CRYSTAL GROWTH & DESIGN
Can the Formation of Pharmaceutical Cocrystals Be Computationally Predicted? I. Comparison of Lattice Energies
2009 VOL. 9, NO. 1 442–453
Nizar Issa,† Panagiotis G. Karamertzanis,‡ Gareth W. A. Welch,† and Sarah L. Price*,† Department of Chemistry, UniVersity College London, 20 Gordon Street, London WC1H 0AJ, U.K., and Centre for Process Systems Engineering, Department of Chemical Engineering, Imperial College, London SW7 2AZ, U.K. ReceiVed June 27, 2008; ReVised Manuscript ReceiVed September 19, 2008
ABSTRACT: A cocrystal is only expected to form if it is thermodynamically more stable than the crystals of its components. To test whether this can be predicted with a current computational methodology, we compare the lattice energies of 12 cocrystals of 4-aminobenzoic acid, 8 of succinic acid and 6 of caffeine, with the sums of the lattice energies of their components. These three molecules were chosen for their potential use in pharmaceutical cocrystals and because they had sufficient determinations of cocrystals and corresponding partner crystal structures in the Cambridge Structural Database. The lattice energies were evaluated using anisotropic intermolecular atom-atom potentials, with the electrostatic model and the intramolecular energy penalty for changes in specified torsion angles derived from ab initio calculations on the isolated molecules. The majority of the cocrystals are calculated to be more stable than their components, but the energy difference is only large in a few of the cases where the partner molecule cannot hydrogen bond to itself. More typically, the cocrystal stabilization is comparable to polymorphic energy differences and some of the specifically identified errors in the computational modeling. The cocrystals will be more stable relative to the observed disordered structures of caffeine and the kinetically preferred polymorph of 4-aminobenzoic acid, highlighting kinetic factors that may be involved in cocrystal formation. Overall, it appears that cocrystal formation should generally be predictable by comparing the relative stability of the most stable cocrystal and its pure components found on the computed crystal energy landscapes, but this is often very demanding of the accuracy of the method used to calculate the crystal energy. Introduction It has been established that cocrystals of an active pharmaceutical ingredient (API) can have advantageous properties,1–15 provided that the coformer is pharmaceutically acceptable. This has led to considerable interest in determining which cocrystals can be formed. Experimentally searching for cocrystals is not trivial: when there is a cocrystal on the phase diagram, it will not be readily formed by solution crystallization methods if the relative solubility of the two components leads to one crystallizing out first.16 This has led to development of various other methods of screening for cocrystals, such as solvent assisted grinding,17–19 and use of the experimentally determined phase diagram for the system.16 Thus, the labor of cocrystal screening would be considerably assisted by a reliable computational prediction of whether a cocrystal should be thermodynamically favorable relative to its components. Recently, there have been considerable advances in the ability to predict crystal structures from the chemical diagram, as evidenced by the success in the most recent20,21 international blind test organized by the Cambridge Crystallographic Data Centre, which for the first time included a cocrystal. There have been other successes in the prediction of the structure of multicomponent systems,22 such as monohydrates,23 diastereomeric salts,24,25 and other structures where there is more than one molecular entity in the unit cell, despite the increased challenge in searching for the relative orientation of the two molecules. Hence, in principle, it is possible to predict the structure and energy of the most stable cocrystal of two given molecules in a fixed stoichiometry, and we are developing a suitable methodology26 in light of this study. However, this cocrystal is only likely to be formed if it is more stable than * Corresponding author. E-mail:
[email protected]. † University College London. ‡ Imperial College.
the sum of its component crystal structures. Thus, the first step toward a computational method of predicting cocrystal formation is to check whether a current method of evaluating crystal energies consistently predicts the cocrystal as more stable than its component crystals. This paper assesses whether the lattice energy of 26 known cocrystals is more stabilizing than the sum of the lattice energies of the crystal structures of their two components, using a computational methodology27,28 that could be feasibly applied in a search so as to provide a genuine prediction. The lattice energies are computed using an anisotropic atom-atom model intermolecular potential, based on the ab initio charge distribution of the isolated molecules, plus ab initio evaluation of the energy penalty for changing flexible torsion angles in response to the crystal packing forces. The cocrystals studied have at least one component that can be used as a coformer in pharmaceutical cocrystals. Method Coformer and Cocrystal Selection. The choice of coformers studied was limited to those approved by the FDA as safe for internal administration, which were also computationally feasible, i.e., had a limited flexibility and contained only carbon, hydrogen, oxygen, and nitrogen in common functional groups. The Cambridge Structural Database (CSD)29 (September 2007) was searched for cocrystals of these coformers that also obeyed these computational restrictions and where there was a crystal structure of the partner molecule also in the CSD. We required that crystal structures of both the cocrystals and at least one polymorph of the component molecules were reasonably well determined (R factor