Hannah E.
Kerr
a,
Lorna K.
Softley
a,
Kuthuru
Suresh
c,
Ashwini
Nangia
*c,
Paul
Hodgkinson
*a and
Ivana Radosavljevic
Evans
*ab
aDepartment of Chemistry, Durham University, Science Site, Durham DH1 3LE, UK. E-mail: paul.hodgkinson@durham.ac.uk; ivana.radosavljevic@durham.ac.uk
bBragg Institute, Australian Nuclear Science and Technology Organisation, New Illawarra Road, Lucas Heights, NSW, Australia
cSchool of Chemistry, University of Hyderabad, Central University P.O., Prof. C. R. Rao Road, Hyderabad 500 046, India. E-mail: ashwini.nangia@gmail.com
First published on 30th July 2015
Furosemide is a loop diuretic drug marketed in solid form which suffers from low solubility and low permeability. The pharmaceutically relevant properties of a recently described furosemide–isonicotinamide 2:
1 cocrystal (2FS–INA) were investigated and compared with those of other known furosemide cocrystals. The intrinsic dissolution rate of 2FS–INA was found to be very similar to that of commercial FS, while its equilibrium solubility was 5.6 times higher than that of pure FS. The extensive structural disorder in 2FS–INA observed by diffraction methods was also investigated by variable-temperature solid-state NMR in conjunction with first principles calculations. 15N NMR confirmed the absence of proton disorder in the short OH⋯N hydrogen bond. The disordered sulphonamide group was found to be dynamic by variable temperature 2H experiments, involving fast exchange of the sulphonamide NH2 protons combined with a rotation of the whole sulphonamide group about the C–S bond. The disorder of the furan rings of both the unique furosemide molecules was also found to be dynamic by 13C experiments, with approximately the same activation barrier for both rings.
There have been numerous successful attempts to improve the properties of FS. Mesoporous materials, such as SBA-15 containing silica walls with large pores, have been used as drug delivery systems to improve the release of FS in site-specific areas. The large surface area resulting from the pores allow facile adsorption of FS onto its surface via weak interactions (allowing rapid drug release), while the silanol groups provide hydrophilic character to carry and deliver the drug in aqueous media.5 Alternatively, surfactant-based methods have been used in self-(micro)emulsifying formulations which contain a mixture of oil, surfactant and hydrophilic cofactant. The formulation is able to form an oil/water emulsion in the stomach or intestine. A solid form of the formulation, loaded with FS, has been seen to improve both permeability and solubility of the drug.6 Most recently, improved solubility has been achieved in Na and K salts of FS.7 Finally, numerous cocrystals of FS with a range of coformers, as well as cocrystal polymorphs, hydrates and solvates have been reported.3,8–12
The FS molecule contains potential hydrogen bond donor and acceptor groups: COOH, NH and SO2NH2, and additional potential for halogen bonding from chlorine. The groups COOH and SO2NH2 are known to give robust heterosynthons via O–H⋯O and N–H⋯O hydrogen bonds, forming a wide range of stable supramolecular structures. The propensity for hydrogen bonding enables FS to exist in different polymorphic forms, of which three have been reported.13 Polymorphism is a result of the combination of functional groups to form different synthons, as well as the conformational flexibility about the sulphonamide and furan torsion angles. Higher dissolution rates were observed with the metastable polymorphs, but their use as a marketable drug is limited by their tendency to convert to the stable polymorph during experiments or storage.13
To date, six FS cocrystals (not including hydrates or solvates) with a range of coformers have been prepared and fully structurally characterised.3,8,12 In all cases studied, solubility was improved over pure FS. The only crystal form for which the pharmacologically relevant properties have not been reported is the 2FS–INA cocrystal, which is reported here.
The coformers and synthons relevant to the FS cocrystal structures are summarised in Fig. 2 and Table 1. FS–CAF and FS–CYT crystallise in the triclinic space group P. FS–CYT is distinct from the other structures due to the occurrence of proton transfer in synthon N to form the strong, electrostatic hydrogen bond COO−⋯+HN, and so is more appropriately described as a salt. There is no proton transfer in FS–CAF; hierarchical hydrogen bonding means that synthon E is the strongest interaction. The three pyridine carboxamide-based cocrystals (FS–NA, FS–INA and 2FS–INA) all crystallise in the monoclinic space group P21/n, but have distinctly different structures, despite sharing the same functional groups for potential hydrogen bonding. Synthon E is common to all three cocrystals, but the remaining synthons differ as a result of isomeric change of the coformer or stoichiometry. The molecules in FS–PABA form homodimers via synthon A, which are connected by a weak interaction via synthon K.
![]() | ||
Fig. 2 The coformers in structurally characterised FS cocrystals and the synthons present in FS cocrystals. |
The structural flexibility to which FS owes its rich supramolecular chemistry can also manifest itself as crystallographic disorder, as evidenced in a number of the known solid forms. The crystal structure of FS form II possesses one unique FS molecule with the furan ring disordered over two positions (differing by an angle of 172° between the ring planes) with 0.75:
0.25 occupancies at 298 K. The structure of form I was initially solved with similar disorder, but further analysis revealed the true structure to have a doubled unit cell containing two FS molecules that differed only in the furan ring orientation.13 Several cocrystals of FS also show ring disorder.14–16 For example, the structure of FS–CAF was solved at 100 K with the ring disordered over two positions (differing by an angle of 175° between the ring planes) with equal occupancy.15 Disorder of the whole SO2NH2 sulphonamide group is not commonly observed in FS crystal forms, presumably because this group is usually involved in hydrogen bonding networks, as evidenced by a systematic study of aromatic primary sulphonamide pyridine-N-oxides synthons.14 It has been noted, however, that the NH2 of the sulphonamide can adopt different orientations through rotations about the S–N bond in FS13 and FS-containing cocrystals,9 indicating that there is conformational flexibility.
The recently reported 2FS–INA cocrystal exhibits a particularly complex disorder, which affects both crystallographically unique FS molecules (FS1 and FS2, see Fig. 3), both the sulphonamide group and the furan ring, and potentially a short strong hydrogen bond.8 This single crystal X-ray diffraction study was carried out at 120 K. The sulphonamide group on FS1 is disordered over two positions with 0.51(1):
0.49(1) occupancies, differing by a 32° rotation about the C–S bond. The furan ring on FS2 was modelled using split atomic sites, as disordered over two positions with 0.67(2)
:
0.33(2) occupancies. The furan ring on FS1 could be modelled satisfactorily using single atomic sites, albeit with the anisotropic atomic displacement parameters somewhat elongated in a direction perpendicular to the ring. A further interesting feature of the 2FS–INA structure is the presence of a short strong hydrogen bond (SSHB) between the COOH on FS1 and the ring nitrogen of INA (donor acceptor distance, rOH–N = 2.546 Å), similar in length to the NH⋯O SSHB found in 3,5-pyridinedicarboxylic acid, which showed temperature-dependent proton migration across the SSHB,17,18 demonstrating the potential for proton dynamics in 2FS–INA.
![]() | ||
Fig. 3 The asymmetric unit of 2FS–INA with the major occupied furan ring and sulphonamide positions drawn as free atoms. The SSHB is shown by the dashed red line and with hydrogen atoms are removed for clarity except for those on the carboxylic acids. Thermal ellipsoids are shown at the 50% probability level. FS site numbering follows that of ref. 13 with labels followed by 1 or 2 to distinguish FS1 and FS2. A trailing apostrophe identifies INA sites. |
Previous X-ray diffraction studies provided an average structural picture for 2FS–INA at a single temperature. Solid-state NMR is highly complementary to X-ray diffraction, particularly for characterising polymorphic behaviour and disorder in pharmaceutically relevant materials.19–21 In particular, NMR relaxation times and 2H NMR spectra are very sensitive to dynamic processes.22,23 Here we use 13C, 2H and 15N solid-state NMR, supported by DFT calculations, to elucidate the nature of three different aspects of structural disorder in this furosemide crystal form.
13C spectra were obtained using either direct excitation or cross polarization (CP) with sideband suppression26 at 8 kHz MAS (5 kHz MAS for spectra recorded below −20 °C), with a recycle delay of 1 second in the direct excitation experiment and 10–35 s in the CP experiments, depending on temperature. The contact time was 4 ms, acquiring over 48–180 transients. SPINAL64 with 78 kHz 1H nutation rate was used for heteronuclear decoupling. Spectra were referenced by setting the carbonyl resonance of replacement sample of α-glycine to 176.5 ppm. Dipolar dephasing spectra were acquired with a dephasing delay of 80 μs. T1 relaxation times were measured using a saturation-recovery scheme with 20 saturation pulses separated by a delay of 10 ms at 8 kHz MAS, where the τ delay was varied between 0.05–3 s in 25 steps. The peak heights were picked using the Topspin T1/T2 module and fitted to a single decaying exponential function in QtiPlot.27
1H-13C heteronuclear correlation spectra were recorded using Lee–Goldburg CP with 109 kHz 1H nutation frequency at 9.5 kHz MAS under the following conditions: recycle delay 10 s, contact times of 0.2 ms and 1 ms over 64t1 increments using 128 transients. An FSLG pulse train was used for homonuclear decoupling, while SPINAL64 was used for heteronuclear decoupling during acquisition.
The 15N spectrum was recorded using CP and 8 kHz MAS with the following conditions: 30 s recycle delay and 4 ms contact time acquiring over 8640 transients. SPINAL64 with 55.6 kHz 1H nutation rate was used for heteronuclear decoupling and the spectrum was referenced by setting the signal from an external sample of 15N-labelled α-glycine to −346.8 ppm. An apodization function corresponding to 60 Hz Lorentzian line broadening was applied prior to Fourier transformation.
Selective deuteration of the most labile protons was achieved by storage in a D2O atmosphere for two weeks. The samples were then kept sealed in a rotor to minimise exchange with H2O in air. 2H spectra were recorded using a simple pulse-acquire scheme without decoupling at 10 kHz MAS, with a 1 s recycle delay and acquiring over 100 transients. T1 relaxation times were measured using a saturation-recovery scheme with 20 saturation pulses separated by a delay of 0.2 ms at 10 kHz MAS without decoupling, where the τ delay was varied between 0.0001–2 s in 30 steps. The peak heights were picked and fitted using the Topspin T1/T2 module. Band shape analysis of the spinning sideband manifolds was performed in GSim28/pNMRsim29 by simultaneously fitting the peak linewidths (using a Lorentzian shape function) and quadrupolar parameters for each resolved site. Flat baselines, necessary for fitting, were achieved using the spline baseline correction in Topspin. Spectra were also acquired from a sample that had been deuterated by shaking in D2O for ten minutes followed by storage in a D2O atmosphere, but the cocrystal was unstable in slurry in D2O and the results were not reproducible (Fig. S6 of the ESI†).
![]() | ||
Fig. 4 13C CPTOSS spectrum (top) and direct-excitation spectrum (bottom) of 2FS–INA at 8 kHz MAS with carbon atoms labelled as in Fig. 3. |
Fig. 5 shows the 15N spectrum and a comparison with calculated 15N chemical shifts for the structures in which the CASTEP geometry optimisation transferred H11 across the SSHB to the INA molecule (red) and where H11 was constrained to the position obtained from single crystal X-ray diffraction (green). The calculated shieldings were converted to chemical shifts using δiso = σref − σiso, where σiso is the CASTEP-calculated shielding value and σref was calculated to equate the average calculated shift and average experimental shift39 (σref was −161.7 ppm when H11 was transferred to INA and −169.0 ppm when H11 was fixed). The experimental shifts of the four overlapped peaks were estimated by spectral deconvolution. The experimental shift of N1′ (−116.0 ppm) is a much better match with the predicted shift range when H11 is associated with the FS1 molecule and not when proton transfer had occurred. This shows that H11 is associated with only FS1 and not the INA (supporting the XRD data), nor is the position averaged over fast transfer. The CASTEP-calculated difference in chemical shift upon protonation of the INA molecule is 36.3 ppm. This is within the experimental range found for a cocrystal/salt transition. For example, protonation of INA in a salt with 2,4,6-hydroxybenzoic acid compared to a cocrystal with 4-hydroxy-3-nitrobenzoic acid results in a change in chemical shift of −68.8 ppm, while the 15N shift in a 3,5-dinitrobenzoic acid salt compared to a cocrystal with 4-aminobenzoic acid is 18 ppm.40 Such changes in chemical shift would be expected to be observed in the experimental spectrum.
The 15N spectrum is too heavily overlapped to draw conclusions about the disorder. On the other hand, the deuterium spectrum is less crowded and the disordered sulphonamide group can be investigated explicitly. As shown in Fig. 6, the shape of the deuterium spectra clearly change with temperature, with the bands broadening as the temperature is lowered to −60 °C. The change in linewidth of the disordered sulphonamide peak was particularly notable and indicative of the freezing out of a dynamic process at low temperature (see Fig. S9 of the ESI† for further discussion). The loss in signal intensity on cooling is due to the noted broadening and all changes are reversible. A representative example of the fitting of the spectra to obtain the quadrupolar parameters is given in Fig. S8 of the ESI.†
![]() | ||
Fig. 6 Variable temperature 2H spectra at 10 kHz MAS. The centre-band of each spectrum is shown on the right. Baselines were corrected for this figure using the Whittaker Smoother in MestReNova. |
Table 2 shows that H12 and H11 are fit with rather different CQ values, with the H11 value being markedly low. Chiba, and others, have reported that reducing the donor–acceptor distance (rD–A) in a hydrogen bond lowers the CQ parameter,41–44 which correlates with the finding that CQ(H11) < CQ(H12); rD–A(H11) is 2.546 Å while rD–A(H12) is 2.619 Å by single crystal XRD. Most studies of HBs in molecular solids typically include a variety of types of hydrogen bonds with different strengths, which limits the use of the data in determining quantitative trends in CQ with structural parameters.43 While quadrupolar parameters have been quantitatively correlated with hydrogen bond geometry individual cases, such as an isolated molecule containing a CO⋯HO hydrogen bond, in which η was shown to be more sensitive to bond angles than rD–A,45 these are not directly relevant to H11 which is in an N⋯HO hydrogen bond.
Deuterium site | −50 °C | 20 °C | 40 °C | CASTEP averageb | ||||
---|---|---|---|---|---|---|---|---|
C Q/kHz | η | C Q/kHz | η | C Q/kHz | η | C Q/kHz | η | |
a One-standard deviation uncertainties are given in parentheses, although these are likely to be underestimated. The −60 °C spectrum could not be fitted satisfactorily due to the broadening of the sulphonamide and H12 peaks. b Full details of CASTEP-calculated values are given in Table S4 in the ESI. c Structures showing proton transfer of H11 are not included in the CASTEP average. | ||||||||
Sulphonamide | 88.4(6) | 0.51(2) | 87.1(2) | 0.417(6) | 87.8(2) | 0.421(7) | 222 | 0.16 |
H12 | 170(2) | 0.22(5) | 161(2) | 0.34(4) | 168(2) | 0.26(5) | 138 | 0.16 |
H11c | 68.8(5) | 0.39(3) | 78.9(5) | 0.34(3) | 81.0(9) | 0.32(5) | 48 | 0.52 |
The CASTEP-calculated CQ and η values for H12 and H11 are comparable with the experimental fits. However, the calculated CQ of the disordered sulphonamide site is around 2.5 times larger than the values determined from the −50 °C data. This can only be explained in terms of extensive dynamics. NH2 dynamics have previously been studied in pure FS by monitoring 1H relaxation times and were described in terms of hindered NH2 jumps that were rapid at ambient temperature.46 Estimating the underlying quadrupolar parameters from the average of the CASTEP-calculated parameters, i.e. CQ = 222 kHz, η = 0.16, and averaging the tensor over a 112° jump (corresponding to exchange of the two hydrogens of the NH2) gives values which are still too high, CQ = 112 kHz, η = 0.66. Values that are more consistent with the experimental results are obtained if the tensor is averaged successively over two jump motions. As an illustrative example, jumps of both 112° and 40° give CQ = 89 kHz, η = 0.57 and jumps of both 112° and 30° give CQ = 100 kHz, η = 0.61. This is consistent with the magnitude of the angle measured crystallographically between the two disordered sulphonamide positions. The quadrupolar parameters are also affected by high frequency librational motions,47 and the actual atomic motion at ambient temperature will be more complex than instantaneous jumps between discrete sites. We can still conclude that the observed deuterium quadrupolar parameters are consistent with rapid rotations between the two orientations of the sulphonamide observed by XRD plus exchange of the hydrogen sites on the NH2, both at a relatively high frequency (at least 100 s kHz at ambient temperature). It should also be noted that the dynamic sulphonamide groups are relatively close to each other (rNH2−NH2 between sulphonamide groups in two orientations of FS1 = 2.452 Å), and so the motion may be further complicated by correlation between the orientations.
Complementary information is provided from the temperature dependence of the 2H T1 relaxation times. The relaxation data was fitted (see Fig. 7) to direct calculations of the 2H relaxation rate under MAS using a simple two-site jump model48 and the CASTEP-calculated CQ of 222 kHz. Using an Arrhenius-type model for the temperature dependence of the jump correlation time with temperature, τc = τ0exp(Ea/kT), the fit shown corresponds to log10(τ0/s) = −12.0 ± 0.1, Ea = 16.7 ± 0.7 kJ mol−1 and a jump angle of 26.5 ± 0.5°. The one standard deviation errors bars are based on an averaged uncertainty of the fitted T1 values of 0.7 ms. Note that including a correction for sample heating due to MAS results in a slight increase in the fitted Ea (from 14.9 kJ mol−1) and only marginal effect on the other fitting parameters. These values should, however, be treated with caution since the actual molecular motion is likely to be more complex. The non-zero asymmetry of the 2H site was not accounted for since the expressions of Torchia and Szabo49 assume η = 0, but this will not have a significant impact. This fitting corresponds to a τc at 298 K of 86 × 10−11 s, and so an effective jump rate, 1/τc, of 1.2 GHz, consistent with observing an averaged CQ in the 2H spectrum. The fitted activation barrier is somewhat smaller than the activation energy of the hindered NH2 jumps measured in pure FS (31 kJ mol−1),46 but is consistent with essentially the same basic process being involved. The estimated jump rate at −60 °C is 80 MHz and it is likely that the broadening of the 2H spectrum in Fig. 6 at low temperature is associated with freezing out some aspect of the sulphonamide dynamics.
The 13C CPTOSS spectrum in Fig. 4 is too strongly overlapped to provide clear conclusions about the ring disorder, and 2D 1H-13C HETCOR and 1H DQ/SQ experiments were also inconclusive (see ESI† for more details). However, in a direct excitation experiment with a short recycle delay to select only dynamic carbon sites only the carbon atoms of both furan rings are observed (bottom of Fig. 4). This indicates that both furan rings are dynamically disordered.
As anticipated from the direct excitation spectrum, the 13C T1 relaxation times for the furan ring carbon atoms are significantly shorter than those of the other carbon sites (see Table S5 of the ESI† for the full data). Since T1 relaxation of is driven by motion on the order of the NMR frequency of the spins involved, this implies that these carbon sites are dynamic at rates of the order of 125 MHz at ambient temperature. The magnitude and the temperature dependence of the 13C T1 times, Fig. 8, is similar between the two furan rings, suggesting that similar processes are involved. The variable temperature data can be fitted using expressions for dipolar relaxation in solid sample due to rotational diffusion on a cone49 to yield the parameters given in Table 3.
![]() | ||
Fig. 8 The temperature dependence of the 13C T1 relaxation times for the furan rings on FS1 and FS2. Temperatures include a correction of +10 °C for sample heating due to MAS. |
E a/kJ mol−1 | log![]() |
Cone angle/degrees | ||
---|---|---|---|---|
a 13C T1 data for C111 could not be satisfactorily fit due to the low signal intensity. | ||||
FS1 | C121 | 20(2) | −12.7(3) | 19.8(5) |
C101 | 22(1) | −13.3(2) | 21.6(4) | |
FS2 | C122 | 25(1) | −13.7(2) | 31.9(9) |
C102/C112 | 22(1) | −13.2(2) | 21.3(4) |
Given the simple model used, the agreement between motional parameters for the carbon atoms of the same ring is quite satisfactory. The differences in parameters between the rings are generally not statistically significant, although the cone angle for C12 is arguably larger (corresponding to a greater motional amplitude) in FS2 compared to FS1.
In the structure determination from single crystal X-ray diffraction, resolved electron density in the difference Fourier map prompted the modelling of the FS2 ring using split atomic sites and isotropic thermal parameters, while the FS1 furan ring was modelled with single atomic sites and anisotropic atomic displacement parameters (ADPs). As seen in Fig. 3, the resulting ADPs are slightly elongated along one axis, suggesting the presence of small amplitude dynamics. It is important to note, however, that an essentially equally good fit to the experimental XRD data can be obtained using single sites and anisotropic ADPs to describe both furan rings, as shown in Fig. S11.† In this fit, the ring disorder on FS1 and FS2 appears very similar, with only a subtle difference in amplitude. In other words, in this case a single-temperature XRD experiment was unable to distinguish between small and large amplitude re-orientations of the ring. As discussed above, the calculated 13C chemical shifts are also not particularly sensitive to the orientation of the ring, and it is the relaxation data that strongly suggests that both FS1 and FS2 are likely to be undergoing similar small amplitude motion.
Three specific questions concerning the extensive structural disorder in the 2FS–INA cocrystal, posed by the previous single crystal low-temperature X-ray diffraction study, have been investigated using variable-temperature solid-state nuclear magnetic resonance spectroscopy supported by first principles calculations: the nature of the disorder of the sulphonamide group, the furan ring, and the potential proton disorder in a short strong hydrogen bond.
No evidence for dynamic proton transfer in the short OH⋯N hydrogen bond was found. The disorder on the sulphonamide group could be readily probed using variable temperature 2H experiments and was found to be consistent with fast exchange of the NH2 protons as well as a rotation of the whole sulphonamide group about the C–S bond. The empirical activation energy for this process is estimated to be 17 kJ mol−1, although the motion is expected to be more complex than a simple two-site jump. A direct excitation 13C spectrum showed that the carbon atoms on the furan rings of both unique FS molecules had short relaxation times, and 13C spin–lattice relaxation times measured as a function of temperature provided estimates for the energy barriers to rotation of the two furan rings, which were found to be the same within experimental uncertainty.
In general terms, this work illustrates the complementarity of diffraction and NMR in studying structural disorder. While disorder can be readily observed by diffraction methods and in certain cases some insight into its nature can be gained from careful variable-temperature experiments analysed using the translation-libration-screw (TLS) formalism,50 the quality of diffraction data generally deteriorates as the temperature is raised. In contrast, dynamic disorder at ambient temperatures often simplifies NMR spectra by sharpening lines and shortening relaxation times. Hence NMR aided by DFT calculations provides a straightforward way to establish the nature of disorder and provide quantitative information on the relevant processes.
Footnote |
† Electronic supplementary information (ESI) available: Containing supporting ssNMR experiments as detailed in the text. Details of the raw data and fitting codes are available, see DOI: 10.15128/br86b3677. See DOI: 10.1039/c5ce01183c |
This journal is © The Royal Society of Chemistry 2015 |