Integration of Spatial Metabolomics and Machine Learning Reveal Glycogen as an Actionable Target for Pulmonary Fibrosis

Abstract

Spatial metabolomics is revolutionizing our understanding of the tissue microenvironment. To this end, the application of artificial intelligence (AI) to spatial metabolomics for clinical research are beginning to emerge. Here, we demonstrate the application of machine learning to high-dimensionality reduction and spatial clustering (HDR-SC), histopathological annotation, and histopathological prediction of matrix-assisted laser desorption/ionization-mass spectrometry imaging (MALDI-MSI) datasets for the comprehensive assessment of tissue metabolic heterogeneity in human formalin-fixed paraffin-embedded (FFPE) sections of lung diseases. Using this approach, we identified a class of carbohydrate features unique to pulmonary fibrosis (PF) with nearly 100% accuracy at histopathological identification of fibrotic regions within both human and mouse lung diseases. Further, HDR-SC observed decreasing levels of glycogen from early- to end-stage fibrosis, suggesting glycogen is utilized during disease progression. A mouse model of glycogen utilization deficiency exhibited a nearly 70% reduction in fibrosis development when compared to WT animals in the bleomycin-induced PF model. This study establishes a new workflow for the integration of AI and MALDI-MSI complex carbohydrate datasets and identifies glycogen metabolism as a previously unknown metabolic event and a future therapeutic target for the treatment of PF.

Keywords: MALDI-imaging, digital pathology, glycogen, N-linked glycosylation, pulmonary fibrosis, machine learning