Discriminant analysis in the urine of patients with glycemic alterations based on Nuclear Magnetic Resonance.
Type 2 diabetes, metabolomics, urine, nuclear magnetic resonance.
Metabolomics is applied to investigate the pathophysiological mechanisms of several diseases, such as diabetes. The metabolomic profile is obtained from biological samples, with urine being preferred due to reliability and ease of obtaining. Improving strategies to discriminate between normal and pathological metabolites of certain diseases is essential for the clinical application of metabolomics. The study's objective was to characterize the population formed by individuals with glycemic alterations and to test the feasibility of using Nuclear Magnetic Resonance as a spectroscopic method for analyzing metabolomics in urine. The study population was divided into three groups: DM2 Group - patients with type 2 diabetes (DM2); the PDM Group - patients with prediabetes (PDM); and the control group (C) formed by healthy individuals. Participants were characterized by demographic, clinical, lifestyle, anthropometric, diet, glycemic, and lipid profiles. The 24h urine spectra were acquired from NMR, and later, the data were analyzed by Principal Component Analysis (PCA), followed by unsupervised analysis. A series of algorithms were tested to verify which model had better accuracy, sensitivity, and specificity. The fasting glucose, HbA1c, and HOMA-IR values of the T2DM and PDM groups were significantly different from the control. In addition, the PDM and DM groups had high waist circumference values. Diet composition did not differ between groups and was adequate in terms of the proportion of macronutrients and inadequate in total fiber. The set of signals from the metabolites based on the chemical shifts (peaks) obtained resulted in identifying 21 distinct regions of metabolites, among which glycine, urea, glucose, acetate, citrate, and creatinine stand out. Four showed a significant difference in the following ppm ranges (3.27, 3.37, 3.5.9, 3.75, and 9.29). APC scores showed partial separation between groups. The GA-LDA model was the most responsive algorithm for the discrimination of the urine of the groups in terms of accuracy, sensitivity, and specificity.