De novo carbon monoxide dehydrogenase and carbonic anhydrase using molecular dynamics and deep generative model

Abstract

Carbon monoxide dehydrogenase (CODH) and carbonic anhydrase (CA) play crucial roles in cellular metabolism by catalyzing the interconversion of carbon monoxide, carbon dioxide, and bicarbonate. However, the diversity of both enzymes remains unclear. This study integrates deep generative models and molecular dynamics simulations to streamline the design of novel CODH and CA variants. Using highly active enzymes from Carboxydothermus hydrogenoformans (PDB: 1SU8) and human carbonic anhydrase II (PDB:1HEB) as templates, we engineered de novo protein structures with enzymatic activities. Deep generative models including RFdiffusion, ProteinMPNN, CLEAN, and AlphaFold3 were employed to design novel CODH variants. Among all candidates, CODH2206 showed superior stability and activity in simulations but protein expressed as inclusion bodies in E. coli BL21(DE3) and improved in C43(DE3). Further characterization revealed that CODH2206 exhibited higher activity at pH 8. To resolve the quality and quantity of de novo enzymes, we applied SoDoPe solubility and trRosetta structure prediction for pixel-to-protein creation. Finally, hCAd3 activity increased 5-folds when chaperones and rare codons were involved in the system. This pipeline has high potential to generate diverse enzymes, advancing protein engineering for the creation of biocatalysts in the future.