Metabolomic Markers of Kidney Function Decline in Patients With Diabetes: Evidence From the Chronic Renal Insufficiency Cohort (CRIC) Study

Abstract

Rationale & objective: Biomarkers that provide reliable evidence of future diabetic kidney disease (DKD) are needed to improve disease management. In a cross-sectional study, we previously identified 13 urine metabolites that had levels reduced in DKD compared with healthy controls. We evaluated associations of these 13 metabolites with future DKD progression.

Study design: Prospective cohort.

Setting & participants: 1,001 Chronic Renal Insufficiency Cohort (CRIC) participants with diabetes with estimated glomerular filtration rates (eGFRs) between 20 and 70mL/min/1.73m² were followed up prospectively for a median of 8 (range, 2-10) years.

Predictors: 13 urine metabolites, age, race, sex, smoked more than 100 cigarettes in lifetime, body mass index, hemoglobin A_1c level, blood pressure, urinary albumin, and eGFR.

Outcomes: Annual eGFR slope and time to incident kidney failure with replacement therapy (KFRT; ie, initiation of dialysis or receipt of transplant).

Analytical approach: Several clinical metabolite models were developed for eGFR slope as the outcome using stepwise selection and penalized regression, and further tested on the time-to-KFRT outcome. A best cross-validated (final) prognostic model was selected based on high prediction accuracy for eGFR slope and high concordance statistic for incident KFRT.

Results: During follow-up, mean eGFR slope was-1.83±1.92 (SD) mL/min/1.73m² per year; 359 (36%) participants experienced KFRT. Median time to KFRT was 7.45 years from the time of entry to the CRIC Study. In our final model, after adjusting for clinical variables, levels of metabolites 3-hydroxyisobutyrate (3-HIBA) and 3-methylcrotonyglycine had a significant negative association with eGFR slope, whereas citric and aconitic acid were positively associated. Further, 3-HIBA and aconitic acid levels were associated with higher and lower risk for KFRT, respectively (HRs of 2.34 [95% CI, 1.51-3.62] and 0.70 [95% CI, 0.51-0.95]).

Limitations: Subgroups for whom metabolite signatures may not be optimal, nontargeted metabolomics by flow-injection analysis, and 2-stage modeling approaches.

Conclusions: Urine metabolites may offer insights into DKD progression. If replicated in future studies, aconitic acid and 3-HIBA could identify individuals with diabetes at high risk for GFR decline, potentially leading to improved clinical care and targeted therapies.

Keywords: Biomarker; Chronic Renal Insufficiency Cohort (CRIC); chronic kidney disease (CKD); diabetes; end-stage renal disease (ESRD); estimated glomerular filtration rate (eGFR); incident kidney failure; kidney disease progression; kidney function decline; longitudinal study; metabolomics; multivariate model; prediction; prognosis; risk factor.

Figure 1:

Histogram of predicted eGFR slopes (N=1001). The eGFR slopes were estimated via linear mixed effects model with a mean of −1.83 +/− 1.92 (SD) ml/min/1.73m² per year.

Figure 2:

Boxplots of model prediction performance: 100 repeats of 5-fold cross-validated (a) Mean-squared error (MSE) for eGFR slopes and (b) C-statistics for time-to-incident KFRT. Model type: C: Clinical-variables only CM-A: Clinical-variables + All 13 metabolites CM-L: LASSO – Forced clinical-variables + Selection from 13 metabolites CM-S: Stepwise AIC – Forced clinical-variables + Selection from 13 metabolites M-A: All 13 Metabolites M-L: LASSO – Selection from 13 metabolites M-S: Stepwise AIC - Selection from 13 metabolites

Figure 2:

Boxplots of model prediction performance: 100 repeats of 5-fold cross-validated (a) Mean-squared error (MSE) for eGFR slopes and (b) C-statistics for time-to-incident KFRT. Model type: C: Clinical-variables only CM-A: Clinical-variables + All 13 metabolites CM-L: LASSO – Forced clinical-variables + Selection from 13 metabolites CM-S: Stepwise AIC – Forced clinical-variables + Selection from 13 metabolites M-A: All 13 Metabolites M-L: LASSO – Selection from 13 metabolites M-S: Stepwise AIC - Selection from 13 metabolites