From growth hormones to cancer drugs, small molecules play a crucial role in our health. Monitoring them is essential to keeping us healthy; it enables physicians to calculate dosages and patients to monitor their medical conditions at home, for example.
Monitoring small molecules depends on sensing where they are, and in what concentrations. While scientists have developed sensors to detect some small molecules, these sensors are used primarily in research and drug discovery and can only detect a limited range of molecules with particular qualities. There is a compelling need for sensors that can detect and signal the presence of diverse small molecules of different shapes, sizes, flexibility and polarity.
Using artificial intelligence (AI), a team of scientists led by Nobel Prize winner David Baker at the University of Washington has created a computational method for generating proteins that bind and signal a wide range of small molecules with great effectiveness. Baker won the 2024 Nobel Prize in Chemistry for computational protein design.
The research described here, published in Science and conducted in part at the Advanced Photon Source (APS), exemplifies that approach. The APS is a U.S. Department of Energy (DOE) Office of Science user facility at DOE’s Argonne National Laboratory.
The sensor design problem
Creating a protein sensor for small molecules is very difficult. The protein must first bind to the small molecule, then signal its presence.
The team solved both problems with modular design strategies. Their AI-generated proteins consist of identical repeating subunits surrounding a central cavity. The cavity holds a pocket where the small molecule binds.
The subunits, being modular, are easily disassembled. In this way, the small molecule binding proteins can be treated like Lego blocks and be connected to well-established signaling proteins (such as split green fluorescent protein, or GFP), to make a full sensing protein device. When a small molecule binds in the pocket, the subunits reassemble, which leads to the signaling module sending a signal that the small molecule is present.
First step: Binding
The team chose a diverse spectrum of ligands (molecules that bind to protein receptors to send signals between cells), including cholic acid, a biomarker for liver disease; methotrexate, a cancer drug, which requires regular monitoring; thyroxine, a human hormone that indicates thyroid conditions; and a cyclic peptide.
The scientists constructed a machine learning algorithm based on AlphaFold2 (a protein structure predictor whose developers, John Jumper and Demis Hassabis, shared the Nobel Prize in Chemistry with Baker) and other machine learning protein design algorithms to generate thousands of proteins to bind the small molecules.
After computational design, the team tested the designed proteins in the laboratory and identified binders to particular ligands, following computational design and using machine learning methods to choose the best designs for experimental tests.
To confirm the accuracy of their design approach, the Baker team turned to the APS. They used the ultrabright X-ray beams to collect data on the atomic structure of the binding proteins. Using the Northeastern Collaborative Access Team (NE-CAT) beamlines at 24-ID at the APS, the team determined the structures of crystals formed from one of the designed proteins.
“Prediction algorithms are excellent tools, but without verification of the structures, there’s no proof that the predictions match reality,” said Kay Perry of Cornell University, staff scientist at NE-CAT. “X-ray crystallography remains one of the best ways to make that confirmation, and the team was able to do so in this case.”
Second step: Signaling
The next challenge was turning the binding proteins into signaling proteins. The scientists took advantage of their modularity to create two different types of signaling events.
The team built ligand-induced dimerization proteins from the binders. Linna An, the first author of this study, said the technology can be used in many health-related applications, such as regulating the release of drugs in cancer therapies.
In a different type of signaling event, the scientists fused the binding proteins to a newly designed nanopore, a protein creating a channel allowing ion flow. The fused unit was constructed in such a way that when a small molecule blocked the binding pocket, the whole nanopore was blocked, preventing the flow of ions and loss of current. Loss of current signaled the presence of the small molecule.
Read more on APS website
Image: The crystal structure of CHD_r1 (gray) is very similar to the computational design model (colored).
Credit: Linna An, et al., Science.
