SDPD-SX: combining a single crystal X-ray diffraction setup with advanced powder data structure determination for use in early stage drug discovery

We report a method for routine crystal structure determination on very small (typically 0.1 mg or less) amounts of crystalline material using powder X-ray diffraction data from a laboratory-based single-crystal diffractometer. The solved structures span a wide range of molecular and crystallographic complexity.


Introduction
Crystal structures provide not only an accurate description of molecular connectivity and conformation, but also a basis for the understanding of physical properties such as solubility.In the pharmaceutical industry, three-dimensional information obtained from the solid state is useful at all stages of drug development, from hit identification through to lead optimisation, and beyond into formulation.Whilst single crystal X-ray (SX) diffraction is the "gold standard" method of characterisation, in many cases it is not possible to easily grow single crystals large enough to permit SX experiments.In such cases, and providing the material is polycrystalline, structure determination from powder X-ray diffraction (SDPD) data is a powerful alterative.However, most conventional lab-based powder X-ray diffraction (PXRD) setups require ca. 10 mg of material for a good quality diffraction pattern to be collected; such amounts are simply not available in the very early stages of drug development.This is unfortunate, because 3Dstructural information has significant potential to inform design (especially conformational design) earlier in the drug discovery process.][10] Collecting PXRD data using SX instrumentation is well-established 11 and has found applications in phase identification, QPA and structure refinement, 12 but reports of its use for structure determination are extremely rare. 13The focus in this study is to demonstrate a method for crystal structure determination from powder diffraction data 14,15 using very small amounts of polycrystalline material and a laboratory-based SX diffractometer for data collection.The broad applicability of this approach (henceforth referred to as SDPD-SX) is demonstrated with the crystal structure determination of 14 known compounds of pharmaceutical interest, chosen to represent the chemical and crystallographic complexity of real-world drug-like molecules.

Methodology
The 14 previously published crystal structures listed in Table 1 were first validated by periodic dispersion-correct DFT (DFT-D) calculations, following the approach of van de Streek. 16,17owder X-ray diffraction data were collected on the University of Manchester's Rigaku FR-X laboratory single crystal X-ray diffractometer using CuKα (λ = 1.5418Å) radiation.‡ A submilligram (typically <0.1 mg) amount of polycrystalline sample was mounted on a 100 μm glass fibre and secured with a minimum possible amount of Fomblin® YR-1800 oil (Fig. 1).Five 300°ϕ scans (beam divergence 1.0 mrad, detector distance 150 mm, 300 s per frame, range 1.8-60°2θ) were collected at room temperature using Rigaku's CrysalisPro 18 software and diffraction rings integrated using its built-in routines.For comparison purposes, PXRD data were collected from the same materials in transmission mode on a Bruker D8 Advance PXRD diffractometer using CuKα 1 radiation.§ Samples were contained in a 0.7 mm borosilicate glass capillary.
This journal is © The Royal Society of Chemistry 2022 All powder indexing and crystal structure solution attempts were carried out using the DASH software, 19,20 utilising the optimised simulated annealing parameter settings and recommended number of runs/simulated annealing moves reported by Kabova. 21 The best (i.e., lowest profile χ 2 ; see Fig. S1-S14 †) crystal structures resulting from each of the DASH runs were compared with the known single-crystal structures taken from the Cambridge Structural Database using the "Crystal Packing Similarity" feature of Mercury. 22

Results and discussion
Of the 15 materials studied, 14 (Table 1) were solved to a high degree of accuracy using DASH.Only γ-carbamazepine (P1 ¯, 28 DoF, Z′ = 4, N at = 120), which could be solved from capillary data, could not be solved using SDPD-SX; the low space group symmetry and large asymmetric unit led to a degree of reflection overlap that precluded stable Pawley fitting in DASH.
Fig. 2 shows that whilst the SDPD-SX data are in generally good agreement with the capillary PXRD data, they do not exhibit as good instrumental resolution.This, coupled with the CuKα 2 contribution leads to a higher degree of reflection overlap, limiting the accuracy with which individual reflection intensity information can be extracted.Despite this, the crystal structures obtained using DASH are in predominantly very good agreement with their known SX counterparts; DFT-D optimisation confirms this high level of accuracy in the solved structures (see Fig. S15-S28 †).The relatively low real-space resolution of the SDPD-SX datasets is therefore not a serious impediment to structure solution using a global optimisation approach such as the one implemented in DASH and subsequent Rietveld refinement is straightforward (see Fig. S29-S31 †) with results in good agreement with those obtained from   capillary data.The solved structures span a wide range of molecular and crystallographic complexity and include highly flexible molecules such as ritonavir (Fig. 3), demonstrating a broad range of applicability of SDPD-SX for synthetic and structural chemists.Additionally, sample preparation does not require light grinding and, in general, preferred orientation is not a significant issue.Some spottiness was observed in the diffraction rings of a few samples, suggesting the presence of larger crystallites in the sample, but this did not hamper structure determination.We have also obtained comparable success (results not shown) with a Rigaku Synergy diffractometer equipped with a sealed microfocus Cu source (0.1 mrad divergence) and a Hypix 6000HE single-photon counting detector.

Conclusions
The SDPD-SX approach outlined here shows the capability to determine crystal structures from sub-milligram amounts using powder X-ray diffraction on a single crystal X-ray diffractometer.The method can be justifiably classified as routine, in that it uses standard instrumentation and software at all stages of the process.There is no doubt that data collection in transmission capillary mode on a dedicated powder diffractometer (be it laboratory-based or central facility based) is the preferred option for SDPD from polycrystalline materials.However, we envisage that the SDPD-SX approach will be of significant value to those who only have access to sub-milligram amounts of powder that are insufficient to fill a capillary or whose loss (e.g. by radiation damage on a synchrotron beamline) cannot be risked.Furthermore, the wide availability of appropriate SX instrumentation provides a very useful route to structure determination for those who do not have easy access to suitable, dedicated PXRD instrumentation.Whilst it is unlikely that it will give sufficiently good data for the successful application of direct-methods based structure solution (at least, for crystal structures of the complexity used herein), any global-optimisation based SDPD method should prove effective.

Fig. 1
Fig.1A very small amount of powder (ca.0.01 mg) mounted on the top of a glass fibre in the single crystal X-ray diffractometer.Each major tick represents 0.1 mm.

Fig. 2
Fig.2Capillary PXRD data (black) for cefadroxil monohydrate overlaid upon PXRD data obtained from the equivalent SDPD-SX experiment (red).Note that the y-axis has been arbitrarily scaled to facilitate overlay and does not represent raw counts.

Table 1
Crystallographic details of the structures used in this study N at = number of atoms in asymmetric unit; DoF = total degrees of freedom in DASH run; d = resolution of solved structure; RMSD = crystal packing similarity value of DASH solution with CSD structure, and DFT-optimised DASH solution with the DFToptimised CSD structure.N/A: the water molecule was not accurately located, precluding DFT optimisation.