AI-Guided Hydrophobic Core Design of Robust Six-Helix Bundle Proteins.

Abstract

α-Helical domains are widespread and versatile, yet typically fail under low mechanical load because backbone hydrogen bonds unzip sequentially, limiting their use in force-bearing nanomaterials and molecular devices. We present an AI-guided strategy to design six-helix bundle proteins with densely packed hydrophobic cores that co-optimize mechanical and thermal stability. Backbones were generated with RFdiffusion, sequences designed with ProteinMPNN, and structures validated by AlphaFold2/ESMFold; steered and annealing molecular dynamics simulation identified designs with high predicted unfolding forces and heat resilience. Three selected constructs (HP149, HP206, HP347) expressed solubly and folded as predominantly α-helical by circular dichroism. AFM-based single-molecule force spectroscopy revealed unfolding forces approaching 100 pN, much higher than typical α-helical domains (∼20 pN). All three retained substantial helical content to ≥100 °C. Mutating buried hydrophobic residues (V17S, L104R in HP149) reduced unfolding forces, confirming core packing as an important determinant. These results establish hydrophobic-core design as a promising route to robust α-helical scaffolds.