Diverse Intracellular Trafficking of Insulin Analogs byMachine Learning-based Colocalization and DiffusionAnalysis
Insulin signaling is crucial for maintaining cellular function and systemic homeostasis, with its dysregulation leading to metabolic disorders in particular diabetes. While insulin analogs are essential in type 1 diabetes treatment, their intracellular trafficking and sorting compared to endogenous insulin remains largely unresolved at the molecular level. Understanding these are important in improving therapeutics and guiding future drug development. However, current methods rely on static imaging and bulk receptor assays conducted in non-physiological conditions, which often disrupt native insulin signaling and provide only snapshots that fail to capture the temporal dynamics of insulin trafficking and characteristic sorting pathways. Here, we directly observe insulin trafficking and intracellular sorting and compare the characteristics of Atto655 labeled recombinant human insulin (HI655) with insulin aspart (IAsp655), a clinically approved, rapid-acting insulin analog. We developed a combined approach integrating Colocalization Fingerprinting, a machine learning framework for reliable, time-resolved colocalization analysis, together with our recently developed deep learning-assisted single-particle diffusional analysis (DeepSPT). Our analysis revealed subtle, yet significant differences in intracellular behavior between IAsp655 and HI655, particularly in diffusional behavior and lysosomal colocalization, highlighting the potential of our approach to decipher subtle differences in intracellular trafficking and sorting characteristics. In addition to contributing to a more detailed understanding of biology of insulin analogs and intracellular sorting, we provide a reliable machine-learning methodology to study intricate cellular processes.