Revealing the early stages of carbamazepine crystallization by cryoTEM and 3D electron diffraction

A time-resolved crystallization study of early-stage carbamazepine crystallization by cryoTEM reveals two crystallization pathways, one classical and the other non-classical. Four forms of carbamazepine are also identified via 3D electron diffraction from a single crystallization.


ICP-MS
In addition to the development of CBZDH crystals in the samples imaged at cryogenic and RT conditions after 20 seconds of crystallization. Several very thin crystals with an unusual tear-drop morphology were also observed ( Figure S5). These crystals were shown by 3DED to have a cubic F lattice with unit cell dimension a = 4.21 Å, suggestive of a simple salt. ICP-MS was carried out as described below and reveals the presence of Mg 2+ as a trace contaminant (Table S6) and the unit cell dimensions are consistent with MgO.
Contaminants in EtOH were detected using ICP-MS using an Agilent 7500ce (with octopole reaction system), employing a radio frequency forward power of 1500 W and reflected power of 1 W, with argon gas flows of 0.8 L min −1 and 0.2 L min −1 for carrier and makeup flows, respectively. Using a peristaltic pump, sample solutions were taken up into the Mira mist nebuliser at a rate of 0.2 ml min −1 .
Skimmer and sample cones were made of nickel. The instrument was operated in 'spectrum multitune acquisition mode' and three replicate runs per sample were employed. Each isotope was integrated for 0.1s per point giving a total integrations time of 0.3 s per unit mass. Three replicate runs per sample were employed to give good counting statistics. To ensure good crossover between the potential switch from pulse counting (P) to analogue counting (A) mode, a P/A factor was determined across the entire mass range at the start of each analysis day. The switch occurs at about 20-30 μg L −1 under the current instrument condition. The isotopes: 7 Li, 9 Be, 11 B. 23 Na, 24

Notes on model refinement from electron data
Carbamazepine dihydrate (CBZDH) -Seven datasets were merged in XSCALE. (Kabsch, 2010) The structure was solved using SHELXT (Sheldrick, 2015a) and refined using the kinematical model of scattering in SHELXL (Sheldrick, 2015b). Molecular graphics were produced using XP (Sheldrick, 2008). The structure was modelled in space group P21/c with the amide and solvent water disordered over two orientations. Chemically equivalent bond distances and angles were restrained to be similar in the disordered region of the structure. H atoms attached to atoms C3 -C14 were treated as variable metric rigid groups in which the CH distance was allowed to refine. Amide H-atoms in the disordered region were placed in calculated positions. H-atoms attached to water were located in a difference map. The water molecules were treated as rigid groups with hydrogen bond distances (O1 ... H3B and O2 ... H4C) restrained to the distances in the neutron structure of CBZDH (CSD refcode FEFNOT08), 1.871 and 1.910 Å, respectively. All full-occupancy atoms were refined with anisotropic displacement parameters subject to enhanced rigid-body restraints (Thorn et al., 2012). Disordered atoms were refined isotropically. The final value of R1 was 16.31%.
Form III -A single dataset was indexed and integrated in PETS 2.0 (Palatinus et al., 2019). As this was the only crystal observed, merging of data sets was not possible. Solution and refinement procedures were similar to those described above for CBZDH. H atom distances were constrained to typical electron CH and NH distances from the structure determination of glycine from electron diffraction data (refcode KUFDOH, 1.168 and 1.184 Å, respectively). CH distances involving H7, H10 and H11 were allowed to refine freely in variable metric rigid groups. The final value of R1 was 19.68%.
Form IV -Two datasets were merged and processed in the same way as CBZDH. Starting atomic coordinates were taken from (Lang et al., 2002) (refcode CBMZPN12). Refinement procedures were similar to those described above. Most H atoms were allowed to refine freely as variable metric rigid Electron scattering factors were taken from Doyle and Turner (Doyle & Turner, 1968

Powder Diffraction
Powder diffraction data was collected on two samples. For the first, a sample was prepared by allowing a 3 ml drop of the saturated carbamazepine-ethanol solution (see Section 2 of the main text) to crystallize for three minutes directly on the powder sample holder. The samples were exposed to Cu Kα radiation (30 kV, 10 mA) in a wide-angle powder x-ray diffractometer (D2 Phaser, Bruker, USA). The instrument was operated in Bragg-Brentano geometry with steps in increments of 0.08° 2θ. The angular range was 5 -50° in 2θ and the counts were accumulated for 1 second at each step.
The total time for each data collection was ten minutes.
A second sample was prepared by dipping a MiTeGen mounting loop into the saturated carbamazepineethanol solution and then allowing to crystallize for 30 seconds before starting the data collection on a single crystal X-ray diffractometer (Rigaku Supernova, Oxford Diffraction, UK). The instrument was operated in powder measurement mode using CrysalisPro in the default settings (CrysAlis, 2009). The total time for this data collection was 15 minutes. The sample obtained in this way consisted of a small collection of crystallites and was not large enough to study on the powder diffractometer.
Carbamazepine dihydrate (CBZDH) has a lath-like habit where the dominant face is 020 (Kachrimanis & Griesser, 2012). The powder pattern obtained after 3 min could mostly be indexed and CBZDH with 0k0 reflections with even values of k from 4 to 14. The pattern could be successfully fitted using the Pawley method (Coelho, 2018)  The data obtained after 30 sec could also be modelled as CBZDH. The first peaks were observed at 2θ ~20°. Several peaks below this are missing indicating that texture effects are significant. A Pawley fit ( Figure S7) using cell dimensions fixed to their literature values (CSD refcode FEFNOT04) converged to Rwp = 1.07%. A TCHZ function was used for the peak-shape and a six-term Chebychev polynomial for the background. The profiles in the pattern are quite broad and the structures relatively complex and reasonable Pawley fits could also be obtained with other phases. Notably a fit IUCrJ (2022). 9, https://doi.org/10.1107/S2052252521010101 Supporting information, sup-5 to CBZ-I which is even more complex than CBZDH, being triclinic with Z = 8, converged to Rwp = 1.10%. While characterisation of the 30 sec sample by this method can thus not be regarded as definitive, these findings do illustrate the power of electron diffraction for phase identification of small samples.   (Lisgarten et al., 1989).  Mass of 1 L of wet ethanol = 789.45 g (assuming the density is the same as pure ethanol (0.78945 g cm −3 )) Mass of water in 1 L = 219 x 10 −3 g (average value from runs 1 -3) % water = 100 x (219 x 10 −3 ∕ 789.45) = 0.028 % water by mass.

Supporting Figures
Figure S1 Schematic of sample preparation procedure used in this work.     (Table S6) and the unit cell dimensions are consistent with MgO. All scale bars = 2 μm.

Figure S6
Pawley fit of data obtained after 3 min crystallisation using parameters for CBZDH.

Figure S7
Pawley fit of data obtained after 30 s crystallisation using parameters for CBZDH.