Modeling Solvent Effects in Quantum Chemical Calculation of Relative Energies and NMR Chemical Shifts for...

This work presents an investigation into the effect of including explicit solvent molecules in DFT (Density Functional Theory) calculations of relative energies (ΔE_rel) and NMR (Nuclear Magnetic Resonance) chemical shifts for organic molecules in chloroform, water, and dimethyl sulfoxide (DMSO) solution using the PCM (Polarizable Continuum Model) approach. The large and flexible molecule of the antibiotic azithromycin (AZM) containing five OH groups and other polar centers susceptible to interacting with solvent molecules was used here as a working example. An increasing number of explicit solvent molecules was used in the geometry optimization of the supermolecule (n = 5, 15, 25, and 50), which accurately reproduces the first solvation shell. We optimized a large AZM trimer (3AZM-75CHCl₃) structure at the ωB97x-D/6-31G(d,p)-PCM level, containing 747 atoms, which may roughly simulate a 0.1 M dilute solution, with a good agreement with experimental NMR data. The supermolecule approach offers a robust description of solute-solvent intermolecular interactions, effectively accounting for both short- and long-range effects, making it a reliable method for selecting the predominant conformer in solution. While the effect of including explicit solvent molecules on the DFT calculation of ΔE_rel and ¹H NMR chemical shifts (OH protons) is remarkable, it is only moderate for the evaluation of ¹³C NMR spectra, providing support for the use of the continuum solvation model in this case. For highly solvated structures, a degree of arbitrariness in the calculation of relative energies is naturally introduced, mainly due to solvent-solvent interaction, causing a strong dependence of total energies on the initial guess structure used in the geometry optimization procedure, with variation in ΔE_rel around 70 kcal mol^-1 being predicted, and, therefore, it may not be quite suitable as a criterion to find the predominant conformer in solution. This does not happen with the DFT calculation of ¹H NMR chemical shifts (RMSD variations less than 0.1 ppm were observed for distinct initial guess structures), which are more strongly influenced by the local chemical environment. This is an interesting result regarding the use of an explicit solvent model in DFT calculations for organic molecules.