Taming Tautomerism in Organic Crystal Structure Prediction

Tautomerism influences the solubility, stability, efficacy, and even toxicity of drug formulations. Computational solid-form screening is increasingly being used to help identify effective drug formulations and de-risk against undesirable crystal forms. Here, however, we demonstrate that widely-used density functionals predict tautomeric crystal polymorph energetics poorly, largely due to density-driven delocalization error. We first show how these DFT models incorrectly identify the preferred tautomeric polymorphs of 2-thiobarituric acid and the antihelmintic drug mebendazole, and that some functionals substantially underestimate the barrier to solid-state tautomerization. Surveying 19 additional examples, we find that tautomeric polymorph energy differences defy conventional wisdom by regularly and substantially exceeding the 10 kJ/mol energy window associated with crystal polymorphism. Moreover, the DFT errors in the polymorph energy differences are strikingly large: they can reach tens of kJ/mol, and they produce the wrong qualitative stability ordering in ~20-30% of the cases. We address the challenge of predicting tautomeric polymorphs through a general solution that combines periodic hybrid DFT with intramolecular coupled-cluster theory corrections to substantially improve agreement with experiment. Most notably, our refined models reveal that the experimentally reported crystal structure of mebendazole form B features the incorrect tautomer--a prediction that is confirmed via solid-state NMR analysis.