Longitudinal non-targeted metabolomic profiling of urine samples for monitoring of kidney transplantation patients

Abstract

The assessment of kidney function within the first year following transplantation is crucial for predicting long-term graft survival. This study aimed to develop a robust and accurate model using metabolite profiles to predict early long-term outcomes in patient groups at the highest risk of early graft loss. A group of 61 kidney transplant recipients underwent thorough monitoring during a one-year follow-up period, which included a one-week hospital stay and follow-up assessments at three and six months. Based on their 12-month follow-up serum creatinine levels: Group 2 had levels exceeding 1.5 mg/dl, while Group 1 had levels below 1.5 mg/dl. Metabolites were detected by mass spectrometer and first pre-processed. Univariate and multivariate statistical analyses were employed to identify significant differences between the two groups. Nineteen metabolites were found to differ significantly in the 1st week, and seventeen metabolites in the 3rd month (adjusted p-value < 0.05, quality control (QC) < 30, a fold change (FC) > 1.1 or a FC < 0.91, Variable Influence on Projection (VIP) > 1). However, no significant differences were observed in the 6th month. These distinctive metabolites mainly belonged to lipid, fatty acid, and amino acid categories. Ten models were constructed using a backward conditional approach, with the best performance seen in model 5 for Group 2 at the 1st-week mark (AUC 0.900) and model 3 at the 3rd-month mark (AUC 0.924). In conclusion, the models developed in the early stages may offer potential benefits in the management of kidney transplant patients.