#2565 Urine amino acid profiles enlighten new prediction models for renal Fanconi syndrome and its association with eGFR level

Abstract Background and Aims Renal Fanconi syndrome (FS) presented dysfunctions of renal proximal tubular (PT) transport with or without excessive urine amino acids (AAs) excretion, which might contribute to the progression of eGFR decline. This study aimed to explore the clinical benefit of 21 urinary AAs excretion patterns in renal FS and eGFR decline. Method Liquid chromatography-tandem Mass spectrometry (LC-MS-MS) is currently quantitative and detects over 21 types of amino acids in 157 patients with renal proximal tubular dysfunctions. To evaluate the diagnostic values of urine AAs in renal FS, we established a machine learning algorithm to perform binary classification using Random Forest Regressor in Python (version 3.8.8) with a training set and a testing set (8:2). We used Randomized search to find out the best model hyperparameters, calculated feature importance for contribution evaluation, and Receiver Operating Characteristic (ROC) curves to evaluate the performance of classification algorithms. Results A model based on a machine learning algorithm of all urine AAs to diagnose renal FS established with sensitivity = 0.71, specificity = 0.73, and AUC = 0.83 (95% CI 0.648-0.957) (Fig. 1), with similar performance in combined six urine AAs (citrulline, asparagine, threonine, phenylalanine, ornithine, and proline) models. The urine proline level independently correlated with eGFR < 60 ml/min/1.73 m2 (bias regression coefficient 0.84, P = 0.001, OR = 2.31, 95% CI 1.41-3.80) when adjusted by age, etiology, gender, and blood urea nitrogen level. For the primary diseased diagnosis, the most susceptible urine AAs were citrulline (56.7%), proline (56.0%,) threonine (52.2%), and ornithine (45.9%) and the less susceptible ones were histidine (24.8%), aspartate (16.6%), and arginine (13.4%). Accordingly, the most common affected AA transporters were SLC6A19 (B0AT1) and etiology-specific urine AA atlas major in the eight urine AAs (isoleucine, glutamine, methionine, proline, alanine, ornithine, lysine, and arginine). Different from primary Sjӧgren syndrome patients (n = 26) with less susceptible, the patients with plasmacyte diseases (n = 8) showed elevations of almost all the urine AAs except glycine and aspartate. The connection between urine AAs excretion and hypouricemia showed gender differences, with 20 urine AAs significantly negatively correlating with serum uric acid levels in males but only five in females. Conclusion We first established a model based on the urine AA atlas detected by (LC-MS-MS) which contributed to differentiating the etiologies of renal tubular dysfunctions and associated well with eGFR. It shed light on the clinical practice in precision evaluation of tubule injury.