5-amino-2-methylpyridinium hydrogen fumarate: an XRD and NMR crystallography analysis

Single-crystal X-ray diffraction structures of the 5-amino-2-methylpyridinium hydrogen fumarate salt have been solved at 150 and 300 K (CCDC 1952142 and 1952143). A base-acid-base-acid ring is formed through pyridinium-carboxylate and amine-carboxylate hydrogen bonds that hold together chains formed from hydrogen-bonded hydrogen fumarate ions. H and C chemical shifts as well as N shifts that additionally depend on the quadrupolar interaction are determined by experimental magic-angle spinning (MAS) solid-state nuclear magnetic resonance (NMR) and gauge-including projector augmented wave (GIPAW) calculation. Two-dimensional homonuclear H-H doublequantum (DQ) MAS and heteronuclear H-C and N-H spectra are presented. Only small differences of up to 0.1 ppm and 0.6 ppm for H and C are observed between GIPAW calculations starting with the two structures solved at 150 and 300 K (after geometry optimisation of atomic positions, but not unit cell parameters). A comparison of GIPAW calculated H chemical shifts for isolated molecules and the full crystal structures is indicative of hydrogen bonding strength. Introduction Salt formation has been common practice within the pharmaceutical industry for more than 25 years as a method of altering the biophysical characteristics of products without altering their pharmacology. An NMR crystallography approach, where solid-state NMR and density functional theory (DFT) calculations are employed alongside complementary techniques, is increasingly utilised to characterise the solid form. In the case of crystalline solids, it generally accompanies a structure solution from Xray diffraction (XRD), providing further insight into the intermolecular interactions and allowing validation and refinement of atomic positions. This is particularly crucial for systems solved from only powder XRD (PXRD) data. The development of salt systems can be complicated by hydrate formation, as charged ions can interact strongly with polar molecules such as water, which generally leads to instability issues and room temperature phase transitions. Many pharmaceutically acceptable counter ions contain


Introduction
Salt formation has been common practice within the pharmaceutical industry for more than 25 years 1 as a method of altering the biophysical characteristics of products without altering their pharmacology.
An NMR crystallography approach, [2][3][4][5][6] where solid-state NMR and density functional theory (DFT) calculations are employed alongside complementary techniques, is increasingly utilised to characterise the solid form. In the case of crystalline solids, it generally accompanies a structure solution from Xray diffraction (XRD), providing further insight into the intermolecular interactions and allowing validation and refinement of atomic positions. This is particularly crucial for systems solved from only powder XRD (PXRD) data.
The development of salt systems can be complicated by hydrate formation, as charged ions can interact strongly with polar molecules such as water, 7 which generally leads to instability issues and room temperature phase transitions. 8 Many pharmaceutically acceptable counter ions contain carboxylate groups which are strongly charged but studies of the Cambridge Structural Database (CSD) show that the incidence of hydrate formation is significantly lower for crystallisations involving pyridine derivatives. 9 Here we report on the crystallisation of 5-amino-2methylpyridine (52AMP) with fumaric acid (FA), a pharmaceutically acceptable co-former (see Scheme 1). A new salt form, 5-amino-2methylpyridinium hydrogen fumarate (52AMP:F) is reported from single crystal XRD (SXRD) and characterised using a multi-nuclear NMR crystallography approach.

Experimental details
All chemicals were obtained from Sigma Aldrich (UK) at purities of 98% or higher and used without further purification. 52AMP:F was crystallised by mixing the base (81 mg) and fumaric acid (87 mg) in the minimum amount of hot methanol required to dissolve all the solutes (~ 15 mL) and then allowing the resulting solution to cool slowly at room temperature. Crystals began to form after two days. Crystal growth was improved by co-grinding before dissolution and the addition of seed crystals to subsequent crystallisations.
Crystals were selected for SXRD using polarised light microscopy with an Olympus SZ61 Stereomicroscope. Those that appeared by shape and birefringence to be single crystals were chosen. SXRD was carried out using Cu K α1 (1.5406 Å) on a Rigaku Oxford Diffraction SuperNova diffractometer with an Atlas S2 CCD detector equipped with an Oxford Cryosystems N-Helix cooling system. Crystal screening was conducted at room temperature. CrysAlisPro 10 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 also used to screen a minimum of 10 crystals from each crystallisation. Following full data collection, ShelXL 11 was used for structure solution and a least-squares refinement was run, using the Olex2 12 software. Following screening by SXRD, the most crystalline components of each crystallisation were ground to a fine powder and the structure was checked by PXRD to determine bulk purity and ensure no changes had occurred under grinding, by comparing the experimental powder pattern to the pattern predicted from the crystal structure. PXRD was performed on a Panalytical X'Pert Pro MPD equipped with a curved Ge Johansson monochromator, giving pure Cu K α1 radiation and a solid state PiXcel detector. The powder samples were mounted on a zero-background offcut-Si holder and spun at 30 rpm.
Each sample was run with a step size of 0.013° and time per step of 2850 s. A comprehensive analysis of the powder patterns was undertaken using TOPAS Academic v6, 13   experiments were performed on a Bruker Avance III spectrometer, operating at 1 H and 13 C Larmor frequencies of 500.0 MHz and 125.8 MHz, respectively, using a 3.2 mm HX probe. A 13 C CP MAS spectrum was acquired with 16 coadded transients, a CP contact time of 1500 μs and a recycle delay of 80 s. A 2D 1 H-13 C HETCOR spectrum was acquired with 24 transients coadded for each of 120 t 1 FIDs using a recycle delay of 80 s, a t 1 increment of 36 μs and a CP contact time of 500 μs (corresponding to a total experiment time of 64 hrs). eDUMBO-1 22 24, 25 homonuclear decoupling was used with a 32 μs cycle, with 320 divisions of 100 ns each. The scaling factor was determined to be 1.6. In the HETCOR pulse sequence, the following phase cycling was employed: 1 H 90° pulse (90º 270°), 13 38 It is noted that it is common practice to calculate a specific reference shielding for each system 39 , though average values over a range of compounds are also available. 40 By comparing the parameters in the full crystal structure with those for the isolated molecule, insight is provided into the intermolecular interactions responsible for maintaining the crystal structure. 41

XRD
The crystal structure of 52AMP:F has been determined, as described below, at both 150 K and 300 K and the structures deposited with the CCDC, no. 1952142 and 1952143, respectively. Selected crystal data for the structure at each temperature are given in Table 1. A small thermal expansion occurred on heating from 150 K to 300 K but with no evident change in the molecular packing. Hydrogen atoms were found in the electron density map. Initial verification of proton transfer was completed by comparison of the carboxylate C-O bond lengths and was then confirmed by 14 N-1 H HMQC NMR experiments, as discussed below.
52AMP:F has a stoichiometry of 1 : 1, base : acid, and crystallises in the triclinic space group P1. Hydrogen fumarate molecules form acid chains along the a-axis with graph set notation C 1 1 (7) (   (Fig. 1b). Through pyridinium-carboxylate and aminecarboxylate H-bonds, a base-acid-base-acid ring is formed, R 4 4 (18), which supports this pairing (Fig.   1c). 42 A H-bond via the other amino proton, to the carboxylic acid O=C, allows crosslinking between paired chains, forming a H-bonded layer on the (010) crystal plane (Fig. 1d). These layers then stack to form the 3D structure. Neither fumaric acid nor doubly ionised fumarate are present, with the occurrence of hydrogen fumarate instead preventing the formation of the base-acid-base units seen in other related systems, such as Bis-(2-amino-5-methylpyridinium) fumarate fumaric acid. 45,46 The H-bond parameters at 300 K for the significant H-bonding motifs identified are given in Table 2.
PXRD of 52AMP:F showed no evident change in structure on grinding (SI, Fig. S1). A Rietveld refinement of the experimental powder pattern against the SXRD structure gave R wp = 8.59% and R Bragg = 1.24% (SI, Tables S1 and S2). Fig. 2 shows 1D 1 H MAS and 1 H-13 C CP MAS spectra for 52AMP:F. Assignments are based on both the GIPAW calculated chemical shifts (Tables 3 and 4) and the 2D MAS NMR spectra, discussed in detail below. It is interesting to note that, despite the noticeable thermal expansion of the unit cell between 150 K and 300 K (Table 1), any changes in interatomic distances are too small to result in significant changes in the GIPAW calculated chemical shifts, with differences limited to 0.1 ppm and 0.6 ppm for 1 H and 13 C, respectively.

NMR
52AMP:F show good agreement for 1 H between experimental NMR chemical shifts and calculated GIPAW chemical shifts. Surprisingly, as no indication was seen in the PXRD pattern in Fig.   S1, the 1 H MAS spectrum for 52AMP:F has a resonance thought to correspond to crystalline fumaric acid at 12.8 ppm (Fig. 2) as seen previously for 2,6-lutidinium hydrogen fumarate. 44 Other than this minor secondary phase, the only discrepancy for 52AMP:F (the established discrepancy between GIPAW calculations and experimental values is only around 1% of the chemical shift range, ~ 0.2 ppm for 1 H) corresponds to the apparent overcalculation of H3, the OH proton involved in an OH···O Hbond, which experimentally is 1.6 ppm lower than the calculated value (δ iso calc = 18.4 ppm compared to . This is the same environment for which a larger than anticipated discrepancy was previously described for 2,6-lutidine hydrogen fumarate. 44 Despite the excellent agreement seen for the 1 H chemical shifts for 52AMP:F, there is relatively poor agreement for the 13 C chemical shifts. The 1% rule of thumb stated above gives expected error between GIPAW calculations and experimental values of up to ~2 ppm for 13 C. It can be clearly seen in the 1D 1 H-13 C CP MAS spectrum in Fig. 2 that this is exceeded for numerous carbon environments in 52AMP:F. For the highest and lowest chemical shifts, as usual, GIPAW calculation underestimates and overestimates the low ppm and high ppm values, respectively. In addition, while C10 is calculated to lie 2.8 ppm lower than it is seen experimentally, both C3 and C4 are calculated to lie more than 3 ppm higher than their experimental chemical shifts. These latter two are the CH carbons in the hydrogen fumarate anion.  peaks are thought to correspond to the H12-H9 proximity (δ DQ = 15.0 + 6.1 = 21.1 ppm). The H12-H9 correlation is expected to be weaker as H9 is significantly further away at 3.38 Å as compared to 2.71 Å for H3 (see Table 5). A 2D SQ 1 H-1 H NOESY spectrum (Fig. 3b) was used to confirm the presence of fumaric acid as it shows the existence of two distinct phases. A mixing time of 500 ms was used to allow spin diffusion throughout the entirety of each phase, 47,48 with the clear separation of cross-peaks indicating the occurrence of more than one phase. Only a single correlation was seen for the resonance at δ exp = 12.8 ppm, corresponding to a proximity to a proton with a chemical shift in the CH region, as    Table S3). The absence of the NH 2 correlation is likely due to the difference in build-up of the recoupled signal between nitrogen environments, with the NH 2 signal already decaying before the NH signal reaches its maximum.
A 2D 1 H-13 C HETCOR MAS NMR spectrum of 52AMP:F is shown in Fig. 4, recorded using a CP contact time of 500 μs such that cross-peaks for longer-range C···H proximities are apparent as well as direct one-bond C-H connectivities. The 1 H-13 C HETCOR spectrum is shown together with crosses that represent the GIPAW calculated chemical shifts for the C-H dipolar correlations up to 3.3 Å (see Table S4 in the SI). More correlations with the methyl protons are present experimentally than expected for this cut off distance, with small cross-peaks apparent for C3 (δ exp = 132.5 ppm, δ calc = 135. The aforementioned disparities between experiment and calculation for the 13 C chemical shifts mean that C10 and C3 are seen experimentally in the opposite order to which they are calculated. Their calculated 13 C chemical shifts lie only 0.4 ppm apart with C10 at the lower chemical shift. The assignment of C10 to the resonance at δ exp = 138.2 ppm is, however, confirmed by its cross-peak with H12 at δ exp = 15.0 ppm, with a C10-H12 proximity of 2.08 Å compared to a distance of 3.68 Å for C3- respectively. The absence of the C3-H3 cross-peak is expected as not only is C3 quaternary, but this corresponding distance is 3.29 Å, on the limit of what is seen for any other carbon environment. The absence of the C14-H12 cross-peak is rather more inexplicable as this is a methyl carbon and the C-H separation is 2.60 Å.

Intermolecular interactions
Comparing the GIPAW chemical shifts calculated for the entire crystal structure to those calculated for individual isolated molecules, as extracted from the geometry optimised crystal structure, is useful as significant differences between the two indicate the presence of intermolecular interactions. 34,41,[49][50][51][52] For 1 H, changes are considered significant for δ Cryst-Mol exceeding 1 ppm. The isolated molecule calculations for 52AMP:F identify only the four classical H-bonds ( Table 6) that were assumed from proximities and angles within the crystal structure ( Table 2). The OH···O interaction involving H3 is by far the strongest with δ Cryst-Mol = 11.9 ppm and corresponding to both the shortest distance, at 2.50

Conclusions
The crystallisation of 5-amino-2-methylpyridine with fumaric acid results in a 1:1 salt, 52AMP:F, the structure of which was solved by SXRD. The molecular packing is based upon a hydrogen fumarate acid chain akin to that seen for 2,6-lutidinium hydrogen fumarate. 43 PXRD was utilised to confirm the composition of the bulk prior to analysis by solid-state NMR.
As well as exhibiting similar structural patterns as the 2,6-lutidinium salt, the OH···O interaction was also found to show a comparable discrepancy between the experimental and GIPAW calculated chemical shifts, with the hydrogen bonded proton calculated 1.6 ppm higher than observed experimentally for 52AMP:F (a difference of 1.9 ppm was recorded for 2,6-lutidinium hydrogen fumarate). 44 Excluding this one exceptional proton, there was excellent agreement between GIPAW calculated and the experimentally observed resonances of both 1 H chemical shifts and pyridinium 14 N shift. The errors for the 13 C chemical shifts were slightly higher than expected difference of ~2 ppm, with the largest difference (excluding the methyl carbon which differs due to temperature effects) being The key intermolecular interactions supporting the structure were confirmed, by the isolated molecule GIPAW NMR chemical shift calculations, to consist of only four classical hydrogen bonds expected from analysis of the crystal structure, with no CH donors or π interactions apparent. The Hbond strength, according to the change in calculated chemical shift between crystal and isolated molecule, matched that expected from the hydrogen bond parameters. The relative strength of the two weaker NH···O hydrogen bonding interactions, formed by each of the two NH 2 protons, also implies that a slight lengthening of the hydrogen bond has a larger impact on hydrogen bonding strength than a small change in NH…O angle further from linearity.