An XRD and NMR crystallographic investigation of the structure of 2,6-lutidinium hydrogen fumarate

Fumarate is a pharmaceutically acceptable counterion often used to modify the biophysical properties of active pharmaceutical ingredients (APIs) through salt formation. With 2,6-lutidine (2,6-dimethylpyridine), fumaric acid forms the salt 2,6-lutidinium hydrogen fumarate. An NMR crystallography approach was employed to investigate the salt structure and the intermolecular interactions involved in its formation and stability. The crystallographic unit cell was determined by both single crystal XRD (SXRD) and synchrotron powder X-ray diffraction (PXRD) to contract at low temperature with a skew in the β angle. Density functional theory (DFT)-based geometry optimisations were found partially to replicate this. A second room temperature structure was also identified which exhibited a similar skew of the β angle as the low temperature structure. DFT calculation was also employed, alongside 2D 1 H double-quantum (DQ) magic angle spinning (MAS) and 14 N- 1 H HMQC MAS NMR spectra, to investigate the hydrogen bonding network involved in the structure. DFT-based gauge-including projector-augmented wave (GIPAW) calculations highlighted both strong N + -H···O − and O-H···O intermolecular hydrogen bonds between the molecules, as well as several weaker CH···O hydrogen bonds. Both PXRD and solid-state MAS NMR, supported by thermal gravimetric analysis (TGA) and solution-state NMR analysis, show formation of fumaric acid within samples over time. This was evidenced by the identification of reflections and peaks associated with crystalline fumaric acid in the PXRD pattern and in 1 H MAS and 13 C cross polarization (CP) MAS solid-state NMR spectra, respectively.


Introduction
The analytical characterisation of the solid-state structures adopted by an active pharmaceutical ingredient (API) or agrochemical ingredient (AI) is an important step in the process of optimising the design and efficacy of products. A key area of research pertains to methods of enhancing an API's and AI's physical properties without detrimentally affecting its bioactivity. Increasingly, the APIs being developed are larger and more insoluble. 1, 2 Thus, developing approaches to improve their solubility and rate of dissolution are of particular importance. The development of salt forms offers the possibility of altering biophysical properties, 3 such as solubility, bioavailability and stability, although other methods for enabling formulation like co-crystallization and rendering the material amorphous are also employed. [2][3][4][5][6] Salts and cocrystals are often distinguished by their ionicity, with cocrystals containing neutral molecules.
One formal definition of a cocrystal, according to Dunitz,7 also includes solvates, although they can alternatively be defined separately by the physical distinction that one of the components is liquid at room temperature. 4,8 Solvates are intrinsically unstable and generally undergo some form of phase change close to room temperature, 8 making both their use and characterisation more challenging. They are still of interest though as they represent another potential route to additional biologically active forms that may have desirable properties. The characterisation of such systems may also be relevant in cases where an API/AI is liquid at room temperature, but it is preferable to be able to store or administer it in solid form (for cost, convenience and/or ease of use).
Traditionally, X-ray diffraction (XRD) is employed for structure determination of crystalline materials. Single crystal XRD (SXRD) is the gold standard for organic molecules, with powder XRD (PXRD) mainly used to fingerprint or refine the resulting structure. However, where suitable single crystals are not available, structure determination has been achieved from powder patterns. [9][10][11][12][13][14] Distinguishing between salts and cocrystal forms by XRD can be difficult, however, as it is inherently insensitive to low atomic number elements. As a result, it is not always possible to provide accurate proton positions, particularly if the quality of single crystals is poor. 6 This limits the ability of XRD to give a detailed representation of the intermolecular interactions present within a system. A more thorough investigation of these interactions can inform how the crystal structure is being held together, aiding in the development of methods to reliably predict the stability of multicomponent solid forms. 8 NMR crystallography [15][16][17][18][19] is a growing field in which solidstate NMR, under magic-angle spinning (MAS), and calculations of NMR parameters from first principles using density functional theory (DFT), notably the gauge-including projector-augmented wave (GIPAW) method, are combined. 20,21 It complements other techniques (generally XRD) to either facilitate structure determination, 9,16,[22][23][24] interrogate the nature of both weak and strong intermolecular interactions [25][26][27][28] or investigate the structure of localised disorder within systems. 29 NMR crystallography of organic systems benefits particularly from the intrinsic sensitivity of 1 H chemical shifts to local interactions. 30 In this work, NMR, XRD and DFT are combined and utilised to investigate the intermolecular interactions and stability of 2,6lutidinium hydrogen fumarate. 31 Fumarate is a pharmaceutically acceptable counterion for salt formation and is already employed to improve the properties of various APIs, including iron, bisoprolol and tenofovir disoproxil. 32 Fumaric acid is a crystalline solid at room temperature and can exist in a variety of crystal forms with different isomers of lutidine, as investigated by Haynes et al. 33 Conversely, 2,6-lutidine is liquid at room temperature, which allows the system to be described as either a API/AI-fumarate or as a solvate.

Sample preparation
All chemicals were obtained from Sigma Aldrich (UK) at purities of 98% or higher and used without further purification.
Purity was verified by 1 H solution-state NMR. Fumaric acid was dissolved in isopropanol and 2,6-lutidine was added in a 1:1 molar ratio. Samples were made on a ~50 mg scale with 24 mg fumaric acid and 26 μL 2,6-lutidine (with density 0.92 mg / L).
Crystallisation was achieved by slow solvent evaporation over approximately 4 days.

Calculations
Density functional theory (DFT) calculations were performed using CASTEP 34 Academic Release version 16.1. All calculations used the Perdew Burke Ernzerhof (PBE) exchange correlation functional, 35 a plane-wave basis set with ultrasoft pseudopotentials and a plane-wave cut-off energy of 700 eV.
Integrals over the Brillouin zone were taken using a Monkhorst-Pack grid of minimum sample spacing 0.1 × 2π Å −1 (see Fig. S1 for convergence of energy with decreasing spacing). The literature structure, 31  Geometry optimisations were also run for each structure allowing the unit cell parameters to vary, with a dispersion correction applied under the scheme proposed by Tkatchenko and Scheffler. 37 MOGUL 38 searches were performed both before and after geometry optimisation to ensure that the bond lengths, torsion and angles were consistent with the CCDC database.
Distances stated in this paper are for the DFT optimised structure calculated with fixed unit cell (unless otherwise stated).
NMR parameters were calculated using the gauge-including projector-augmented wave (GIPAW) 20 method and were performed for both the geometry optimised crystal structures as a whole and for each of the isolated molecules in the asymmetric unit. For the isolated molecule calculations, each molecule in the asymmetric unit was extracted from the geometry optimised unit cell and placed in a sufficiently large box such that it could not interact with repeated molecules across periodic boundary conditions 28 (unit cell dimensions increased by 10 Å in each direction). As each individual molecule carried a charge, this was specified in the .param file. 39 The calculated isotropic chemical shifts (δiso calc ) were determined from the calculated chemical shieldings (σcalc) by δiso calc = σref -σcalc, with σref values of 30.5, 169.7 and −153 ppm for 1 H, 13 C and 14 N, respectively. σref was determined for 1 H and Please do not adjust margins Please do not adjust margins resulting y-intercept was taken as σref. 40,41 A literature value was used for 14 N. 42 It is noted that it is common practice to calculate a specific reference shielding for each system 43 (see, e.g., with an Atlas S2 CCD detector. Crystal screening was conducted at room temperature. CrysAlisPro 46 data-collection and processing software was used, allowing crystals to be checked for quality and giving a preliminary unit cell determination by using a short pre-experiment prior to full data collection. This pre-experiment was used to screen a large number of crystals from each crystallisation, with full data collection only run if a discrepancy was identified between the experimental unit cell parameters and those found in the CCDC. Following full data collection, ShelXL 47 was used for structure solution and a leastsquares refinement was run, using the Olex2 48 software.

Supplementary
Temperatures down to 100 K were also employed to monitor the contraction of the unit cell. A comprehensive analysis of the powder patterns was undertaken using TOPAS Academic v6, 49

XRD
Single crystals of 2,6-lutidinium hydrogen fumarate were successfully obtained by slow evaporation. Crystals exhibited a thin plate morphology and, upon closer inspection with an optical microscope, thicker regions of the specimen could be seen as stacks of multiple thin plates, with some evidence of intergrowths.
To ensure that each sample was homogeneous, room temperature SXRD at 293 K was utilised to screen each crystallisation for unit cell variations and polymorphism. Most of the crystals found to be suitable for diffraction refined to the structure, published by Pan et al. 31  Following grinding to a powder, room temperature PXRD patterns were run to ensure no changes had been induced during the grinding process. Fig. 2 shows the result of a Rietveld refinement against a high-resolution synchrotron scan of 2,6lutidinium hydrogen fumarate carried out under ambient conditions with a Rbragg of 5.59%. As can be seen in Table 1, there is good agreement between the refined unit cell (MAC 300 K) and that of structure 181445 published by Pan et al., 31 with only small differences in the unit cell parameters consistent with the small difference in temperature between data collections. A more detailed comparison of atomic coordinates is presented in Table   S1 (SI).
In the Rietveld refinement all the experimental reflections (as shown by the tick marks) are replicated in the calculated pattern and their 2θ positions are in excellent agreement. Although there was some evidence of differences in peak height and shape between experiment and calculation, as seen in the difference plot in Fig. 2, this can be explained by residual preferred orientation effects. The natural plate morphology of the crystals results in strong preferred orientation effects due to the alignment of crystallites. Minimisation of this by grinding the powder more finely, to allow the collection of better PXRD data, was hindered by breakdown of the crystal structure if the sample was ground for too long (discussed below). Most powders analysed therefore Please do not adjust margins Please do not adjust margins still contained microcrystallites, exacerbating the preferred orientation effects. Taking these effects into account, the Rietveld refinement of the PXRD data was therefore considered sufficient to confirm that no structural change had occurred on grinding and the published structure of 2,6-lutidinium hydrogen fumarate is a suitable model. Fig. 3a and 3b presents 1 H one-pulse MAS and 1 H-13 C CP MAS NMR spectra of 2,6-lutidinium hydrogen fumarate together with stick spectra that represent the 1 H and 13 C chemical shifts calculated using the GIPAW method for the geometry optimised crystal structure (see Table S6  calculation is the occurrence of two distinct peaks at high chemical shift rather than one, as discussed below.

NMR
In the 1 H-13 C CP MAS NMR spectrum (Fig. 3b) 71 an alternative approach would be to use different reference shieldings for different parts of the spectrum. 70 As noted above, two peaks are observed above 10 ppm in the 1 H MAS NMR spectrum (Fig. 3a) whereas, as can be seen from the stick spectra, the GIPAW calculated chemical shifts predict that both H13 and H10 (the OH and NH protons, respectively) are at the same value of 17.7 ppm. In the 2D 1 H- 13  as up to a distance of 3.5 Å). 30 Table 2 lists the H···H proximities respectively. The next closest proximity to the NH, H10, is to H11, which also falls within 3.5 Å of the OH, H13 (see Table 2

GIPAW calculations of NMR chemical shifts for isolated molecules
A comparison of the chemical shifts calculated for the full crystal structure of 2,6-lutidinium hydrogen fumarate to those calculated for individual isolated molecules, as extracted from the geometry optimised crystal structure, can provide additional insight into the crystal packing. 27,28,45,[72][73][74] The difference between the crystal and isolated molecule chemical shifts ( H1 and H2 exhibit weak ring current effects (−1.0 ppm), possibly due to the stacking of the pyridine rings within the crystal structure, although it is interesting to note that this effect seems to be offset for H3 which instead shows a very slight positive change (0.6 ppm). This is due to weaker effects from ring currents, as the stacking is slightly misaligned (Fig. 1d, bottom right), placing it further out from the π-π stack.

Skewed cell contraction
Please do not adjust margins Please do not adjust margins Fig. 6 shows a Rietveld refinement against a PXRD high resolution synchrotron scan that was collected at 100 K. As with the room temperature high resolution scan shown in Fig. 2, residual preferred orientation effects prevent accurate refinement of atomic positions, but the tick marks, corresponding to 2θ values of the reflections expected for the refined structure, are in excellent agreement with those recorded experimentally. As can be seen in Table 1, this Rietveld refinement showed a contraction accompanied by an increase in the β angle. This effect was also observed in SXRD crystal screens conducted at a range of temperatures between 100 K and 300 K (Fig. S3).
Compared to 181445, the Rietveld refinement for the 300 K PXRD pattern presented above (Fig. 2) shows a small, but similarly skewed expansion, with a marginal decrease in β angle, alongside increases in cell lengths, consistent with being recorded at a slightly higher temperature than the literature structure (300 K compared to 292 K). The skew in the cell, when going to low temperature during contraction, is also evident in the DFT calculations (also shown in Table 1). Geometry optimisations, performed allowing the unit cell parameters to vary, showed convergence towards the low temperature structure. As no external temperature is included in the calculations, they are effectively performed at 0 K, so it is unsurprising they would exhibit such a tendency.
During the room temperature crystal SXRD screening to check each crystallisation for variations or new polymorphs, one crystal was identified that differed slightly from the previously identified structure. This newly identified structure has been deposited with the CCDC (no. 1876100) and selected crystal data is given in Table 4. Although the molecular packing of the crystal remained unchanged, with only small shifts in relative atomic positions (Fig. 7), the unit cell parameters presented in Table 1 show both a slight contraction in the unit cell lengths (the most A GIPAW calculation for 1876100, following geometry optimisation, showed that the minimal shifts in atom position produced only small changes to the calculated chemical shifts (the largest being 0.2 ppm for 1 H and 0.6 ppm for 13 C, with mean differences of 0.01 ppm and 0.1 ppm, respectively, see Table S6). Extensive PXRD analysis failed to provide conclusive evidence for even a minor second structural phase within the powdered compound, suggesting that 1876100 may have been an isolated case. However, in the case of only partially ground powders (so as to as minimise break down of the co-crystal, see below discussion), some of the reflections appeared to show a splitting. Several larger crystallites were observed in static transmission PXRD of these samples, using a 2D detector, that may explain this. They lie slightly off the main powder ring, as predicted due to a small change in β angle (Fig. 8), and might therefore be linked to 1876100. As 1876100 seems to be related to the low temperature contraction, it is unclear how it existed within a room temperature SXRD screen. The energies of 181445 and 1876100, determined by DFT (see Table 1), differed by only 0.9 kJ/mol after geometry optimisation (allowing both atom positions and unit cell parameters to vary).  converged towards the 100 K structure but neither reached it, with each satisfying the convergence conditions of the DFT calculation with larger volume unit cells than that of the 100 K structure. This discrepancy is probably due to the limitations of the dispersion correction scheme. Although the difference between the two unit cells was smaller following variable cell geometry optimisation, the output of the calculation based on 1876100 was still slightly more contracted than that based on the original CCDC structure, 181445. It can also clearly be seen (Table 1) that despite effectively being at a lower temperature (0 K), the outputs of neither calculation exhibit as significant a contraction as is evident for the 100 K synchrotron case, although the optimisation of 1876100 did produce a similar skew of the β angle.

Formation of fumaric acid
Samples that had been stored for more than a week as powders rather than single crystals showed additional peaks in the 1 H MAS and 1 H -13 C CP MAS NMR spectra, as shown in Fig. 3c and 3d. The high chemical shift of the new 1 H peak, 12. hrs, which then plateaued (Fig. 11). The loss in mass corresponds to 96.3% of the 2,6-lutidine that was present in the complete crystal structure originally. If the mass loss is due to evaporation of 2,6-lutidine, as proposed, the plateau prior to complete loss suggests that the remaining 3.7% of 2,6-lutidine molecules are trapped in the centre of the crystallites, with insufficient energy at 70 °C to escape. This could be due to the collapse of the majority of the structure preventing evaporation of this residual 3.7% of 2,6-lutidine molecules.

Conclusions
A combined NMR and XRD crystallographic investigation of 2,6-lutidinium hydrogen fumarate has been presented together with a computational study based on DFT geometry optimisation and GIPAW calculation of NMR chemical shifts.
The use of this combined approach enabled the identification of fumaric acid formation within powder samples over time.
Based on a corresponding reduction in the ratio of 2,6-lutidine to

Conflicts of interest
There are no conflicts to declare.
12 | J. Name., 2012, 00, 1-3 This journal is © The Royal Society of Chemistry 20xx Please do not adjust margins Please do not adjust margins      Table 2. The dashed diagonal line indicates the δDQ = 2δSQ diagonal, while horizontal lines indicate a DQ peak at the sum of the two SQ peaks for dipolar coupled unlike protons.