Within each individual’s immune system is a unique repertoire of billions of T cells patrolling the body to uncover and eliminate foreign threats. Yet this army is not foolproof and can miss the subtle signs of cancer or viral infection that can lead to advanced disease.
In a step toward enhancing disease detection for scalable personalized immunotherapy, researchers from the lab of Nobel Laureate David Baker, PhD, have now leveraged artificial intelligence (AI) to design highly specific, low immunogenicity protein binders to recognize evasive disease markers for targeted cell killing. The work was published in a new study in Science titled, “Design of high specificity binders for peptide-MHC-complexes.”
Co-lead authors Julia Bonzanini, Nathan Greenwood, and Bingxu Liu, PhD (left to right) [Ian Haydon]Diseased cells are identified by T-cell receptors (TCRs) through peptide antigens presented on the cell surface by major histocompatibility complex (MHC) proteins. While therapeutic targeting of MHCs has been effective in specific disease indications within small patient cohorts, this silver bullet has struggled to capture the exponentially large MHC allele diversity required to scale to broader populations.
“During every immune response, we’re rolling dice to see if we’re lucky enough to have a TCR that recognizes the threat,” said Bingxu Liu, PhD, a postdoctoral researcher in the Baker lab, in an interview with GEN. “Using protein design, we can get highly specific binders to these MHCs in a fast and cheap way for a generalized platform.”
The Baker lab’s AI-based pipeline facilitates rapid design and validation of protein binders that can recognize peptides from a diverse array of viral and tumor-associated proteins, including fragments from HIV and the tumor antigen, PRAME. PRAME is a historically challenging target as the atomic details of how the protein gets displayed to the immune system have not been experimentally determined.
Baker, who is a Howard Hughes Medical Institute (HHMI) investigator and the director of the Institute for Protein Design at the University of Washington (UW), was awarded the 2024 Nobel Prize in Chemistry for his work in pioneering AI for protein design. In recent years, his lab has developed RFdiffusion, a diffusion-based AI model that generates proteins from scratch (de novo) to enable broad applications across cancer immunotherapy, enzyme catalysis, and neutralizing snake toxin. Recently, the Baker lab published a new AI framework to drug historically “undruggable” intrinsically disordered proteins.
That’s so specific
The study’s computational approach offers an advance from labor-intensive and time-consuming experimental methods that rely on engineering new TCRs to enhance immune function. TCR screening is additionally limited to the availability of donors with relevant human leukocyte antigen (HLA) alleles for the disease indication, leading to low hit rates and lack of specificity.
The authors generated binders for eleven total target peptide-MHC complexes (pMHCs). Upon activation into chimeric antigen receptors (CARs), designs for eight targets successfully triggered T-cell activation. Two designs enabled targeted killing of human cells bearing disease markers in the lab.
Notably, binders demonstrated impressive specificity with minimal off-target effects to confer low immunogenicity risk. When placed in CAR T constructs, binders triggered T-cell activation for the target peptide, but not for closely related peptides that were only one amino acid away.
Nathan Greenwood, a research scientist in the Baker lab and co-lead author of the study, recalled that the beginnings of the project excelled at designing binders that hit targets, yet tuning the pipeline for high specificity was another milestone.
“We had a really hard time trying to get those binders to be super specific so that they wouldn’t go downstream and be toxic to other cells,” Greenwood told GEN. “Once we started refining our pipeline, we were able to get successful hits on almost all targets to see that the platform was truly generalizable.”
Julia Bonzanini, a graduate student in the Baker lab and another co-lead author on the paper, highlights that this precision immunotherapy project is distinct from the group’s other protein design efforts because of its highly translational focus.
“We eventually want these de novo proteins to end up in humans. That’s the biggest difference from a lot of other projects in the lab that may focus on optimizing a protein to be stable or perform a function,” she told GEN.
Looking ahead, Bonzanini is interested in expanding the work to monomorphic MHCs that have limited variation in their structure within a population, offering a powerful cost-effective path to extend patient reach. She also emphasizes that seeing AI-designed proteins come to fruition in the lab is a technological feat that should not be understated.
“It’s mind–blowing that something you were working on in a computer weeks ago, that was just a bunch of helices and beta sheets, is now in an actual cell doing something that might benefit cancer patients downstream,” she said.
The team plans to launch a company based on the technology, with public details to come at a later date.
The post AI-Designed Proteins Enable Scalable Precision Immunotherapy appeared first on GEN – Genetic Engineering and Biotechnology News.
