AI-Guided De Novo Design of a Caffeine-Induced Protein Dissociation System

Chemically induced proximity (CIP) and chemically disrupted proximity (CDP) technologies have transformed biological research by enabling precise temporal control of protein-protein interactions and cellular functions. However, despite the extensive development of CIP systems, CDP tools with strong translational potential remain comparatively underdeveloped, calling for the need to expand the CDP toolkit. Here, we present a caffeine-operated dissociation system (CODS), a CDP platform activated by caffeine-an inexpensive and widely available small molecule from food or beverages with a well-established safety profile. Using an AI-guided de novo protein design framework, we reprogrammed an existing caffeine-responsive CIP module into a ligand-dependent dissociation system. CODS exhibits high sensitivity and rapid functional switching, with an EC₅₀ below 90 nM and minute-scale dissociation and reassociation kinetics. We further demonstrate the utility of CODS across diverse cellular contexts, including caffeine-dependent suppression of gene expression, induction of programmed cell death via pyroptosis, and conditional deactivation of chimeric antigen receptor (CAR) T-cell activity. CODS establishes a broadly applicable CDP platform for tunable control of cellular functions and provides a potentially generalizable strategy for engineering chemically controlled dissociation systems from existing CIP architectures.