Solving molecular crystal structures from X-ray powder diffraction data: The challenges posed by γ-carbamazepine and chlorothiazide N,N,-dimethylformamide (1/2) solvate

Authors

  • Philippe Fernandes,

    1. Solid-State Research Group, Strathclyde Institute of Pharmacy and Biomedical Sciences, University of Strathclyde, 27 Taylor Street, Glasgow G4 0NR, UK
    Search for more papers by this author
  • Kenneth Shankland,

    1. CCLRC ISIS Facility, Rutherford Appleton Laboratory, Chilton, Didcot, Oxon OX11 0QX, UK
    Search for more papers by this author
  • Alastair J. Florence,

    1. Solid-State Research Group, Strathclyde Institute of Pharmacy and Biomedical Sciences, University of Strathclyde, 27 Taylor Street, Glasgow G4 0NR, UK
    Search for more papers by this author
  • Norman Shankland,

    Corresponding author
    1. Solid-State Research Group, Strathclyde Institute of Pharmacy and Biomedical Sciences, University of Strathclyde, 27 Taylor Street, Glasgow G4 0NR, UK
    • Solid-State Research Group, Strathclyde Institute of Pharmacy and Biomedical Sciences, University of Strathclyde, 27 Taylor Street, Glasgow G4 0NR, UK. Telephone: +44-0141-548-4877; Fax: +44-0141-552-6443
    Search for more papers by this author
  • Andrea Johnston

    1. Solid-State Research Group, Strathclyde Institute of Pharmacy and Biomedical Sciences, University of Strathclyde, 27 Taylor Street, Glasgow G4 0NR, UK
    Search for more papers by this author

Abstract

The crystal structures of γ-carbamazepine (Pequation image, Z′ = 4) and chlorothiazide N,N-dimethylformamide solvate (1/2) (P21/c, Z′ = 2) have been determined from synchrotron and laboratory X-ray powder diffraction data, respectively, using simulated annealing. Both structures represent a significant challenge for global optimization and the successful solutions and subsequent refinements highlight the ever-expanding range of applicability of powder diffraction to structural problems of pharmaceutical relevance. © 2007 Wiley-Liss, Inc. and the American Pharmacists Association J Pharm Sci 96: 1192–1202, 2007

INTRODUCTION

Global optimization methods for crystal structure determination from powder diffraction data (SDPD) have become widely available in recent years and have successfully been applied to solving organic crystal structures, often in the context of polymorphism investigations.1, 2 The basis of global optimization strategies has been fully described elsewhere3, 4 and software implementing global optimization methods is now widely available, including DASH;5 Endeavour;6 ESPOIR;7 FOX;8 PowderSolve;9 PSSP;10 Topas.11

It is the systematic investigation and development of simulated annealing (SA), as implemented in DASH, for pharmaceutical structure determination that concerns us here. Early structure determinations of capsaicin, thiothixene and promazine hydrochloride,12 zopiclone,13 and famotidine14 have stimulated numerous follow-on DASH successes—recent examples include hydrochlorothiazide form II;15 hydrochlorothiazide methyl acetate solvate;16 ampicillin trihydrate;17 pamoic acid;18 chlorothiazide N,N-dimethylformamide solvate;19 amodiaquinium dichloride dihydrate.20

The work reported here continues this theme with a redetermination of the anticonvulsant γ-carbamazepine (γ-CBZ, also known as form I) and a determination of the previously unreported chlorothiazide N,N-dimethylformamide solvate (1/2) (CT-DMF2) (Fig. 1). The single-crystal structure of γ-CBZ21 is triclinic (Pequation image, Z′ = 4), with a = 5.1705(6), b = 20.574(2), c = 22.245(2) Å, α = 84.124(4), β = 88.008(4), γ = 85.187(4)°, and V = 2344.8(5) Å3 at 158 K. Although the publication of this structure preempted an SDPD investigation that was ongoing at the time, the powder solution of a structure with this complexity (seven degrees of freedom (DOF) for each of the four fragments) is a valuable milestone in global optimization and is reported for that reason. In the case of CT, two DMF solvates were crystallized during a preliminary solvent screen run in preparation for an automated parallel crystallization22 study of this diuretic. The crystal structure of the (1/1) solvate (CT-DMF1; Pequation image, Z′ = 1) was solved using DASH and subsequent Rietveld refinement,23 using data collected to 1.5 Å resolution, yielded an Rwp of 0.050.19 The second solvate (CT-DMF2) was shown to be Z′ = 2 and, with three-times the number of DOF of the (1/1) solvate (seven DOF for each of the six fragments), represents another significant challenge for global optimization.

Figure 1.

CBZ (top) and CT-DMF2 (bottom).

MATERIALS AND METHODS

A polycrystalline sample of γ-CBZ, prepared by heating the monoclinic β-form in an oven at 165°C for 48 h, was lightly ground and placed into a 0.9 mm borosilicate glass capillary. Powder diffraction data were collected at 160 K on beamline ID31 at the ESRF in Grenoble (λ = 0.51561 Å). CT-DMF2 was crystallized from a warm (35°C) saturated solution of chlorothiazide in DMF, cooled to room temperature and allowed to stand for 3 months in a partially sealed, narrow-necked test-tube that allowed only very slow evaporation of DMF. A sample was loaded into a 0.7 mm borosilicate glass capillary, along with a small amount of saturated DMF solution to prevent desolvation. Room temperature powder diffraction data were collected on a Bruker-AXS D8 Advance diffractometer (Cu Kα1, λ = 1.54056 Å) equipped with a Braun metal wire linear position sensitive detector. Both data collections utilized a variable count time (VCT) Scheme24, 25 in which the step time is increased with 2θ (Tab. 1). Powder pattern indexing, space group determination, and structure determination were carried out using the DASH computer program.5 Internal coordinate descriptions for γ-CBZ and CT-DMF2 were constructed with the aid of previously determined crystal structures and the SA component of DASH used to optimize the position, orientation, and conformation of these models against the relevant diffraction data sets. In the case of CT-DMF2, the tautomeric H(N)-atom on the thiadiazine ring was placed as shown in Figure 1, consistent with the single-crystal structure of CT.26 Given the complexity of the structure solutions being attempted, the number of SA moves allowed for each run was set to 20 × 106 and 800 × 106 for γ-CBZ and CT-DMF2, respectively. Rietveld refinement of the best solutions obtained was carried out using TOPAS.11

Table 1. VCT Schemes Used During Data Collections
γ-CBZ (Step Size = 0.003° 2θ)CT-DMF2 (step size = 0.014° 2θ)
Solution and RefinementSolutionRefinement
Range (° 2θ)Scan Rate (°/min)Range (° 2θ)Sec./StepRange (° 2θ)Sec./Step
2–32104.3–101.54–153
16–36110–19.5415–256
20–360.219.5–28925–3514
  28–321735–5032
  32–433550–7064

RESULTS

The refined lattice parameters for γ-CBZ (Tab. 2) are in excellent agreement with the previously reported single-crystal structure.21 Of the twenty DASH structure determination runs performed, 14 returned favorable equation image ratios of ∼3, suggesting that the structure had been solved. Confidence in the solution was enhanced by the fact that the structure corresponding to the best χ2 had a chemically reasonable lattice packing arrangement, with no significant misfit to the diffraction data. In essence, the solution was superimposable on the single-crystal structure. This solution was refined against data in the range 1–30° 2θ (5252 reflections), using a restrained Rietveld method as implemented in TOPAS. All atomic positions (including H-atoms) were refined, subject to a series of restraints on bond lengths, bond angles and group planarity, where appropriate. The distance and angle restraints were based on the single-crystal structure of γ-CBZ,21 as was the orientation of each amide functional group. A spherical harmonics correction of intensities for preferred orientation was applied in the final refinement.27 An excellent fit to the data was achieved (Fig. 2) and the refined crystal structure is shown in Figure 3. Each of the four molecules in the asymmetric unit forms a bimolecular hydrogen-bonded ring motif (Fig. 4). The hydrophobic aromatic interactions appear to have little significance with regard to directing molecular stacking: (a) offset face-to-face π − π interactions, with centroid-to-centroid distances of 5.670(1) and 5.960(1) Å, falling just within the 6 Å cut-off for significance applied in PLATON;29 (b) edge-to-face C−H···π interactions, with H···centroid distances in the range 2.890(7) – 2.969(7) Å, falling just within the 3 Å cut-off for consideration.

Table 2. Refined Unit Cells, Extracted Intensity Information and Final χ2 for the Compounds Shown in Figure 1
Parametersγ-CBZCT-DMF2
  1. Nreflections is the number of reflections in the data range shown, equation image is the χ2 for the Pawley-type fit to the data range shown, whilst Rexp, Rp, and Rwp are standard profile R-factors for the Pawley-type fit to the data range shown.

a (Å)5.1853(3)12.3609(8)
b (Å)20.5777(5)8.5629(7)
c (Å)22.2387(5)37.3091(5)
α (°)84.1959(14)90
β (°)87.9872(11)92.8769(26)
γ (°)85.1009(13)90
V3)2351.27(16)3944.0(4)
T(K)160293
Space groupPequation imageP21/c
Data range (° 2θ)1–12.294–36.6
Resolution (Å)2.412.45
Nreflections349281
equation image3.941.82
Rexp9.057.72
Rp18.0415.5
Rwp17.9610.42
Figure 2.

Observed profile (○), calculated profile (–), and difference [(yobs − ycalc)/σ(yobs)] plot for the Rietveld refinement of γ-CBZ. Bottom plot: high-angle data in the range 14–30° 2θ. Refined lattice parameters: a = 5.18561(9), b = 20.57575(29), c = 22.24106(30) Å, α = 84.1942(8), β = 87.9756(7), γ = 85.1053(8)°, V = 2351.44(7) Å3 at 160 K (Rexp = 1.654; Rwp = 2.600; Rp = 4.549).

Figure 3.

Molecules A–D in the asymmetric unit of γ-CBZ. Displacement ellipsoids are shown at the 50% probability level.

Figure 4.

Formation of equation image (8) rings amongst molecules A–D in γ-CBZ (following the graph-set notation of reference28), showing the H···O distances.

The CT-DMF2 diffraction pattern indexed to a monoclinic cell (F(20) = 52.5, M(20) = 15.1; DICVOL-9130) and space group P21/c was assigned from volume considerations and a statistical consideration of the systematic absences.31 Pawley32 refinement in DASH gave an excellent fit to the data, confirming the assignment of the unit cell and space group (Tab. 2). Of the 99 DASH runs performed, 4 solutions were obtained with favorable equation image ratios of ∼6–10, suggesting that the structure had been solved. Confidence in the solution was enhanced by the fact that the structure corresponding to the best χ2 had a chemically reasonable lattice packing arrangement, with no significant misfit to the diffraction data. This solution was refined against data in the range 4–65° 2θ (1422 reflections) using the same procedure as outlined above for γ-CBZ, with distance and angle restraints based on the single-crystal structure of CT.26 An excellent fit to the data was achieved (Fig. 5) and the refined crystal structure is shown in Figure 6. The structure is stabilized by six intermolecular N−H···O hydrogen bonds (Tab. 3; Fig. 7) and offset face-to-face π − π interactions (Fig. 8).

Figure 5.

Observed profile (○), calculated profile (–), and difference [(yobs − ycalc)/σ(yobs)] plot for the Rietveld refinement of CT-DMF2. Bottom plot: high-angle data in the range 40–65° 2θ. Refined lattice parameters: a = 12.3586(2), b = 8.5619(2), c = 37.3043(7) Å, α = 90, β = 92.8786(13), γ = 90°, V = 3942.30(13) Å3 at 293 K (Rexp = 1.409; Rwp = 2.031; Rp = 1.837).

Figure 6.

The asymmetric unit of CT-CMF2, showing CT molecules A and B plus solvent molecules C–F. Displacement ellipsoids are shown at the 50% probability level.

Table 3. Hydrogen Bond Geometry in CT-DMF2
 D−H···AD−H (Å)H···A (Å)D···A (Å)D−H···A (°)A (Symmetry Code)
1N2[BOND]H4···O3S0.951 (14)1.762 (14)2.697 (4)166.7 (15)−1 + x, y, z
2N3[BOND]H5···O2S0.949 (16)2.106 (16)3.030 (5)164.1 (12)1 − x, 1 − y, −z
3N3[BOND]H6···O2S0.947 (16)2.012 (16)2.848 (4)146.1 (13)x, y, z
4N2A[BOND]H4A···O1S0.947 (14)1.753 (14)2.672 (4)162.9 (15)x, y, z
5N3A[BOND]H5A···O4S0.949 (16)1.944 (16)2.825 (5)153.5 (13)−1 + x, y, z
6N3A[BOND]H6A···O4S0.955 (15)2.085 (16)3.024 (5)167.1 (12)1 − x, −y, −z
Figure 7.

Hydrogen bonds 13 (Tab. 3) in CT-DMF2, showing the equation image (8) motif. The centrosymmetric arrangement of six molecules shown here is also observed with CT molecule B, plus DMF molecules E and F, combining via hydrogen bonds 4–6. As reported elsewhere for CT-DMF1,19 rotating the sulfonamide group of CT molecule A in increments of 120° about the S2[BOND]C5 bond (Fig. 6) results in three orientations that are similar in the sense that the X-ray scattering power of N3(H2) is on a par with that of atoms O3 and O4 (the same applies to CT molecule B by rotation around bond S2A[BOND]C5A). In this case, the correctness of the orientation shown above was confirmed by the intermolecular hydrogen-bonding pattern. The H-atoms on N3 were placed in calculated positions that minimize the N−H···O hydrogen bond distances.

Figure 8.

Hydrophobic offset face-to-face π − π interactions between CT molecules A and B in the crystal structure of CT-DMF2.

DISCUSSION

It has been known for some time that global optimization methods are capable of solving relatively complex crystal structures routinely from laboratory XRPD data, even in the presence of complicating factors such as disorder and preferred orientation.33 One may go so far as to say that SDPD has lived up to its “routine” designation,12, 25 but the technique does have limitations and determinations may fail, typically when the data cannot be indexed or when the starting molecular model is deficient in some respect. It is also significant that the upper limit of effectiveness of the SA algorithm has yet to be established, although a previous investigation33 demonstrated that, given XRPD data collected to 2 Å resolution or better: (a) structures with <15 DOF present little challenge, yielding accurate structure solutions reproducibly; (b) for structures with greater complexity (DOF = 15–20), where the preponderance of local minima in the agreement hypersurface reduces the frequency of success, the SA algorithm is still able to locate the global minimum with a reasonable frequency. Higher levels of complexity (DOF > 20) remain to be explored systematically; hence γ-CBZ and CT-DMF2 represent significant contributions to this wider effort (Tab. 4).

Table 4. A Summary of the Most Complex DASH Structure Solution Successes to Date
 C9H11N2O2S · Cl · H2Oaγ-CBZbCT-DMF2b
  • a

    C9H11N2O2S · Cl · H2O has the highest total DOF attempted by the authors, prior to this work.34

  • b

    CCDC 635853–635854 contain the supplementary crystallographic data for this paper. These data can be obtained free of charge from The Cambridge Crystallographic Data Center via www.ccdc.cam.ac.uk/data_request/cif.

Total DOF282842
External DOF (position)181218
External DOF (orientation)61218
Internal DOF446
Z242
Fragments in asymmetric unit646
Total number of atoms in asymmetric unit5812094
Number of nonhydrogen atoms in asymmetric unit327254

Several strategies were adopted from the outset in order to maximize the chances of successfully solving both structures (Tab. 5), with a heavy emphasis on a requirement to collect the best possible diffraction data, to the highest possible resolution, in order to facilitate structure solution and accurate refinement. The γ-CBZ diffraction data, in particular, are very high quality (∼1 Å resolution) and a free Rietveld refinement (i.e. one that employs no restraints and thus has 216 nonhydrogen atom coordinates refining freely) returns a recognizable structure. However, as always with Rietveld refinement, the caveat is that an improved fit to the diffraction data should not be pursued at the expense of chemical sense and, in the free refinement, the individual CBZ molecules are slightly distorted from chemically acceptable values. Therefore, restrained Rietveld refinement is recommended and is readily implemented within TOPAS. It is also essential to closely examine the final refined structure to ensure that the orientation of functional groups such as [BOND]CONH2 and [BOND]SO2NH2 is correct, acknowledging that there is little to distinguish [BOND]NH2 from an O-atom, in terms of X-ray scattering.

Table 5. Approaches for optimizing Data Quality and Maximizing the Chances of Successfully Solving Crystal Structures from XRPD Data
ApproachAim/AdvantageComment
RecrystallizationMinimize intrinsic sample line width; improve angular resolutionRisk of phase transformation or texture effects
Low-temperature data collectionImprove signal-to-noise, particularly at high 2θ angles; improve accuracy of reflection intensitiesRisk of phase transformation
VCT data collectionAs “low-temperature data collection”Simple to implement and markedly superior to data collections that use equal count times at every point in the pattern
SynchrotronHigh instrumental resolution; rapid data collectionAccess to beamline is by peer-reviewed proposal
Optimize SA control parametersIncrease probability of locating global minimum, that is, correct crystal structureFor example, reduce the “cooling rate” to avoid quenching
Crystallographic constraintsReduce number of DOF to be optimized during search; increase probability of locating global minimumFor example, in space groups such as P1, with a floating origin, fixing the x, y, and z coordinates of an atom in a formula unit removes three DOF
Chemical constraintsAs ‘crystallographic constraints’For example, fixing an amide torsion angle to an exact value of 180° removes one DOF from the optimization

Available computing power is an important factor in tackling complex SDPD problems for two reasons: (a) a large number of SA moves per run is necessary to give the algorithm a chance to explore the complex χ2 space and locate the global minimum; (b) a large number of runs is required, since no one run is guaranteed to find the global minimum and a complex search space inevitably results in a low frequency of success. Fortunately, this type of problem is ideally suited to execution on distributed computing systems35 that take advantage of the vast number of “spare” CPU cycles present in modern PCs, running the individual SA structure solution attempts in parallel. The DASH code has now been adapted to run on the commercially available United Devices Grid MP platform36 that is already widely used within pharmaceutical industry to accelerate drug discovery and clinical modeling and it can easily be run on other well-known noncommercial systems such as Condor.37 This “grid-enabling” of the structure solution stage facilitates rapid turnaround from XRPD experiment to crystal structure and, as such, is expected to add significant value to the physical form screening process.

CONCLUSIONS

The ability to determine molecular crystal structures from powder diffraction data means that structures that may previously have gone unreported can now be solved and refined to the publication standards required by Acta Crystallographica.29 Thus, the structural landscape revealed by a physical form screen that incorporates both SDPD and single-crystal diffraction can be considerably richer than a screen that is dependent upon single-crystal diffraction alone. The structures reported here show that structural complexity need not necessarily be a barrier to successful structure determination from powders.

Acknowledgements

We thank the Basic Technology Program of The Research Councils UK for funding under the project Control and Prediction of the Organic Solid State (http://www.cposs.org.uk); EPSRC for grants GR/N07462/01 and GR/S10162/01; CCLRC Center for Molecular Structure and Dynamics for studentship funding for PF.

Ancillary