A Comparison of Methods for Computing Relative Anhydrous–Hydrate Stability with Molecular Simulation

The transformation of a pharmaceutical solid from an anhydrous crystal into a hydrated form during drug development represents a risk to a product’s safety and efficacy due to the potential impact on stability, bioavailability, and manufacturability. In this work, we examine 10 classical free energy simulation protocols to evaluate the thermodynamic stability of hydrated crystals relative to their anhydrous forms. Molecular dynamics simulations are used to compute the Gibbs free energies of the crystals of three pharmaceutically relevant systems using two fixed-charge potentials, GAFF and OPLS, as well as the polarizable AMOEBA model. In addition, we explore a variety of water models, including TIP3P, TIP4P, and AMOEBA, for both the interstitial water and the effects of ambient humidity. The AMOEBA model predicts free energy values most consistent with experimental measurements among the models examined. The benefits of a fully polarizable water model relative to fixed-charged models appear to derive predominantly from a better treatment of water’s dipole moment in the crystalline phase. Despite this improved physical treatment, we find that no single model produces reliable predictions of the phase boundary between hydrated and anhydrous crystals from theory alone. However, we show that accurate phase diagrams can be constructed from the simulations by introducing a single experimentally determined coexistence point. With this single experimental data point as input, the phase boundary is correctly predicted within 10% relative humidity on the temperature range of 15 to 75 °C for all three systems examined. Furthermore, we demonstrate that with this known coexistence point as an input, the differences between the various potentials and the water models become insignificant, as all models yield accurate phase boundaries regardless of whether polarization is included due to significant temperature-dependent error cancellation between models.