Atomic-level structure determination of amorphous molecular solids by NMR

Structure determination of amorphous materials remains challenging, owing to the disorder inherent to these materials. Nuclear magnetic resonance (NMR) powder crystallography is a powerful method to determine the structure of molecular solids, but disorder leads to a high degree of overlap between measured signals, and prevents the unambiguous identification of a single modeled periodic structure as representative of the whole material. Here, we determine the atomic-level ensemble structure of the amorphous form of the drug AZD4625 by combining solid-state NMR experiments with molecular dynamics (MD) simulations and machine-learned chemical shifts. By considering the combined shifts of all 1H and 13C atomic sites in the molecule, we determine the structure of the amorphous form by identifying an ensemble of local molecular environments that are in agreement with experiment. We then extract and analyze preferred conformations and intermolecular interactions in the amorphous sample in terms of the stabilization of the amorphous form of the drug.

NMR probes the local chemical environment, but interpreting the results is difficult. NMR crystallography employs chemical shift calculations on candidate structures, but that's hard to do for an amorphous system. To make it work, they have to sample large numbers of candidate environments via MD simulations, and then predict chemical shifts cheaply for ~8000 snapshots with their ShiftML machine learning model.

Versions of the individual techniques employed here have all been reported previously.
Rather, the current paper is impressive because of what they accomplish by combining all of these ideas. In addition to gaining insights into the experimental structure, I found the investigation of the similarities and differences between their NMR-derived ensemble and the energy-determined MD one fascinating. They also do a nice job of highlighting the types of conformations and environments of the drug molecule that occur in the system.  (2021)) as that only determined the hydrogen-bonding environments of the three N-H protons in a different amorphous drug. This paper shows how the machine learned chemical shifts can enable the determination of the conformation as well as hydrogen bonding in an amorphous drug. My main concern with this paper is that the general title "of amorphous molecular solids" is not supported by one unverified result on a simpler amorphous drug and so the title would be more accurate as "of an amorphous pharmaceutical solid". This should be published as a communication demonstrating the power of complex multiple NMR experiments combined with theory to elucidate the range of conformations, hydrogen bonding etc seen in an amorphous form of a pharmaceutical, a methodology which I expect will be refined by a wide range of studies. This paper is convincing as being performed on an Astra Zeneca molecule, rather than one chosen for its suitability for the computational modelling, and I expect that further examples refining this approach will follow. Most of my comments are suggestions for clarification to help the reader.
The result of the analysis is only not verified because the experiment that was probably intended to do this (as I cannot think of any other verification approach), the radial distribution functions derived from X-ray diffraction, were unable to discriminate between the ensembles of local molecular environments. This makes the achievement of this paper even more impressive. My initial impression of The methodology is critically dependent on the whether the machine learned chemical shifts and the MD simulations of the amorphous material fully sample all the conformational and hydrogen-bonding space that is available in the amorphous form. Note that the possibility of polyamorphism, in which the amorphous states differ in the dominant conformational region or type of hydrogen bonding present, suggests the experimentally as well as computationally, there may be molecule-dependent barriers to fully sampling all possible molecular environments. Hence the starting points for the 8 MD simulations of the amorphous solid, which are limited to 128 molecules. may be critical in determining the coverage of possible configurations. The SI suggests that all 8 amorphous simulations were all generated from one isolated molecule conformation, whereas the text (around Fig 3) suggests that the starting conformations had a dihedral angle between the aromatic planes of around +90 or -90. (Why does this explain the differences in height in Fig3d?) The SI must contain more details of how the amorphous MD simulations were produced, and why 8 were used. This will be very dependent on the molecule. ADZ4625 has only one hydrogen bond donor OH position, one very flexible dihedral (C15-C16-C18-C19 which gives both R and OH on both sides of the molecule in Fig A), some flexibility in 8 membered ring, boat/chair for 6 membered ring, and enone conformation. This makes for a good test system, and I expect that 8 MD simulations are easily sufficient for ADZ4625's flexibility, but this must be argued so the method can be adapted for other systems.  Understanding of the interactions made to stabilize a neat API could help lead to improved selection of polymers and other excipients to help stabilize an amorphous drug to prevent crystallization.
I think this work is of considerable novelty and addresses a problem long pondered by many in the field of amorphous solids, particularly in the pharmaceutical arena. The molecule itself is of structural relevance to the modern pharmaceutical industry, making the work even more relevant to current investigators. All those in the field will find this work of great interest, in my opinion. The data collected are of high quality and provide ample support for the claims made, supported by statistical analysis and controls, for example the randomly selected 1000 structures from the MD ensemble.
The manuscript is well written and support with sufficient data and references, and I fully support publication without further revision.