Leveraging protein language model for maturation of high-affinity peptide ligand in the purification of an adeno‒associated virus vector.

Computational design has emerged as an important approach for the discovery of peptide ligands in development of affinity chromatography. However, fast and efficient identification of high-affinity and specific ligands is still challenging. Here an artificial intelligence (AI)-based strategy for affinity maturation was proposed for improving the binding affinity of peptide ligands of adeno‒associated virus (AAV). To target AAV receptor-binding region on AAV2 capsid, a de novo design method was applied in the development of peptide ligands and three candidate peptides were identified for affinity maturation. We then introduced a multimodal protein language model, ESM3, to guide peptides generation for the maturation of the binding affinity. After three rounds of docking simulations, a high-affinity peptide A4 with a dissociation constant of 3.8 μmol/L was obtained. Molecular dynamics simulation revealed that the binding of A4 was dominated by electrostatic interactions. By coupling A4 onto Sepharose 4 Fast Flow (Sep4FF) gels, Sep4FF‒A4 gels were synthesized and exhibited good selectivity and serotype specificity to AAV2. AAV2 was adsorbed effectively at pH6.0-8.0 and eluted mildly from the affinity column at 500 mmol/L NaCl. Finally, Sep4FF‒A4 chromatography was successfully used to purify AAV2 with a high transduction efficiency from HEK293 cell culture fluids, and it kept stable with AAV2 yields of 67.4-72.7%, DNA clearance ranging from 47.6-58.7% and HCP clearance ranging from 86.3-91.3% in 20 cycles.