Generalized Born and Explicit Solvent Models for Free Energy Calculations in Organic Solvents: Cyclodextrin...

Evaluation of solvation (binding) free energies with implicit solvent models in different dielectric environments for biological simulations as well as high throughput ligand screening remain challenging endeavors. In order to address how well implicit solvent models approximate explicit ones we examined four generalized Born models (GB(Still), GB(HCT), GB(OBC)I, and GB(OBC)II) for determining the dimerization free energy (ΔG(0)) of β-cyclodextrin monomers in 17 implicit solvents with dielectric constants (D) ranging from 5 to 80 and compared the results to previous free energy calculations with explicit solvents ( Zhang et al. J. Phys. Chem. B 2012 , 116 , 12684 - 12693 ). The comparison indicates that neglecting the environmental dependence of Born radii appears acceptable for such calculations involving cyclodextrin and that the GB(Still) and GB(OBC)I models yield a reasonable estimation of ΔG(0), although the details of binding are quite different from explicit solvents. Large discrepancies between implicit and explicit solvent models occur in high-dielectric media with strong hydrogen bond (HB) interruption properties. ΔG(0) with the GB models is shown to correlate strongly to 2(D-1)/(2D+1) (R(2) ∼ 0.90) in line with the Onsager reaction field ( Onsager J. Am. Chem. Soc. 1936 , 58 , 1486 - 1493 ) but to be very sensitive to D (D < 10) as well. Both high-dielectric environments where hydrogen bonds are of interest and low-dielectric media such as protein binding pockets and membrane interiors therefore need to be considered with caution in GB-based calculations. Finally, a literature analysis of Gibbs energy of solvation of small molecules in organic liquids shows that the Onsager relation does not hold for real molecules since the correlation between ΔG(0) and 2(D-1)/(2D+1) is low for most solutes. Interestingly, explicit solvent calculations of the solvation free energy ( Zhang et al. J. Chem. Inf. Model . 2015 , 55 , 1192 - 1201 ) reproduce the weak experimental correlations with 2(D-1)/(2D+1) very well.