharacteristicAASK (n = 491)Age, y54.1 ± 10.6Black491 (100%)Female184 (37%)History of heart disease247 (50%)History of diabetes0 (0%)Current smoking129 (26%)Former smoking148 (30%)Body mass index, kg/m230.9 ± 6.6Systolic blood pressure, mm Hg151 ± 25Glomerular filtration rate, mL/min/1.73 m244.8 ± 12.7Glomerular filtration rate categories, mL/min/1.73 m2 >900 (0%) 60-9069 (14%) 45-59193 (39%) 30-44147 (30%) <3082 (17%)Urinary protein-creatinine ratio, mg/g90.8 (30.4-401.6)Randomized treatment groups Ramipril; MAP ≤92 mm Hg91 (19%) Ramipril; MAP 102-107 mm Hg99 (20%) Metoprolol; MAP ≤92 mm Hg105 (21%) Metoprolol; MAP 102-107 mm Hg99 (20%) Amlodipine; MAP ≤92 mm Hg48 (10%) Amlodipine; MAP 102-107 mm Hg49 (10%)Note: Data presented as mean ± standard deviation, number (percentage), or median (quartile 1-quartile 3).Abbreviations: AASK, African American Study of Kidney Disease and Hypertension; MAP, mean arterial pressure. Baseline Characteristics of AASK Participants Included in the Present Study Note: Data presented as mean ± standard deviation, number (percentage), or median (quartile 1-quartile 3). Abbreviations: AASK, African American Study of Kidney Disease and Hypertension; MAP, mean arterial pressure. Correlations of Inflammatory Proteins At baseline, TNFR1 and TNFR2 had strongly positive correlations with each other (rs = 0.89), moderately positive correlations with TNF-⍺ (rs =0.44 and 0.47, respectively), and moderately negative correlations with UMOD (rs = −0.62 and −0.53, respectively; Fig 1). GFR was negatively correlated with TNFR1 (rs = −0.76), TNFR2 (rs = −0.68), and TNF-α (rs = −0.35) and positively correlated with UMOD (rs = 0.57). The reverse was observed between log(UPCR) and these inflammatory proteins.Figure 1Spearman correlations and scatterplots between inflammatory proteins and kidney measures at baseline in the African American Study of Kidney Disease and Hypertension. Inflammatory proteins were measured by targeted immunoassays. The following inflammatory proteins had significant correlations with glomerular filtration rate: TNFR1 (P < 0.001), TNFR2 (P < 0.001), TNF-α (P < 0.001), IL-10 (P = 0.03), UMOD (P < 0.001), and EGF (P < 0.001). EGF, epidermal growth factor; GFR, glomerular filtration rate; IFN, interferon; IL, interleukin; TNF, tumor necrosis factor; TNFR, tumor necrosis factor receptor; UMOD, uromodulin; UPCR, urinary protein-creatinine ratio. Spearman correlations and scatterplots between inflammatory proteins and kidney measures at baseline in the African American Study of Kidney Disease and Hypertension. Inflammatory proteins were measured by targeted immunoassays. The following inflammatory proteins had significant correlations with glomerular filtration rate: TNFR1 (P < 0.001), TNFR2 (P < 0.001), TNF-α (P < 0.001), IL-10 (P = 0.03), UMOD (P < 0.001), and EGF (P < 0.001). EGF, epidermal growth factor; GFR, glomerular filtration rate; IFN, interferon; IL, interleukin; TNF, tumor necrosis factor; TNFR, tumor necrosis factor receptor; UMOD, uromodulin; UPCR, urinary protein-creatinine ratio. Cross-sectional Associations of Metabolites With Inflammatory Proteins Among 718 metabolites examined in AASK, the number of metabolites associated with each targeted serum biomarker at baseline ranged from 2 for IL-10 to 103 for EGF (Tables S2-S10). Associations were mostly in the positive direction for TNFR1 (97%), TNFR2 (97%), IL-8 (77%), and IL-10 (100%), mostly negative for UMOD (80%) and EGF (97%), and roughly equal for the remaining proteins. Several metabolites were associated with multiple targeted biomarkers, with substantial overlap for TNFR1 and TNFR2 and for UMOD and EGF (Fig 2). Metabolites in the peptide super pathway tended to have negative associations (ie, higher levels of metabolites with lower levels of inflammatory proteins), whereas those in the lipid super pathways tended to have positive associations (ie, higher levels of metabolites with higher levels of inflammatory proteins). Notable xenobiotics included O-sulfo-l-tyrosine and erythritol (a sugar substitute commonly used in artificial sweeteners),28 which were each associated with higher levels of TNFR1 and TNFR2 and lower levels of UMOD and EGF, as well as the carotene diols, which were generally associated with lower levels of IL-6 and IL-8.Figure 2Circos plot of cross-sectional associations between metabolites and inflammatory proteins at baseline in the African American Study of Kidney Disease and Hypertension. Red denotes positive associations, whereas blue denotes negative associations. Darkest shades of each denote proteins with P values below the Bonferroni threshold of 7.74 × 10−6. White denotes P > 0.05. The remaining colors are graded on the log-scale of the P value. Circos plot of cross-sectional associations between metabolites and inflammatory proteins at baseline in the African American Study of Kidney Disease and Hypertension. Red denotes positive associations, whereas blue denotes negative associations. Darkest shades of each denote proteins with P values below the Bonferroni threshold of 7.74 × 10−6. White denotes P > 0.05. The remaining colors are graded on the log-scale of the P value. Metabolic Pathways Associated With Inflammatory Proteins Overrepresented metabolic pathways were unique for most inflammatory proteins (Table 2). Distinct pathways were noted for TNFR2 (tryptophan metabolism), IFN-γ (pyrimidine metabolism, nitrogen metabolism, glycoxylate and dicarboxylate metabolism, arginine biosynthesis), IL-6 (glycerophospholipid metabolism, glycosylphosphatidylinositol-anchor biosynthesis, retinol metabolism), and IL-8 (pyruvate metabolism, cysteine and methionine metabolism, citrate cycle, sphingolipid metabolism, taurine and hypotaurine metabolism). As shown in Fig S3, several metabolites involved in the kynurenine pathway of tryptophan metabolism were noted to have associations with TNFR1, TNFR2, TNF-⍺, or IFN-γ.Table 2Overrepresented Metabolic Pathways Associated With Each Inflammatory ProteinPathwayTNFR1TNFR2TNF-⍺IFN-γIL-6IL-8EGFInositol phosphate metabolismXXXAscorbate and aldarate metabolismXXTryptophan metabolismXβ-alanine metabolismXXPyrimidine metabolismXNitrogen metabolismXGlycoxylate and dicarboxylate metabolismXArginine biosynthesisXAlanine, aspartate, and glutamate metabolismXXGlycerophospholipid metabolismXGlycosylphosphatidylinositol-anchor biosynthesisXRetinol metabolismXPyruvate metabolismXCysteine and methionine metabolismXCitrate cycle (TCA cycle)XSphingolipid metabolismXTaurine and hypotaurine metabolismXNote: Significant pathways defined by a raw P < 0.05. There were no significant pathways enriched for UMOD and IL-10.Abbreviations: EGF, epidermal growth factor; IFN, interferon; IL, interleukin; TNF, tumor necrosis factor; TNFR, tumor necrosis factor receptor; UMOD, uromodulin. Table 2Overrepresented Metabolic Pathways Associated With Each Inflammatory ProteinPathwayTNFR1TNFR2TNF-⍺IFN-γIL-6IL-8EGFInositol phosphate metabolismXXXAscorbate and aldarate metabolismXXTryptophan metabolismXβ-alanine metabolismXXPyrimidine metabolismXNitrogen metabolismXGlycoxylate and dicarboxylate metabolismXArginine biosynthesisXAlanine, aspartate, and glutamate metabolismXXGlycerophospholipid metabolismXGlycosylphosphatidylinositol-anchor biosynthesisXRetinol metabolismXPyruvate metabolismXCysteine and methionine metabolismXCitrate cycle (TCA cycle)XSphingolipid metabolismXTaurine and hypotaurine metabolismXNote: Significant pathways defined by a raw P < 0.05. There were no significant pathways enriched for UMOD and IL-10.Abbreviations: EGF, epidermal growth factor; IFN, interferon; IL, interleukin; TNF, tumor necrosis factor; TNFR, tumor necrosis factor receptor; UMOD, uromodulin. Overrepresented Metabolic Pathways Associated With Each Inflammatory Protein Note: Significant pathways defined by a raw P < 0.05. There were no significant pathways enriched for UMOD and IL-10. Abbreviations: EGF, epidermal growth factor; IFN, interferon; IL, interleukin; TNF, tumor necrosis factor; TNFR, tumor necrosis factor receptor; UMOD, uromodulin. Associations of Metabolites With Longitudinal Changes in Inflammatory Proteins The mean annual change in inflammatory proteins ranged from −8.2% to 8.6%, with TNFR1, TNFR2, TNF-α, IL-6, and IL-8 having increasing levels and IFN-γ, IL-10, UMOD, and EGF having decreasing levels from baseline to 24 months (Table 3). Forty-five associations representing 41 distinct metabolites were statistically significant. Twenty distinct metabolites were associated with EGF change: taurine had the strongest negative association whereas 2-hydroxydecanoate was the only metabolite with a positive association. Most other metabolites associated with EGF belonged to lipid pathways of sphingolipid metabolism, eicosanoids, or fatty acid metabolism. There were 15 metabolites associated with TNF-α change, 6 with TNFR1 change (eg, tiglylcarnitine and N2,N5-diacetylornithine), 2 with IL-8 change, and 1 each with TNFR2 (ie, tiglylcarnitine) and UMOD (ie, 1,5-anhydroglucitol) change. Baseline and longitudinal associations of metabolites with TNFR1 and TNFR2 were in the same direction, whereas for TNF-α, IL-8, UMOD, and EGF, they were in the opposite direction.Table 3Associations of Baseline Metabolites With 2-Year Change in Inflammatory Proteins in AASKMetaboliteSuper PathwaySub PathwaySD of MetabolitePercentage Annual Change (95% CI)aP valueTNFR1Mean (SD) % change per year: 5.06% (9.31%)Tiglylcarnitine (C5:1-DC)Amino acidLeucine, isoleucine, and valine metabolism0.623.6% (2.2%, 4.9%)1.39 × 10−7N-acetylglutamineAmino acidGlutamate metabolism0.523.3% (1.9%, 4.6%)1.18 × 10−6N-acetylisoleucineAmino acidLeucine, isoleucine, and valine metabolism0.383.3% (1.9%, 4.6%)1.18 × 10−6N2,N5-diacetylornithineAmino acidUrea cycle; arginine and proline metabolism0.763.2% (1.8%, 4.5%)3.81 × 10−6N-acetylleucineAmino acidLeucine, isoleucine, and valine metabolism0.343.1% (1.8%, 4.4%)4.63 × 10−6N-acetylcitrullineAmino acidUrea cycle; arginine and proline metabolism0.863.0% (1.7%, 4.3%)7.66 × 10−6TNFR2Mean (SD) % change per year: 3.39% (6.76%)Tiglylcarnitine (C5:1-DC)Amino acidLeucine, isoleucine, and valine metabolism0.623.3% (2.1%, 4.6%)1.76 × 10−7TNF-⍺Mean (SD) % change per year: 3.54% (10.25%)5-methylthioadenosine (MTA)Amino acidPolyamine metabolism1.00−10.6% (−13.1%, −8.1%)4.49 × 10−15DihydroorotateNucleotidePyrimidine metabolism, orotate-containing1.57−10.2% (−12.7%, −7.6%)8.26 × 10−14CholineLipidPhospholipid metabolism0.2411.0% (7.9%, 14.2%)3.62 × 10−13PhenylalanylglycinePeptideDipeptide0.82−9.6% (−12.1%, −7.0%)4.52 × 10−12DSGEGDFXAEGGGVR∗PeptideFibrinogen cleavage peptide1.28−9.3% (−11.9%, −6.7%)9.64 × 10−12PhenylalanineAmino acidPhenylalanine metabolism0.179.0% (5.9%, 12.1%)4.08 × 10−91-linoleoyl-GPA (18:2)∗aLipidLysophospholipid0.519.1% (5.9%, 12.4%)1.07 × 10−85-oxoprolineAmino acidGlutathione metabolism0.308.1% (5.1%, 11.1%)6.02 × 10−8GlycylvalinePeptideDipeptide1.01−7.1% (−9.7%, −4.4%)6.60 × 10−7SphinganineLipidSphingolipid metabolism0.747.3% (4.3%, 10.4%)1.29 × 10−6Methionine sulfoxideAmino acidMethionine, cysteine, SAM, and taurine metabolism0.407.2% (4.1%, 10.3%)2.43 × 10−6GuanosineNucleotidePurine metabolism, guanine-containing1.65−6.6% (−9.2%, −3.9%)2.78 × 10−6SphingosineLipidSphingolipid metabolism0.787.0% (4.0%, 10.0%)3.50 × 10−6Linoleoyl ethanolamideLipidEndocannabinoid0.677.1% (4.0%, 10.4%)6.02 × 10−6GlycineAmino acidGlycine, serine, and threonine metabolism0.206.8% (3.8%, 10.0%)7.26 × 10−6IL-8Mean (SD) % change per year: 8.17% (14.20%)5-KETELipidEicosanoid1.42−12.2% (−16.7%, −7.5%)1.03 × 10−61-(1-enyl-palmitoyl)-GPE (P-16:0)∗LipidLysoplasmalogen0.39−11.1% (−15.6%, −6.4%)7.28 × 10−6UMODMean (SD) % change per year: −8.24% (7.86%)1,5-anhydroglucitol (1,5-AG)CarbohydrateGlycolysis, gluconeogenesis, and pyruvate metabolism0.474.9% (3.1%, 6.7%)6.01 × 10−8EGFMean (SD) % change per year: −6.99% (7.08%)TaurineAmino acidMethionine, cysteine, SAM, and taurine metabolism0.29−22.7% (−26.4%, −18.7%)4.97× 10−24LactateCarbohydrateGlycolysis, gluconeogenesis, and pyruvate metabolism0.32−19.2% (−23.1%, −15.1%)3.14 × 10−17Sphinganine-1-phosphateLipidSphingolipid metabolism0.49−18.1% (−22.0%, −14.0%)1.35 × 10−15Sphingosine 1-phosphateLipidSphingolipid metabolism0.39−17.1% (−21.0%, −12.9%)3.30 × 10−14SphingosineLipidSphingolipid metabolism0.78−16.8% (−20.8%, −12.7%)1.25 × 10−1312-HHTrELipidEicosanoid1.82−16.1% (−20.0%, −11.9%)8.03 × 10−1312-HETELipidEicosanoid1.36−16.1% (−20.0%, −11.9%)1.01 × 10−12HypotaurineAmino acidMethionine, cysteine, SAM, and taurine metabolism0.57−15.3% (−19.3%, −11.0%)3.23 × 10−11Arachidonoylcarnitine (C20:4)LipidFatty acid metabolism (acyl carnitine)0.64−14.7% (−18.6%, −10.6%)3.63 × 10−111-arachidonoyl-GPA (20:4)LipidLysophospholipid0.60−15.5% (−19.6%, −11.1%)7.02 × 10−11SphinganineLipidSphingolipid metabolism0.74−13.3% (−17.4%, −9.0%)8.78 × 10−9β-citrylglutamateAmino acidGlutamate metabolism0.49−12.9% (−17.2%, −8.4%)6.01 × 10−8Linoleoylcarnitine (C18:2)∗LipidFatty acid metabolism(acyl carnitine)0.36−12.2% (−16.4%, −7.8%)1.78 × 10−7XanthosineNucleotidePurine metabolism, (hypo)xanthine/inosine-containing0.72−11.8% (−16.1%, −7.3%)6.85 × 10−7FumarateEnergyTCA cycle0.33−11.9% (−16.2%, −7.3%)9.92 × 10−7Dihomo-linolenoylcarnitine (20:3–n3 or 6)aLipidFatty acid metabolism(acyl carnitine)0.50−11.2% (−15.5%, −6.8%)1.98 × 10−6AspartateAmino acidAlanine and aspartate metabolism0.24−11.3% (−15.7%, −6.7%)3.70 × 10−65-KETELipidEicosanoid1.42−11.0% (−15.3%, −6.4%)4.76 × 10−62-hydroxydecanoateLipidFatty acid, monohydroxy0.5112.2% (6.7%, 17.9%)5.80 × 10−6MalateEnergyTCA cycle0.31−11.1% (−15.5%, −6.5%)5.96 × 10−6Note: Models adjusted for age, sex, history of heart disease, current smoking, former smoking, body mass index, systolic blood pressure, glomerular filtration rate, log-transformed urinary protein-creatinine ratio, and randomized treatment groups. Significant metabolites defined as having a Bonferroni-corrected P value <0.05/6462 or 7.74 × 10−6. There were no significant associations for IFN-γ mean (SD) change per year: −3.04% (5.35%), IL-6: 8.63% (13.09%), and IL-10: −1.06% (5.78%).Abbreviations: AASK, African American Study of Kidney Disease and Hypertension; CI, confidence interval; DC, dicarboxylic acid; EGF, epidermal growth factor; GPA, glycerophosphate; GPE, glycerophosphoethanolamine; HETE, hydroxyeicosatetraenoic acid; HHTrE, hydroxyheptadecatrienoic acid; KETE, keto-eicosatetraenoic acid; IFN, interferon; IL, interleukin; SAM, S-adenosylmethionine; SD, standard deviation; TCA, tricarboxylic acid; TNF, tumor necrosis factor; TNFR, tumor necrosis factor receptor; UMOD, uromodulin.aDenotes percentage annual change in inflammatory protein per 1 SD higher baseline metabolite level. Table 3Associations of Baseline Metabolites With 2-Year Change in Inflammatory Proteins in AASKMetaboliteSuper PathwaySub PathwaySD of MetabolitePercentage Annual Change (95% CI)aP valueTNFR1Mean (SD) % change per year: 5.06% (9.31%)Tiglylcarnitine (C5:1-DC)Amino acidLeucine, isoleucine, and valine metabolism0.623.6% (2.2%, 4.9%)1.39 × 10−7N-acetylglutamineAmino acidGlutamate metabolism0.523.3% (1.9%, 4.6%)1.18 × 10−6N-acetylisoleucineAmino acidLeucine, isoleucine, and valine metabolism0.383.3% (1.9%, 4.6%)1.18 × 10−6N2,N5-diacetylornithineAmino acidUrea cycle; arginine and proline metabolism0.763.2% (1.8%, 4.5%)3.81 × 10−6N-acetylleucineAmino acidLeucine, isoleucine, and valine metabolism0.343.1% (1.8%, 4.4%)4.63 × 10−6N-acetylcitrullineAmino acidUrea cycle; arginine and proline metabolism0.863.0% (1.7%, 4.3%)7.66 × 10−6TNFR2Mean (SD) % change per year: 3.39% (6.76%)Tiglylcarnitine (C5:1-DC)Amino acidLeucine, isoleucine, and valine metabolism0.623.3% (2.1%, 4.6%)1.76 × 10−7TNF-⍺Mean (SD) % change per year: 3.54% (10.25%)5-methylthioadenosine (MTA)Amino acidPolyamine metabolism1.00−10.6% (−13.1%, −8.1%)4.49 × 10−15DihydroorotateNucleotidePyrimidine metabolism, orotate-containing1.57−10.2% (−12.7%, −7.6%)8.26 × 10−14CholineLipidPhospholipid metabolism0.2411.0% (7.9%, 14.2%)3.62 × 10−13PhenylalanylglycinePeptideDipeptide0.82−9.6% (−12.1%, −7.0%)4.52 × 10−12DSGEGDFXAEGGGVR∗PeptideFibrinogen cleavage peptide1.28−9.3% (−11.9%, −6.7%)9.64 × 10−12PhenylalanineAmino acidPhenylalanine metabolism0.179.0% (5.9%, 12.1%)4.08 × 10−91-linoleoyl-GPA (18:2)∗aLipidLysophospholipid0.519.1% (5.9%, 12.4%)1.07 × 10−85-oxoprolineAmino acidGlutathione metabolism0.308.1% (5.1%, 11.1%)6.02 × 10−8GlycylvalinePeptideDipeptide1.01−7.1% (−9.7%, −4.4%)6.60 × 10−7SphinganineLipidSphingolipid metabolism0.747.3% (4.3%, 10.4%)1.29 × 10−6Methionine sulfoxideAmino acidMethionine, cysteine, SAM, and taurine metabolism0.407.2% (4.1%, 10.3%)2.43 × 10−6GuanosineNucleotidePurine metabolism, guanine-containing1.65−6.6% (−9.2%, −3.9%)2.78 × 10−6SphingosineLipidSphingolipid metabolism0.787.0% (4.0%, 10.0%)3.50 × 10−6Linoleoyl ethanolamideLipidEndocannabinoid0.677.1% (4.0%, 10.4%)6.02 × 10−6GlycineAmino acidGlycine, serine, and threonine metabolism0.206.8% (3.8%, 10.0%)7.26 × 10−6IL-8Mean (SD) % change per year: 8.17% (14.20%)5-KETELipidEicosanoid1.42−12.2% (−16.7%, −7.5%)1.03 × 10−61-(1-enyl-palmitoyl)-GPE (P-16:0)∗LipidLysoplasmalogen0.39−11.1% (−15.6%, −6.4%)7.28 × 10−6UMODMean (SD) % change per year: −8.24% (7.86%)1,5-anhydroglucitol (1,5-AG)CarbohydrateGlycolysis, gluconeogenesis, and pyruvate metabolism0.474.9% (3.1%, 6.7%)6.01 × 10−8EGFMean (SD) % change per year: −6.99% (7.08%)TaurineAmino acidMethionine, cysteine, SAM, and taurine metabolism0.29−22.7% (−26.4%, −18.7%)4.97× 10−24LactateCarbohydrateGlycolysis, gluconeogenesis, and pyruvate metabolism0.32−19.2% (−23.1%, −15.1%)3.14 × 10−17Sphinganine-1-phosphateLipidSphingolipid metabolism0.49−18.1% (−22.0%, −14.0%)1.35 × 10−15Sphingosine 1-phosphateLipidSphingolipid metabolism0.39−17.1% (−21.0%, −12.9%)3.30 × 10−14SphingosineLipidSphingolipid metabolism0.78−16.8% (−20.8%, −12.7%)1.25 × 10−1312-HHTrELipidEicosanoid1.82−16.1% (−20.0%, −11.9%)8.03 × 10−1312-HETELipidEicosanoid1.36−16.1% (−20.0%, −11.9%)1.01 × 10−12HypotaurineAmino acidMethionine, cysteine, SAM, and taurine metabolism0.57−15.3% (−19.3%, −11.0%)3.23 × 10−11Arachidonoylcarnitine (C20:4)LipidFatty acid metabolism (acyl carnitine)0.64−14.7% (−18.6%, −10.6%)3.63 × 10−111-arachidonoyl-GPA (20:4)LipidLysophospholipid0.60−15.5% (−19.6%, −11.1%)7.02 × 10−11SphinganineLipidSphingolipid metabolism0.74−13.3% (−17.4%, −9.0%)8.78 × 10−9β-citrylglutamateAmino acidGlutamate metabolism0.49−12.9% (−17.2%, −8.4%)6.01 × 10−8Linoleoylcarnitine (C18:2)∗LipidFatty acid metabolism(acyl carnitine)0.36−12.2% (−16.4%, −7.8%)1.78 × 10−7XanthosineNucleotidePurine metabolism, (hypo)xanthine/inosine-containing0.72−11.8% (−16.1%, −7.3%)6.85 × 10−7FumarateEnergyTCA cycle0.33−11.9% (−16.2%, −7.3%)9.92 × 10−7Dihomo-linolenoylcarnitine (20:3–n3 or 6)aLipidFatty acid metabolism(acyl carnitine)0.50−11.2% (−15.5%, −6.8%)1.98 × 10−6AspartateAmino acidAlanine and aspartate metabolism0.24−11.3% (−15.7%, −6.7%)3.70 × 10−65-KETELipidEicosanoid1.42−11.0% (−15.3%, −6.4%)4.76 × 10−62-hydroxydecanoateLipidFatty acid, monohydroxy0.5112.2% (6.7%, 17.9%)5.80 × 10−6MalateEnergyTCA cycle0.31−11.1% (−15.5%, −6.5%)5.96 × 10−6Note: Models adjusted for age, sex, history of heart disease, current smoking, former smoking, body mass index, systolic blood pressure, glomerular filtration rate, log-transformed urinary protein-creatinine ratio, and randomized treatment groups. Significant metabolites defined as having a Bonferroni-corrected P value <0.05/6462 or 7.74 × 10−6. There were no significant associations for IFN-γ mean (SD) change per year: −3.04% (5.35%), IL-6: 8.63% (13.09%), and IL-10: −1.06% (5.78%).Abbreviations: AASK, African American Study of Kidney Disease and Hypertension; CI, confidence interval; DC, dicarboxylic acid; EGF, epidermal growth factor; GPA, glycerophosphate; GPE, glycerophosphoethanolamine; HETE, hydroxyeicosatetraenoic acid; HHTrE, hydroxyheptadecatrienoic acid; KETE, keto-eicosatetraenoic acid; IFN, interferon; IL, interleukin; SAM, S-adenosylmethionine; SD, standard deviation; TCA, tricarboxylic acid; TNF, tumor necrosis factor; TNFR, tumor necrosis factor receptor; UMOD, uromodulin.aDenotes percentage annual change in inflammatory protein per 1 SD higher baseline metabolite level. Associations of Baseline Metabolites With 2-Year Change in Inflammatory Proteins in AASK Note: Models adjusted for age, sex, history of heart disease, current smoking, former smoking, body mass index, systolic blood pressure, glomerular filtration rate, log-transformed urinary protein-creatinine ratio, and randomized treatment groups. Significant metabolites defined as having a Bonferroni-corrected P value <0.05/6462 or 7.74 × 10−6. There were no significant associations for IFN-γ mean (SD) change per year: −3.04% (5.35%), IL-6: 8.63% (13.09%), and IL-10: −1.06% (5.78%). Abbreviations: AASK, African American Study of Kidney Disease and Hypertension; CI, confidence interval; DC, dicarboxylic acid; EGF, epidermal growth factor; GPA, glycerophosphate; GPE, glycerophosphoethanolamine; HETE, hydroxyeicosatetraenoic acid; HHTrE, hydroxyheptadecatrienoic acid; KETE, keto-eicosatetraenoic acid; IFN, interferon; IL, interleukin; SAM, S-adenosylmethionine; SD, standard deviation; TCA, tricarboxylic acid; TNF, tumor necrosis factor; TNFR, tumor necrosis factor receptor; UMOD, uromodulin. Denotes percentage annual change in inflammatory protein per 1 SD higher baseline metabolite level. Replication in ARIC and BKBC ARIC visit 5 participants in the replication set had a mean age of 76 years, 17% self-identified as Black, 54% were women; 33% had diabetes, mean eGFR was 69 ± 19 mL/min/1.73 m2, and median UACR was 11 (6-23) mg/g. Compared with those who were excluded, a greater proportion of the study population self-identified as White and male, reported former smoking, and had slightly lower mean systolic BP and median UACR (Table S1). The following number of metabolites replicated for each of the following inflammatory proteins: 49 (83%) for TNFR1, 21 (64%) for TNFR2, and 2 (3%) for IL-8 (Tables S2, S3, and S7). BKBC participants had a mean age of 54 years, 20% self-identified as Black and 7% as Asian, 29% had diabetes, mean eGFR was 50 ± 33 mL/min/1.73 m2, and median UACR was 1058 (230-3070) mg/g. The included study population was older, more likely to have diabetes, and had a lower baseline eGFR than their counterparts who were excluded from the study (Table S1). Replication was observed for 51 (85%), 26 (79%), and 2 (3%) of metabolites for TNFR1, TNFR2, and IL-8, respectively. Given that replication for IL-8 occurred for only 3% of metabolites in ARIC and BKBC, we performed sensitivity analyses that evaluated associations of metabolites with IL-8 measured using the SomaScan V4.1 platform in AASK.13 Replication was high at 97% (Table S7). Replication of associations for TNFR1 and TNFR2 when measured with SomaScan V4.1 were also high, at 78% and 94%, respectively (Tables S2 and S3). Correlations of these inflammatory proteins with kidney measures in each cohort are presented in Figure S2. Discussion In the context of CKD, we identified variations in the metabolic milieu associated with 9 serum inflammatory proteins important for cardiometabolic kidney health: TNFR1,4,9,11,29 TNFR2,4,9,11,29 TNF-α,2,9,30 IFN-γ,31,32 IL-6,30,33, 34, 35 IL-8,10,36 IL-10,10,37 UMOD,12,38, 39, 40 and EGF.41,42 Distinct metabolic pathways were enriched for several inflammatory proteins. For example, TNFR2 elevations were associated with derangements in tryptophan metabolism, and several dietary components were associated with TNFR1 and IL-6. We also observed numerous metabolites that preceded changes in inflammatory markers over time. Taken together, these findings suggest that specific metabolites may be useful in predicting future inflammation. Future mechanistic or patient stratification studies are needed to investigate whether they can be used to identify individuals who are more likely to respond to targeted anti-inflammatory therapies. In cross-sectional analyses, we observed positive associations of kynurenine with TNF-α, its receptors (TNFR1 and TNFR2), and IFN-γ. Additionally, we found that elevated TNFR2, which has been strongly associated with CKD progression,9,11,14 was reflected by alterations in the tryptophan pathway. Prior in vitro studies have demonstrated that TNF-α synergistically enhances IFN-γ-induced expression of indoleamine 2,3-dioxygenase, one of the rate-limiting enzymes in tryptophan catabolism via the kynurenine pathway.43, 44, 45, 46 Our results suggest that similar mechanisms may occur in vivo. Other studies have reported that lower levels of plasma tryptophan and higher kynurenine-to-tryptophan ratios are associated with higher risks of incident CKD,47 progression to KFRT,48 or faster eGFR decline.47,48 It may be that ratios, rather than absolute levels, of metabolites better capture inflammation and disease risk. Various xenobiotics were also associated with circulating inflammatory proteins. For example, erythritol was positively associated with TNFR1. Often used as an artificial sweetener, erythritol has been associated with heightened risk of major adverse cardiovascular events.28 In vitro, incubation of human platelet-rich plasma with erythritol enhanced platelet aggregation in a dose-dependent manner. In vivo, injection of erythritol into mice increased rates of clot formation and cessation of blood flow after carotid arterial injury compared with normal saline. In another example, carotene diols were negatively associated with IL-6. Serum carotenoid levels are thought to reflect diets rich in fruits and vegetables.49 Vast literature exists detailing the anti-inflammatory benefits of such dietary patterns,50, 51, 52, 53, 54 including some suggestion of protection against adverse clinical outcomes.49,55, 56, 57, 58, 59, 60 We evaluated some of our cross-sectional findings in 2 independent cohorts, showing excellent consistency for TNFR1 and TNFR2 but not for IL-8. Replication for TNFR1 and TNFR2 ranged from 64% to 85% across cohorts but, puzzlingly, only 3% for IL-8. To explore whether this low percentage of replication was due to differences in how IL-8 was measured (ie, immunoassay vs aptamer-based), we conducted sensitivity analyses in AASK using SomaScan-measured IL-8 as the outcome. We found that 97% of metabolites replicated, suggesting that the lack of replication in ARIC and BKBC for this protein may have been due to differences in study populations instead. Previous studies have suggested that IL-8 levels may be reflective of the quality of the biospecimen, with higher levels indicating worse quality.61 Alternatively, associations between metabolites and IL-8 may vary by specimen type (eg, serum vs plasma) or population (eg, hypertension-related kidney disease vs general population). Several metabolite alterations preceded changes in inflammatory proteins over time. Lower blood levels of 1,5-anhydroglucitol, which generally signify hyperglycemic excursions,62,63 have been associated with higher risks of adverse cardiovascular outcomes, microvascular events, and all-cause mortality among people with diabetes.63,64 In people without diabetes, plasma 1,5-anhydroglucitol levels can reflect type and amount of dietary carbohydrate intake.65 We expand the evidence base in nondiabetic CKD populations with the observation that lower levels of 1,5-anhydroglucitol were associated with greater declines in the protective UMOD protein. Higher levels of tiglylcarnitine and N2,N5-diacetylornithine were each associated with greater increases in TNFR1 or TNFR2. Tiglylcarnitine, a short-chain acylcarnitine, transports acyl groups from the cytoplasm into mitochondria where it undergoes β-oxidation to produce energy.66,67 Tiglylcarnitine may be a marker of mitochondrial health; when mitochondrial β-oxidation is impaired, short- and intermediate-chain acylcarnitines accumulate. Among individuals with type 2 diabetes (mean eGFR 77 mL/min/1.73 m2), higher levels of tigylcarnitine were associated with annual eGFR decline; the addition of tigylcarnitine, along with kynurenine, to a clinical model improved prediction of rapid eGFR decline.67 N2, N5-diacetylornithine, a byproduct of the urea cycle, was the only metabolite to have overlapping associations with an inflammatory protein (ie, TNFR1) in both cross-sectional and longitudinal analyses. N2, N5-diacetylornithine has been associated with increased risk of kidney failure in the ARIC study and higher left ventricular filling pressures in the Bogalusa Heart Study.21,68 Proposed mechanisms for the latter include endothelial dysfunction and microvascular disease.68 Thus, our data suggest that mitochondrial or endothelial dysfunction are signaled by TNFR1- or TNFR2-related inflammation. Our study has several strengths. First, we used an untargeted approach to comprehensively assess cross-sectional associations of 718 metabolites with 9 inflammatory proteins. Second, we adjusted for baseline kidney function to account for the potential for decreased glomerular clearance of metabolites and inflammatory proteins. Third, several metabolite–inflammatory protein associations were replicated in 2 independent cohorts of adults, suggesting applicability of findings to the general population (ARIC) and other forms of kidney disease (BKBC). Nevertheless, there are important limitations that warrant mention. We were unable to evaluate cross-sectional associations in ARIC and BKBC for TNF-α, IFN-γ, IL-6, IL-10, UMOD, and EGF because of the inaccuracy of aptamer-based measurements for these proteins. Metabolite data in AASK were available only at baseline. Despite accounting for GFR and eGFR in our models, there remains the potential for residual confounding. The sample sizes for AASK and BKBC were relatively small, which may have limited the ability to detect weaker associations. Some differences in baseline characteristics were also noted between participants who were included versus excluded for ARIC and BKBC, which may have introduced a selection bias. Finally, with the exception of N2,N5-diacetylornithine and TNFR1, there was no overlap between cross-sectional and longitudinal analyses, thus suggesting that many of the cross-sectional associations were not causal. Perhaps, the observed associations reflect shared, latent exposures or that the inflammatory biomarker causes changes in the metabolite instead. In conclusion, using an untargeted metabolomics approach, we identified numerous metabolites associated with inflammatory proteins both cross-sectionally and over time. Future studies are needed to determine whether metabolic profiling can be applied to clinical practice to individualize treatment decisions and ultimately improve clinical care.