AI-designed molecular switch uses caffeine to control engineered cells

For many of us, a warm cup of coffee is how we start our day. For Texas A&M Health researchers, it may also offer a new way to control engineered cells in future medicines.

A team at the Texas A&M Health Institute of Biosciences and Technology has developed an artificial intelligence-designed molecular switch that uses caffeine to rapidly separate engineered proteins inside living cells and trigger cellular responses on demand. The platform, called CODS, short for caffeine-operated dissociation system, could help scientists build safer and more controllable gene and cell therapies.

The study, published in the Journal of the American Chemical Society, was led by Yubin Zhou, MD, PhD, FAAAS, FAIMBE, FRSC, director of the Center for Translational Cancer Research at the Institute of Biosciences and Technology and professor in the Texas A&M Naresh K. Vashisht College of Medicine, together with Tianlu Wang, PhD, and colleagues. Graduate students Brendan McKee and Tatsuki Nonomura played central roles in the work, with McKee driving the AI-guided protein design and computational modeling efforts and Nonomura leading key molecular engineering and live-cell validation studies.

AI is changing how we design biology. Instead of relying only on protein parts that already exist in nature, we can now design new mini proteins with specific behaviors. Here, we used AI to help turn caffeine into a precise trigger for controlling engineered cells."

Yubin Zhou, MD, Professor, Texas A&M Naresh K. Vashisht College of Medicine

AI as a molecular architect

The new work builds on Zhou's earlier caffeine-responsive technologies but moves in a distinctly different direction.

Previous systems showed that caffeine could help pull engineered proteins together. However, CODS does the opposite: It uses caffeine to pull proteins apart. That difference matters because future therapies may need ways not only to activate cells, but also to pause, quiet or reset them when needed.

To build CODS, the team used AI-guided protein design to create a small synthetic binder which recognizes a caffeine-responsive protein module. The binder holds the system together when caffeine is absent, and when caffeine is added, the proteins separate.

In this way, CODS acts like a molecular clasp. Without caffeine, the clasp stays closed. With caffeine, the clasp opens.

"Many genetically-encoded molecular tools act like accelerators," Wang said. "CODS gives us something closer to a brake or pause button."

High-performance computing

The AI-driven design process required substantial computational power. The team used protein-design algorithms and molecular simulations to identify, evaluate and refine synthetic binders before testing the most promising candidates in living cells.

This work was enabled by the Texas A&M High Performance Research Computing (HPRC) service, which provided the computing power needed to run advanced AI-driven protein design workflows at scale.

"High-performance computing was essential for this project," Zhou said. "AI protein design is computationally demanding. The Texas A&M HPRC service helped us move from a conceptual idea to a functional molecular switch much faster."

The resulting system responded to very low caffeine concentrations, worked within minutes and could be reversed repeatedly by adding or removing caffeine.

Controlling genes, cell death and immune cells

The researchers demonstrated CODS in three major ways.

First, they used it to control gene activity. Without caffeine, an engineered gene circuit remained active. When caffeine was added, CODS separated the target proteins needed to keep the gene turned on, sharply reducing gene activity. Removing caffeine allowed the system to recover.

Second, the team used CODS to control programmed cell death. By rewiring a cell-death protein with the caffeine-responsive switch, they created a system in which caffeine could trigger inflammatory cell death, known as pyroptosis. This could help scientists study inflammation and may one day support the design of therapeutic cells that can be eliminated when needed.

Finally, the most translational demonstration involved CAR T-cells-immune cells engineered to recognize and attack cancer. CAR T-cell therapies have produced remarkable results in some blood cancers, but they can also cause serious side effects when immune cells become too active. A caffeine-induced safety switch could give clinicians a way to temporarily reduce CAR T-cell activity without permanently destroying the therapeutic cells.

Using CODS, this Texas A&M team built a split CAR system that remains active when caffeine is absent but remains passive when caffeine is added. In laboratory tests, caffeine strongly reduced CAR T-cell activation, suggesting that CODS could become a practical safety OFF switch for engineered immune cells.

Beyond coffee: Toward programmable medicine

Zhou emphasized that caffeine itself is not a cancer treatment. Instead, caffeine serves as a safe and familiar signal that can communicate with specially engineered cells.

"Coffee will not replace medicine," Zhou said. "But caffeine can help us imagine medicines that are more controllable, more responsive and safer for patients."

The broader advance is the use of AI to design new proteins that behave in ways nature does not readily provide. Similar strategies could eventually be used to build switches controlled by other familiar molecules, over-the-counter drugs or clinically approved medicines.

Before CODS can move toward clinical use, the system will need further testing in therapeutic cells, animal models and disease-relevant settings. Still, the study marks an important step toward programmable medicine in providing a framework for designing therapies that can be adjusted after they are delivered.

"Powerful therapies need powerful control," Zhou said. "By combining AI-designed proteins, high-performance computing and familiar small molecules, we are building a new language for communicating with engineered cells."