Abstract
KEY POINTS: We provide an atlas of cross-sectional and longitudinal serum and urine metabolite associations with eGFR and urine albumin-creatinine ratio in an older community-based cohort. Metabolic profiling in serum and urine provides distinct and complementary insights into disease. BACKGROUND: Metabolites represent a read-out of cellular processes underlying states of health and disease. METHODS: We evaluated cross-sectional and longitudinal associations between 1255 serum and 1398 urine known and unknown (denoted with “X” in name) metabolites (Metabolon HD4, 721 detected in both biofluids) and kidney function in 1612 participants of the Atherosclerosis Risk in Communities study. All analyses were adjusted for clinical and demographic covariates, including for baseline eGFR and urine albumin-creatinine ratio (UACR) in longitudinal analyses. RESULTS: At visit 5 of the Atherosclerosis Risk in Communities study, the mean age of participants was 76 years (SD 6); 56% were women, mean eGFR was 62 ml/min per 1.73 m(2) (SD 20), and median UACR level was 13 mg/g (interquartile range, 25). In cross-sectional analysis, 675 serum and 542 urine metabolites were associated with eGFR (Bonferroni-corrected P < 4.0E-5 for serum analyses and P < 3.6E-5 for urine analyses), including 248 metabolites shared across biofluids. Fewer metabolites (75 serum and 91 urine metabolites, including seven metabolites shared across biofluids) were cross-sectionally associated with albuminuria. Guanidinosuccinate; N2,N2-dimethylguanosine; hydroxy-N6,N6,N6-trimethyllysine; X-13844; and X-25422 were significantly associated with both eGFR and albuminuria. Over a mean follow-up of 6.6 years, serum mannose (hazard ratio [HR], 2.3 [1.6–3.2], P = 2.7E-5) and urine X-12117 (HR, 1.7 [1.3–2.2], P = 1.9E-5) were risk factors of UACR doubling, whereas urine sebacate (HR, 0.86 [0.80–0.92], P = 1.9E-5) was inversely associated. Compared with clinical characteristics alone, including the top five endogenous metabolites in serum and urine associated with longitudinal outcomes improved the outcome prediction (area under the receiver operating characteristic curves for eGFR decline: clinical model=0.79, clinical+metabolites model=0.87, P = 8.1E-6; for UACR doubling: clinical model=0.66, clinical+metabolites model=0.73, P = 2.9E-5). CONCLUSIONS: Metabolomic profiling in different biofluids provided distinct and potentially complementary insights into the biology and prognosis of kidney diseases.