Optimizing PET hydrolase stability through an in silico evolution using a Genetic Algorithm

Proteins are highly sensitive to mutations, with even single point mutations affecting their structural stability. Protein engineering frequently employs mutations to improve stability; nevertheless, the vast sequence space makes exhaustive searches costly. To address this, we propose using an in silico genetic algorithm (GA) for a more efficient search. Here, we aim at enhancing the stability of a PET hydrolase using a GA to iteratively evolve the enzyme stability. The crystal structure of PETase from Ideonella sakaiensis (PDB: 6ANE) was used as the initial structure. Then, 3 distinct initial populations were generated: i. natural sequences from hydrolase family; ii. natural sequences + sequences generated by ProteinMPNN; and iii. natural sequences + PETases with random mutations. The Rosetta scoring function ref2015_cart was used as an objective function to estimate the ΔG for each individual. Then, the algorithm performs tournament selection, followed by crossover and mutation stages. The new sequences are concatenated to the previous population, with the best individuals forming the new population, and the cycle repeats iteratively. Residues within a 10 Angstrom radius of the active site were immobilized during the initial population generation, recombination, and mutation processes. In terms of ΔG, three replicas of 60 cycles of evolutions converged to similar solutions, for all conditions tested, with the  ΔG going from -600 Rosetta Energy Units (REU) to -880 REU. To evaluate the stability in higher temperatures of the best candidate for each condition and the WT, we ran a molecular dynamics (MD) simulation in GROMACS at 333.15K for 500ns. Although the  ΔG were close between the conditions, the sequence from the ProteinMPNN condition presented a more stable RSMD during the simulation. The RSMD distribution shows a high frequency in 2 Angstrom, while the other conditions have a higher frequency between 1 and 4 Angstrom. The RMSF for the ProteinMPNN condition was also lower than the others. We demonstrate the potential of a GA in enhancing protein stability by efficiently exploring the sequence space, even though further analysis is necessary to confirm the improvement in thermostability.