Taming Disorder in Pharmaceutical Solids

The higher bioavailability of the amorphous form of pharmaceutical materials make them an
attractive goal for drug formulation. This bioavailability is associated with greater solubility in
the amorphous form, which in itself is a product of the higher free energy of these disordered
molecules. The inherent downside to this free energy is a tendency towards crystallisation into
less bioavailable forms, reducing the efficacy of the drug formulation over time. It is therefore
important that these materials can be reliably characterised and the factors affecting the
stability in the amorphous forms be determined.
This research project establishes a joint computational and experimental workflow whereby
amorphous models are constructed using molecular dynamics simulation. The power of ma-
chine learning is then leveraged to produce rapid chemical shift predictions of the atomic sites
within these amorphous structures, which would be computationally prohibitive using con-
ventional means of chemical shift calculation using DFT. Qualitative comparison between the
simulated machine learning derived spectra with experimental spectra obtained via solid-state
NMR can be made. The abundant information present in the trajectories of these amorphous
model systems can be used to explain key features associated with the amorphous form that
are present in the NMR spectra, and differentiate it from the crystalline form. Such features
include the relationship between tautomerisation in irbesartan and the torsion angle of the
tetrazole ring, which manifests as a splitting of the C23 peak in the amorphous 13C spectrum.
Also probed is the hydrogen bonding associated with the tetrazole 1H, the relationship to
conformation and the 1H experimental NMR spectrum.
This workflow has been applied to a much larger PROTAC molecule AZ1, in which it was
able to differentiate molecular dynamics simulations run with different charge models, via a
qualitative comparison of the resultant machine learning predicted spectra.The difference in
spectra was once again associated with differing hydrogen bonding behaviour of the charge
models used in the simulation, obtained by probing the MD trajectories.
Lastly, parallel variable temperature experiments have also been performed using solid-
state NMR and molecular dynamics simulations in order to analyse the dynamics of amorphous
irbesartan, above and below Tg. Mobility in the form of bulk translation as well as local
reorientation have been probed. These experiments represent another tool in the repertoire
of experiments used to understand the nature of disorder in amorphous drug materials.