Non-invasive urinary FTIR spectroscopy for early diagnosis and monitoring of diabetic nephropathy.

Diabetic nephropathy (DN) is a progressive and life-threatening complication of diabetes mellitus (DM). However, developing sensitive and specific non-invasive diagnostic approaches remains a critical clinical challenge. In this study, we characterized the biochemical and metabolic alterations in the kidney and urine using attenuated total reflection-Fourier transform infrared (ATR-FTIR) spectroscopy in a type 2 diabetic db/db mouse model at 7, 12, and 21 weeks. Spectral analysis revealed distinct, time-dependent disruptions in the vibrational bands associated with protein, lipid, and carbohydrate components in both kidney and urine samples. Principal component analysis (PCA) effectively distinguished diabetic mice from controls, capturing spectral alterations associated with the progression of DN. Based on these discriminative features, we established two independent machine learning frameworks for kidney and urine data to classify temporal groups and to evaluate whether urinary spectral profiles could non-invasively reflect renal pathology. Among all evaluated classifiers, Partial Least Squares Discriminant Analysis (PLS-DA) exhibited the highest classification accuracy across time points, achieving 97.50% for renal spectra and 92.50% for urinary spectra. To quantitatively assess renal injury, we developed a partial least squares regression (PLSR) model using renal spectra to predict hematoxylin and eosin (H&E)-derived scores. Furthermore, urinary FTIR spectra were used to predict renal H&E scores, and a strong linear correlation was observed between predicted and reference values. This analysis revealed a strong association between urinary FTIR spectra and histologically derived renal injury scores (R2CV = 0.846), indicating that urinary spectral signatures are significantly associated with renal pathological changes in this model. Collectively, these findings support urinary FTIR profiling as a rapid, cost-effective, and non-invasive complementary approach to traditional tissue-based assessments of DN, enabling early detection and longitudinal monitoring of the disease. Future studies should evaluate the clinical validity and translational applicability of this technique in independent cohorts.