Abstract
AI-driven biologics manufacturing demands an efficient protein production platform. In this study, we optimized scFv secretion in Pichia pastoris through three strategies: gene dosage optimization, expression cassette design, and endoplasmic reticulum (ER) secretory pathway reprogramming. Using two structurally homologous scFv variants-PR961 and PR953-with divergent basal secretion levels (12.35:1 ratio), we demonstrate that protein-specific thresholds govern optimization efficacy. While increasing gene copy numbers yielded limited improvements (PR961: 1.25-fold at four copies; PR953: 2.37-fold at six copies), reconfiguring the expression cassette to a V(H)-linker-V(L) orientation significantly enhanced secretion (11.18-fold for PR961; 5.09-fold for PR953). Twenty-one genes in three functional modules of the secretory pathway were knocked out or overexpressed. The pathway reprogramming results revealed distinct regulatory dependencies: PR961 secretion relied on ER-to-Golgi trafficking (SEC23 overexpression: 1.20-fold), whereas PR953 depended more on upstream translocation (SEC62: 1.66-fold) and oxidative folding (ERO1: 1.81-fold) enhancements. Notably, both variants exhibited a glycosylation-dependent regulation through CNE1. Our findings challenge the assumption that structural homology (63% amino acid identity; RMSD 0.47 Å) ensures consistent optimization outcomes, highlighting the imperative for protein-tailored engineering strategies in synthetic biology.