Tautomerism plays an important role in drug formulation, including solubility, stability, efficacy, and even toxicity. Crystal structure prediction is increasingly being used successfully as a complement to experimental solid-form screening in the pharmaceutical industry to de-risk formulations. However, we will show how the density-driven delocalization present in commonly-used density functionals can lead to incorrect tautomer energies and poor crystal energy landscapes that could serious impact efforts to de-risk the solid formulation of tautomeric species. A number of examples will be presented, ranging from the crystal structure prediction of 2-thiobarbituric acid, analysis of a sizable data set of tautomeric polymorphs from the Cambridge Structure Database, and the antihemlitic drug mebendazole. In mebendazole, for example, we will show how generalized gradient approximation functionals incorrectly predict the tautomeric state for two of the three polymorphs, dramatically underestimate the proton-transfer barriers associated with solid-state tautomeric conversion, and erroneously predict the crystals to contain mixtures of tautomers. Applying local energy corrections with higher-level methods addresses these problems. Importantly, our calculations reveal that the previously reported experimental crystal structure for one of these mebendazole polymorphs contains the wrong tautomer, as confirmed both by the computed energetics and careful analysis of solid-state nuclear magnetic resonance data.
