Rapid Molecular Evaluation of Human Kidney Tissue Sections by In Situ Mass Spectrometry and Machine Learning to Classify the Nephrotic Syndrome

Nephrotic syndrome (NS) is classified based on morphological changes of glomeruli in biopsied kidney tissues evaluated by time-consuming microscopy methods. In contrast, we employed desorption electrospray ionization mass spectrometry (DESI-MS) directly on renal biopsy specimens obtained from 37 NS patients to rapidly differentiate lipid profiles of three prevalent forms of NS: IgA nephropathy (n = 9), membranous glomerulonephritis (n = 7), and lupus nephritis (n = 8), along with other types of glomerular diseases (n = 13). As we noted molecular heterogeneity in regularly spaced renal tissue regions, multiple sections from each biopsy specimen were collected, providing a total of 973 samples for investigation. Using multivariate analysis, we report differential expressions of glycerophospholipids, sphingolipids, and glycerolipids among the above four classes of NS kidneys, which were otherwise overlooked in several past studies correlating lipid abnormalities with glomerular diseases. We developed machine learning (ML) models with the top 100 features using the support vector machine, which enabled us to discriminate the concerned glomerular diseases with 100% overall accuracy in the training, validation, and holdout test set. This DESI-MS/ML-based tissue analysis can be completed in a few minutes, in sharp contrast to a daylong procedure followed in the conventional histopathology of NS.