Generative Diffusion-Based De Novo Protein Design of Lipid/CD1-Restricted TCR Mimics

De novo protein design has experienced a renaissance in recent years due to advances in generative diffusion-based approaches to develop high affinity, target-specific binders. However, the ability of these tools to generate “new-to-nature” proteins with unique backbones and amino acid sequences that target lipid/protein assemblies has not yet been explored. Here, we apply an end-to-end design pipeline using RFdiffusionAA, LigandMPNN, Chai-1, and Rosetta to develop lipid antigen/CD1-restricted T cell receptor (TCR) mimics. As a proof of principle, the pipeline is applied to design TCR mimics targeting the α-GalCer/CD1d complex that is the prototypical iNK T cell ligand. Designed TCR mimics were recombinantly expressed in E. coli with secondary structure folds and thermal stabilities validated in vitro. Biophysical assays confirmed that the designed TCR mimics bound to their molecular targets in both a lipid and CD1 isoform-restricted manner. Finally, X-ray crystallography and solution NMR provided mechanistic insights into the design/target binding modes, which were consistent with the computational design models. We are now applying the pipeline to target the urushiol (C15:2)/CD1a complex implicated in poison-ivy dermatitis and the DDM/CD1b complex critical for Mycobacterium tuberculosis infection. Together, our results are the first to demonstrate that generative diffusion-based de novo protein design approaches can robustly produce lipid/CD1 targeted TCR mimics with high affinity and specificity. This work will enable new therapeutic avenues for diagnostics, immunotherapy, and vaccine development centered on lipid/CD1 antigen presentation.