Predicting Polymorphism

The Problem
Drug discovery is a phrase used to describe the early stages of the extended process that eventually leads to new cures for disease. At this stage of the enquiry the emphasis is on understanding the protein(s) implicated in the disease expression and in identifying candidate drug structures which could, according to their molecular structure, be possible candidates for arresting the disease process. In practice, molecular structure is not the only feature of importance. Also of interest is the optical activity of candidate molecules (left hand-right hand) and the particular way in which they organise into a crystal lattice. This latter defines the polymorphic form of the crystalline solid. A variety of polymorphic forms for a given molecule are common. The different forms can have widely differing properties such as solubility which have a bearing on the drug’s activity in practice. An ability to predict polymorphs and their properties is therefore of great interest to the pharmaceutical industry as an early indication could eliminate years of development and clinical effort at later stages of the overall process of bringing a drug to market.

The Challenge
AstraZeneca wished to study certain model compounds to develop a validated simulation approach for the prediction of polymorphic forms. This involved the incorporation of atom-atom potentials for describing molecular recognition phenomena including crystallisation. The molecular structures of choice included azoaromatic chlorides such as the fungicide chlorothalonil and theophylline a muscle relaxant (shown).

theophylline showing N-H...N hydrogen bond

The Solution
The Daresbury simulation code DMAREL was chosen to carry out the required molecular modelling and was extended to include atom-atom repulsion potentials. Further extension to minimise under the constraint of space group symmetry and to recognise when this results in a transition state and the symmetry reduction that is required was also implemented successfully.

The interface program NEIGHBOURS, which links DMAREL to a variety of crystallographic databases, was extended to improve automation. These enhanced codes were successfully applied to predict polymorphic forms of theophylline, benzamide (two structures) o-acetamidobenzamide (two structures) and chlorothalonil (two structures). In the case of theophylline a polymorphic form with N-H…O hydrogen bonds was eliminated as a possible structure leaving only the N-H…N structure shown.

The Benefits
• The customer requirement for a proven modelling code for polymorph prediction was met.
• The code was demonstrated to predict polymorphs correctly by comparison with experiment and even predicted a previously unrecognised structure as a component of a powder diffraction result.
• The customer has also used the code to model related phenomena such as protein-drug interactions.