Synthetic Biology · global
Caffeine Makes Proteins Let Go: An AI-Designed Reversible Switch Explores a New Approach to Controlling CAR T Cells
Once cell therapies enter the body, how can their immune attack be reined in at the right time? A Texas A&M University team has designed a caffeine-controlled protein switch that reduces CAR T-cell activation in the laboratory, offering a new approach to adjustable therapies, though multiple challenges remain before clinical use.
Cancer-fighting immune cells need a sufficiently strong attack, as well as a mechanism for pulling back. CAR T-cell therapy has transformed treatment for some blood cancers, but an excessive immune response can cause serious complications. A Texas A&M University research team has therefore explored adding a reversible molecular switch to engineered cells: using caffeine to break connections between proteins and reduce cellular activity.
The research on this caffeine-operated dissociation system, called CODS, was published in the Journal of the American Chemical Society on May 25, 2026. A research overview provided by the university to Phys.org in June, along with interviews published by the campus newspaper The Battalion in September, described how the same work progressed from protein design to tests of cellular function; the evidence so far remains laboratory-based.
At the heart of the switch is an artificially designed binding protein paired with COSMO, a caffeine-responsive protein module previously developed by the team. Without caffeine, the two remain bound; when caffeine is added, COSMO tends to pair with other COSMO molecules, displacing the binding protein. Removing caffeine allows the original complex to form again, making the protein connection reversible.
AI's task in this study was to design protein components that could work with COSMO. According to The Battalion's interviews, the team used tools including BindCraft, AlphaFold, and ProteinMPNN, testing 13 candidate binding proteins after computational screening to identify a workable design. The challenge was to balance binding strength: the connection had to be stable enough under normal conditions, yet able to release in the presence of caffeine. The university's high-performance computing resources also supported the design process and molecular simulations.
The team then incorporated CODS into different cellular systems. According to the university's overview, adding caffeine could suppress the activity of engineered gene circuits, while removing it allowed activity to recover; another configuration could induce an inflammatory form of cell death called pyroptosis. This also shows that the outcome of protein dissociation depends on which cellular pathway the switch is connected to, and cannot universally be regarded as shutting down a cell.
The demonstration closest to a therapeutic application involved integrating the switch into a split CAR system. In laboratory tests, caffeine reduced CAR T-cell activation, supporting the concept of temporarily regulating immune cells with an externally supplied molecule. However, reduced activation does not mean the system has been shown to reduce cytokine release syndrome or neurotoxicity in patients, nor has it been established whether anticancer effects can be maintained after a pause.
Caffeine serves here as a control signal for engineered cells; the research did not establish that drinking coffee can regulate existing CAR T therapies. Before moving toward clinical use, CODS still needs validation in therapeutic cells, animal models, and disease-relevant settings. The concentrations required in the human body, whether control is sufficiently reliable, and the safety of the newly designed proteins will determine whether this molecular switch can become a practical therapeutic tool.