Protein language model-guided generative design of affinity peptides for chromatographic purification of lentiviral vectors.

Lentiviral vectors (LVs) have emerged as the most promising tool for cell and gene therapy. However, LVs are very fragile and sensitive to shear stress, buffer pH and salt concentration, resulting in serious hurdles in downstream processing. To address this challenge, mild and specific downstream processing technologies are desired for LVs purification. Aiming at the development of LVs purification by affinity chromatography, we herein proposed an artificial intelligence (AI)-based strategy for finding affinity peptide ligands. The strategy combined the use of protein language model ESM3-sm-open-v1 with in silico screening. Based on a combined peptide library generated via the fine-tuned ESM model, we identified eight peptides targeting the conserved cysteine-rich domain-binding region of the low-density lipoprotein receptor on vesicular stomatitis virus G glycoprotein (VSV-G). Among them, VSV-G-P7 exhibited the highest affinity based on molecular docking and ELISA. Molecular dynamics simulations revealed that the affinity binding of the peptide to VSV-G was driven by multiple factors but dominated by electrostatic interactions. By coupling the peptide to Sepharose 4FF (Sep4FF) gels, Sep4FF‒VP7 gels were synthesized for LVs purification. The chromatography experiments demonstrated that LVs could not only be selectively bound to Sep4FF‒VP7, but also be efficiently eluted by increasing salt concentration. The mild elution preserved the structural integrity and biological activity of the recovered LVs. In the LVs purification from VCCF, the affinity chromatography kept stable in 20 recycled operations, exhibiting steady LV recoveries (78.2%-79.1%), Vero HCP removal (90.6%-91.7%) and residual dsDNA removal (44.7%-47.3%). The results collectively proved the success of the AI-based ligand screening strategy as well as the potential of the affinity chromatography approach for LVs purification.