Structure determination of an amorphous drug through large-scale NMR predictions

Abstract

Knowledge of the structure of amorphous solids can direct, for example, the optimization of pharmaceutical formulations, but atomic-level structure determination in amorphous molecular solids has so far not been possible. Solid-state nuclear magnetic resonance (NMR) is among the most popular methods to characterize amorphous materials, and molecular dynamics (MD) simulations can help describe the structure of disordered materials. However, directly relating MD to NMR experiments in molecular solids has been out of reach until now because of the large size of these simulations. Here, using a machine learning model of chemical shifts, we determine the atomic-level structure of the hydrated amorphous drug AZD5718 by combining dynamic nuclear polarization-enhanced solid-state NMR experiments with predicted chemical shifts for MD simulations of large systems. From these amorphous structures we then identify H-bonding motifs and relate them to local intermolecular complex formation energies.

Fig. 1. Solid-state NMR experiments.

a ¹H, b ¹³C and c ¹⁵N MAS NMR spectra of crystalline (blue) and amorphous (orange) 1. d ¹³C-¹³C DNP enhanced solvent suppressed INADEQUATE and e, ¹H-¹³C HETCOR spectra of crystalline 1. The dashed black line in (a) indicates the chemical shift assigned to the proton bound to N6 in the amorphous sample. In (d), the ¹³C peaks denoted by a star at 60 and 170 ppm are attributed to impurities introduced during the NMR sample preparation. The chemical structure and labelling scheme of 1 is shown in (e).

Fig. 2. NMR crystallography of AZD5718.

a ¹H and b ¹³C chemical shift RMSEs of the ten lowest DFT-D energy CSP candidates and of the crystal structure determined by X-ray diffraction (labelled XRD). The label rank in terms of increasing force-field energy per molecule of (1) of each candidate is indicated above each point. c Two-dimensional projection of the similarity of the computed ¹H and ¹³C chemical shifts of the candidate structures to the experimental data (red cross). The probability of each candidate to match experiment is represented by the area of the blue disk. p(M) denotes the probability that a virtual candidate, which represents structures potentially missing from the CSP candidate pool, matches experiment. The virtual candidate is represented by the mean of the shifts of all CSP candidates, and a high probability p(M) suggests that the CSP candidate pool may not include the experimentally observed structure. d Chemical structures of the two tautomers of 1 considered, labelled as A and B. e Agreement between ¹H, ¹³C and ¹⁵N experimental and DFT computed chemical shifts for the two tautomers. f ORTEP plot of the ADP tensors for the NMR structure of 1 drawn at the 90% probability level.

Fig. 3. Predicted spectra.

Predicted and experimental ¹H NMR spectra of crystalline (blue) and amorphous (orange) 1. The predicted spectrum of amorphous 1 was obtained by considering only the 4% w/w water MD simulations.

Fig. 4. H-bonding motifs in amorphous AZD5718.

a Predicted spectra obtained using the predicted ¹H chemical shifts of the most often occurring N-H bonding motifs involving N6 for 11 evenly spaced snapshots of all amorphous simulations of each water content. The percentages next to the spectra denote the fraction of bonding motifs their corresponding pattern represents, including the instances where no H-bonded neighbour was identified. The dashed vertical line indicates the experimental shift observed in amorphous 1 and assigned to the proton bound to N6. b Hydrogen bonding motifs associated with the spectra in (a). c Number of occurrences of extended H-bonding motifs yielding a predicted chemical shift above 11 ppm for every snapshot of the 4% water simulations. Only the patterns corresponding to the top 75% of all shifts above 11 ppm were selected. The orange bars represent the bonding motifs involving water, and the blue ones correspond to the motifs that do not involve water. Two secondary neighbours from the same molecule are indicated by an asterisk. In (b) and (c), On indicates the oxygen atom bonded to carbon n.

Fig. 5. Complete structures and H-bonding motifs.

a Superposition of 10 instances of the N6-H···O20 bonding motif. b close-up view of the hydrogen bonding region in (a). c Superposition of 10 instances of the N6-H···OH₂···OH₂/O20 bonding motif. d close-up view of the hydrogen bonding region in (c). The red molecule represents 1 bearing the hydrogen bond donor (N6-H), the dark blue molecule represents 1 bearing the hydrogen bond acceptor, water molecules are coloured in cyan and the atoms of 1 involved in the hydrogen bonding motif are coloured in green.

Fig. 6. Relative formation energy of the hydrogen bonding motifs.

Mean computed formation energies of intermolecular complexes for the H-bond acceptor connected to a N6-H, b N21-H or c N28-H of the probe molecule. The percentage under each bar indicates the fraction of the N-H group bonded to the corresponding H-bond acceptor. Only the H-bond acceptors making up at least 1% of all instances analysed are displayed. The error bars indicate the standard error of the mean of the relative formation energies.