Adapted DFTB3 repulsive potentials reach DFT accuracy for hydride transfer reactions in enzymes

Enzymatic hydride transfer reactions play a crucial role in numerous metabolic pathways, yet their accurate computational modeling remains challenging due to the trade-off between accuracy and computational efficiency. Ideally, molecular dynamics simulations should sample all enzyme configurations along the reaction path using post Hartree-Fock or DFT QM/MM electrostatic embedding methods, but these are computationally expensive. Here, we introduce a simple approach to improve the third-order density functional tight binding (DFTB3) semi-empirical method to model hydride transfer reactions in enzymes. We identified deficiencies in DFTB3's description of the potential energy surface for the hydride transfer step in Crotonyl-CoA Carboxylase/Reductase (Ccr) and developed a systematic methodology to address these limitations. Our approach involves modifying DFTB3's repulsive potential functions using linear combinations of harmonic functions, guided by analysis of C-H and C-C distance distributions along the reaction path. The optimized DFTB3 Hamiltonian significantly improved the description of the hydride transfer reaction in Ccr, reproducing the reference DFT activation barrier within 0.1 kcal/mol. We also addressed the transferability of our method by applying it to another hydride transfer reaction bearing the 1,4-dihydropyridine motif but exhibiting distinct structural features of the reactant, as well as the hydride transfer reaction in Dihydrofolate Reductase (DHFR). In both cases our adapted DFTB3 Hamiltonian correctly reproduced the DFT reference and experimentally observed activation barriers. The low computational cost and transferability of our method will enable more accurate and efficient QM/MM molecular dynamics simulations of hydride transfer reactions, potentially accelerating research in enzyme engineering and drug design.