[P002] Poster

1:00 PM - 1:50 PM JST(4:00 AM - 4:50 AM UTC)

[P01-028] AI-Enabled Discovery and Biochemical Characterization of Novel GH5 Cellulases with Secretory Expression in Bacillus subtilis

○Seong-Jin Jeong¹, Min Yoo¹, Hyo-Deok Seo¹, Young Kyoung Rhee¹, Jae Woong Choi¹ (1. Korea Food Research Institute (Korea))

Keywords:

Cellulase,Protein language model,Catalytic efficiency,Secretory expression,Bacillus subtilis

Glycoside hydrolase family 5 (GH5) cellulases play a central role in lignocellulosic biomass saccharification, yet the discovery of industrially viable enzymes remains a major bottleneck due to the vast sequence space of natural diversity and the low throughput of conventional screening methods. Recent advances in protein language models (PLMs) offer a transformative opportunity to navigate this sequence space computationally; however, few studies have validated PLM-based predictions through end-to-end experimental pipelines spanning from gene selection to secretory production in an industrial host. Here, we present an integrated approach that bridges AI-driven enzyme discovery with functional deployment. Using CatPred, a PLM-based catalytic efficiency predictor, we computationally screened GH5 sequences from diverse microbial genomes and ranked candidates by predicted k_cat/K_m values. Twenty-three top-ranked genes were synthesized, cloned, and heterologously expressed in Escherichia coli Rosetta (DE3), and activity was evaluated by DNS assay with carboxymethyl cellulose (CMC). Four enzymes-EglS, AQSP_CL, MACA_CL, and THMA_CL-exhibiting significantly higher activity than the benchmark were selected for comprehensive biochemical characterization. THMA_CL, originating from the hyperthermophile Thermotoga maritima, displayed an optimal temperature of 80°C and retained over 85% residual activity after 120 min at 60°C, representing one of the most thermostable GH5 cellulases reported to date. AQSP_CL exhibited the highest catalytic efficiency (k_cat/K_m = 0.1504 s^-1·(mg/mL)^-1). Notably, Mn²⁺ strongly activated all four enzymes (261–325% of control), suggesting a conserved metal-dependent activation mechanism within this enzyme set. Structural analysis using BioEmu-generated conformational ensembles and molecular docking with AutoDock Vina elucidated substrate-binding geometries consistent with the observed kinetic parameters. To demonstrate industrial applicability, AQSP_CL was further expressed as a secretory protein in Bacillus subtilis, a GRAS-certified host widely used in industrial enzyme production. Signal peptide optimization yielded an approximately 8-fold increase in extracellular activity relative to cytoplasmic expression, confirming efficient secretory production suitable for downstream scale-up. By bridging computational prediction, experimental validation, and industrial expression within a single workflow, this study offers a scalable and generalizable paradigm for AI-assisted biocatalyst development-one that can be readily adapted to diverse enzyme families and host systems to meet growing demands for sustainable bioprocessing.