Diverse intracellular trafficking of insulin analogs by machine learning-based colocalization and diffusion analysis

Insulin signaling is vital for cellular homeostasis, with dysregulation leading to severe metabolic disorders, particularly diabetes. While insulin analogs are crucial in type 1 diabetes treatment, identifying potential variations in intracellular trafficking and sorting from endogenous insulin is challenging. Current methods rely on static imaging and bulk receptor assays in non-physiological conditions, which disrupts native signaling and masks temporal trafficking dynamics. Here, we directly recorded and compared the intracellular trafficking of ATTO⁶⁵⁵-labeled recombinant human insulin (HI⁶⁵⁵) and rapid-acting analog insulin aspart (IAsp⁶⁵⁵) in live cells. We developed a platform combining Colocalization Fingerprinting, a machine-learning framework for time-resolved colocalization, with our deep learning-assisted single-particle diffusional analysis (DeepSPT). Our analysis revealed significant intracellular trafficking differences between IAsp⁶⁵⁵ and HI,⁶⁵⁵ both in diffusional behavior and lysosomal colocalization. The method offers a detailed understanding of insulin analog biology and provides a reliable machine-learning methodology to identify subtle variations in intracellular pathways of intricate cellular processes.