P06.05 De novo designed protein binders against cancer targets of the B7-H family

Background Artificial intelligence-guided protein design offers a powerful approach for developing novel therapeutic agents, particularly in cases where conventional antibody or single-domain antibody development proves challenging. However, computational predictions have to be carefully considered and need to be validated for target affinity and specificity. Suitable targets for such a platform establishment are well known molecules, such as PD-L1, and closely related proteins, here comprising CD276 and VTCN1 from the same B7-H family.Material and Methods Using RFdiffusion, a machine learning-based pipeline derived from RoseTTAFold, we generated libraries of de novo designed proteins targeting members of the B7-H family to develop a screening and validation pipeline. Using a mammalian cell surface system, we screened thousands of potential protein binders for their target recognition. Further validation and characterization of binder subsets was performed using flow cytometry, chimeric antigen receptor (CAR)-T cells, and surface plasmon resonance (SPR) spectroscopy.Results We established a high-throughput, cell-surface expression-based screening platform capable of selecting target-specific binders from libraries comprising thousands of candidates. In a first approach, several binders against PD-L1 (B7-H1) and CD276 (B7-H3) were identified and validated via flow cytometry, SPR spectroscopy, and interface mutant analysis. These mini-proteins demonstrated high target affinity, small molecular size, and remarkable thermal stability, making them versatile tools for biomedical and clinical applications. Using CAR-T cells, we were able to test the binders’ functionality and target specificity in a clinically relevant setting.Conclusion Altogether, this platform offers a versatile and scalable framework for the rapid development of de novo designed proteins with potential applications in research, diagnostics, and therapy.