Deep-Learning-Assisted Single-Molecule Tracking on a Live Cell Membrane.

Abstract

Single-molecule fluorescence imaging is a powerful tool to study protein function by tracking molecular position and distribution, but the precise and rapid identification of dynamic molecules remains challenging due to the heterogeneous distribution and interaction of proteins on the live cell membrane. We now develop a deep-learning (DL)-assisted single-molecule imaging method that can precisely distinguish the monomer and complex for rapid and real-time tracking of protein interaction. This DL-based model, which comprises convolutional layers, max pooling layers, and fully connected layers, is trained to reach an accuracy of >98% for identifying monomer and complex. We use this method to investigate the dynamic process of chemokine receptor CXCR4 on the live cell membrane during the early signaling stage. The results show that, upon ligand activation, the CXCR4 undergoes a dynamic process of forming a receptor complex. We further demonstrate that the CXCR4 complex tends to be internalized at 2.5-fold higher rate into the cell interior than the monomer via the clathrin-dependent pathway. This study is the first example to scrutinize the early signaling process of CXCR4 at the single-molecule level on the live cell membrane. We envision that this DL-assisted imaging method would be a broadly useful technique to study more protein families for elucidating their physiological and pathological functions.