CB26 - Improving the Solubility and Activity of Porcine ST3Gal-I with ProteinMPN
SCURS Disciplines
Biochemistry
Document Type
General Poster
Invited Presentation Choice
Not Applicable
Abstract
The heterologous expression of mammalian glycosyltransferases in Escherichia coli is notoriously challenging, frequently resulting in insolubility and low yields despite the use of solubility tags or engineered host strains. This bottleneck has significantly restricted their utility in the chemoenzymatic synthesis of glycoconjugates. In this study, we leveraged the deep learning–based design framework, ProteinMPNN, to engineer porcine ST3Gal1 for enhanced expression in E. coli. Structural constraints were applied to residues within 10 Å of the catalytic center, while approximately one-third of the remaining sequence space was redesigned.
Of the ten top-scoring variants selected for experimental validation, nine demonstrated markedly improved solubility (4.3–36.2%), with four achieving expression yields of ~10 mg/L. Activity assays using Gal-β1,3-GalNAc-α-Thr-Fmoc as an acceptor substrate confirmed significant catalytic activity in five mutants. Notably, Mutant-9 exhibited a 1.5-fold increase in activity compared to wild-type pST3Gal1. To demonstrate practical utility, Mutant-9 was employed in a gram-scale synthesis, converting 1.1 g of Core1-O-glycan with a 94% yield within 2 hours using only 10 mg of purified enzyme. These findings establish the feasibility of generating soluble, active mammalian glycosyltransferases in E. coli and highlight the transformative potential of AI-assisted protein engineering for chemoenzymatic applications.
Keywords
Mammalian Glycosyltransferases, ProteinMPNN, Deep Learning Protein Design, Improved Solubility, Increased Expression Yield
Start Date
10-4-2026 9:30 AM
Location
University Readiness Center Greatroom
End Date
10-4-2026 11:30 AM
CB26 - Improving the Solubility and Activity of Porcine ST3Gal-I with ProteinMPN
University Readiness Center Greatroom
The heterologous expression of mammalian glycosyltransferases in Escherichia coli is notoriously challenging, frequently resulting in insolubility and low yields despite the use of solubility tags or engineered host strains. This bottleneck has significantly restricted their utility in the chemoenzymatic synthesis of glycoconjugates. In this study, we leveraged the deep learning–based design framework, ProteinMPNN, to engineer porcine ST3Gal1 for enhanced expression in E. coli. Structural constraints were applied to residues within 10 Å of the catalytic center, while approximately one-third of the remaining sequence space was redesigned.
Of the ten top-scoring variants selected for experimental validation, nine demonstrated markedly improved solubility (4.3–36.2%), with four achieving expression yields of ~10 mg/L. Activity assays using Gal-β1,3-GalNAc-α-Thr-Fmoc as an acceptor substrate confirmed significant catalytic activity in five mutants. Notably, Mutant-9 exhibited a 1.5-fold increase in activity compared to wild-type pST3Gal1. To demonstrate practical utility, Mutant-9 was employed in a gram-scale synthesis, converting 1.1 g of Core1-O-glycan with a 94% yield within 2 hours using only 10 mg of purified enzyme. These findings establish the feasibility of generating soluble, active mammalian glycosyltransferases in E. coli and highlight the transformative potential of AI-assisted protein engineering for chemoenzymatic applications.