Although recent advances in deep learning approaches for protein engineering have enabled quick prediction of hot spot residues improving protein solubility, the predictions do not always correspond to an actual increase in solubility under experimental conditions. Therefore, developing methods that rapidly confirm the linkage between computational predictions and empirical results is essential to the success of improving protein solubility of target proteins. Here, we present a simple hybrid approach to computationally predict hot spots possibly improving protein solubility by sequence-based analysis and empirically explore valuable mutants using split GFP as a reporter system. Our approach, Consensus design Soluble Mutant Screening (ConsenSing), utilizes consensus sequence prediction to find hot spots for improvement of protein solubility and constructs a mutant library using Darwin assembly to cover all possible mutations in one pot but still keeps the library as compact as possible. This approach allowed us to identify multiple mutants of Escherichia coli lysine decarboxylase, LdcC, with substantial increases in soluble expression. Further investigation led us to pinpoint a single critical residue for the soluble expression of LdcC and unveiled its mechanism for such improvement. Our approach demonstrated that following a protein’s natural evolutionary path provides insights to improve protein solubility and/or increase protein expression by a single residue mutation, which can significantly change the profile of protein solubility.

No abstract