Edinburgh Research Explorer High-pressure polymorphism in L-threonine between ambient pressure and 22 GPa

The crystal structure of L-threonine has been studied to a maximum pressure of 22.3 GPa using single-crystal X-ray and neutron powder diffraction. The data have been interpreted in the light of previous Raman spectroscopic data by Holanda et al. ( J. Mol. Struct. (2015), 1092 , 160-165) in which it is suggested that three phase transitions occur at ca. 2 GPa, between 8.2 and 9.2 GPa and between 14.0 and 15.5 GPa. In the first two of these transitions the crystal retains its P 2 1 2 1 2 1 symmetry, in the third, although the unit cell dimensions are similar either side of the transition, the space group symmetry drops to P 2 1 . The ambient pressure form is labelled phase I, with the successive high-pressure forms designated I’, II and III, respectively. Phases I and I’ are very similar, the transition being manifested by a slight rotation of the carboxylate group. Phase II, which was found to form between 8.5 and 9.2 GPa, follows the gradual transformation of a long-range electrostatic contact becoming a hydrogen bond between 2.0 and 8.5 GPa, so that the transformation reflects a change in the way the structure accommodates compression rather than a gross change of structure. Phase III, which was found to form above 18.2 GPa in this work, is characterised by the bifurcation of a hydroxyl group in half of the molecules in the unit cell. Density functional theory (DFT) geometry optimisations were used to validate high-pressure structural models and PIXEL crystal lattice and intermolecular interaction energies are used to explain phase stabilities in terms of the intermolecular interactions. peptides and complexes in the CSD (v5.40 November 2018) there are 2105 individual hydrogen bond lengths (ca. 5 %) which are equal to or shorter than the shortest hydrogen bond length in L-threonine at 22.3 GPa. This is in spite of a reduction in unit cell volume by over one-third. Hydrogen bonds are the strongest and most consistently-formed intermolecular interactions in organic crystal structures, and this study reveals that their robustness persists well above 10 GPa, even in relatively complex molecular crystals.


Introduction
The amino acids have been studied extensively in the context of high-pressure polymorphism because they serve as model systems for the behaviour of H-bonding in other, potentially more complex molecules, but also in part because of their biological significance. 1, 2 Glycine, the simplest amino acid, has three ambient pressure polymorphs which show differing stabilities under compression. α-Glycine has been shown to be stable to 23 GPa by Raman spectroscopy. 3 The crystal structure has been determined by X-ray single-crystal and neutron powder diffraction at 6.2 4 and 6.4 GPa, 5 respectively. Very recent angle-dispersive X-ray diffraction measurements show that the phase persists to 50 GPa. 6 By contrast, β-glycine transforms to δ-glycine at 0.8 GPa, while the γ-form gradually yields the ε-polymorph between 2.0 and 4.3 GPa. 4, 7 ε-Glycine transforms back to the γ-form via a, sixth, short-lived ζ-polymorph. 8 L-serine has four high-pressure polymorphs. [9][10][11][12][13][14] The ambient pressure form, L-serine I, transforms to L-serine II and III on rapid compression at ca. 5 and 8 GPa and to Lserine IV above 5.6 GPa on slow compression. 15 L-cysteine I transforms on compression above 1.8 GPa to give L-cysteine III and then again on decompression from 4.2 to 1.7 GPa to form Lcysteine IV. 16 Structural changes in L-α-glutamine were studied to 4.9 GPa but it does not undergo any phase transitions. 17 L-alanine also remains in the same ambient pressure phase on compression to 13.6 GPa but it undergoes reversible amorphisation at 15 GPa. 18,19 The role of H-bonding and other non-bonding interactions are important in understanding phase stability as a structure evolves under compression to fill space more efficiently and avoid short repulsive contacts. 20 Crystallographic studies of complex molecular materials, however, rarely achieve pressures beyond 10 GPa. The Cambridge Structural Database (CSD) (v.4.0, November 2018) contains almost 3000 entries which specify the pressure of the structure determination, and a list of these is included in Microsoft Excel format in the ESI. There are 2561 entries determined at above 10 atm, and 2457 above 1000 atm (0.1 GPa). Only 14 molecular compounds have been studied above 10 GPa (100 000 atm), only one of which is an amino acid, L-alanine (LALNIN51), whose structure was determined by powder methods at 13.6 GPa (see Table S1, ESI). 18 The highest pressure entry is that of CO2 at 28 GPa (SACBAA) which was also obtained by powder diffraction; 21 the next highest pressure entry is that of benzene (BENZEN09) at 24 GPa but the entry lacks 3D coordinates. 22 The highest pressure entry in the CSD of a complex molecular compound, determined by single-crystal diffraction and with refined 3D coordinates is that of palladium (II) oxathioether (NONWES30) at ca. 14 GPa. 23 A search of the Inorganic Crystal Structure Database (ICSD) for non-metal compounds at pressures greater than or equal to 10 GPa results in 187 hits. Amongst the molecular elements, [24][25][26] oxygen becomes metallic at 96 GPa. 27 The structure of a high-pressure phase of molecular nitrogen has been determined by single-crystal X-ray diffraction at 56 GPa, 25 and the structure of polymeric nitrogen, which forms at 110 GPa and 2000 K, has been determined using X-ray diffraction and Raman scattering at 115 GPa. 28 Beyond the elements, there are 84 crystal structures of molecular solids at pressures greater than 10 GPa deposited in the ICSD, comprising 15 different compounds. Some notable examples include the single-crystal structure determinations of arsenolite and its helium clathrate to 30 GPa 29 and the van der Waals compound Kr(H2)4 whose structure was determined at 11.24 GPa. 30 Xenon difluoride has been studied by powder X-ray diffraction and computational methods and is shown to undergo two phase transitions at 28 and 59 GPa, with metallisation predicted to occur at 152 GPa. 31 Although crystal structures of complex molecular solids above 10 GPa are quite rare, spectroscopic methods, particularly Raman spectroscopy, have been very useful in detecting phase transitions. This approach has been used extensively to study amino acids. 32 We now describe the crystal structure of L-threonine ( Figure 1

X-Ray Crystallography
L-threonine (99% Sigma-Aldrich) was recrystallised from a 656 mM ethanol solution by slow evaporation; forming colourless, blade-shaped crystals. Single-crystal diffraction data were collected at ambient pressure and room temperature using a cut crystal measuring 0.2 x 0.2 x 0.1 mm³ on a Bruker 3-circle goniometer APEX-II diffractometer using Mo-Kα radiation (λ=0.71073 Å).
Below 7 GPa diffraction data were measured at room temperature using silicon (111) monochromated synchrotron radiation (λ=0.4780 Å) on Station 9.8 at the Synchrotron Radiation Source, Daresbury, UK. A single crystal of L-threonine measuring ca. 0.1 mm³ was loaded in a Merrill-Bassett type diamond anvil cell (DACs) along with a ruby chip in 4:1 methanol-ethanol as a pressure-transmitting medium. 34,35 A total of six diffraction measurements were carried out between 1.26 to 6.67 GPa using a Bruker-Nonius APEX II diffractometer following the collection strategy of Dawson et al. 36 Single-crystal diffraction data between 4.0 and 22.3 GPa were measured at room temperature on Beamline 12.2.2 at the Advanced Light Source in Berkeley, California, USA, which has been described in detail elsewhere. 37,38 Crystals measuring ca. 50 µm³ were cut from larger single crystals and mounted with a ruby sphere in a BX-90 type DAC 39 consisting of 500 µm Boehler-Almax cut diamonds mounted in tungsten-carbide backing seats. 40 The rhenium gasket hole had an initial diameter 320 µm and thickness of ca. 70 µm. The cell was gas-loaded in neon using a GSECARS/COMPRES gas-loader 41 at the Advanced Light Source.
Data were collected in steps of approximately 1.4 GPa; and on decompression at 13.0 GPa on a custom-built Huber diffractometer with silicon (111) monochromated synchrotron radiation (λ=0.4959 Å) and a Perkin-Elmer amorphous silicon detector, using a combination of shutterless ϕ-scans at 0.25° and 1° step-widths across the half-opening angle (±40°) of the sample chamber and cell body. Additional low-pressure measurements were performed in the same manner from 2.6 to 5.9 GPa on a separate sample.
In all cases, pressure was measured using the ruby fluorescence method. 42

Structure analysis
Diffraction data were processed using the APEX3 suite of programs. 43 Data reduction was carried out using SAINT, 44 employing dynamic masks generated by ECLIPSE 45 to mask shaded detector areas. Absorption and shading corrections were applied using the multi-scan procedure SADABS. 45 Data-sets were initially analysed using XPREP, 46 the structures at ambient pressure and following a phase transition at 18.2 GPa were solved using direct methods (SHELXT) 47 and then refined by full-matrix least-squares on |F| 2 (SHELXL) 48 using the ShelXLe graphical user interface. 49 Otherwise refinements started from the atomic coordinates of the preceding pressure point.
There is a single-crystal to single-crystal phase transition at 18.2 GPa which reduces the space group symmetry from P212121, to P21, though the unit cell metrics are similar either side of the transition. The structure was modelled with a two-fold axis about a as a twin law, but the twin fraction refined to 0.05 (5). The orthorhombic cells of the structure below 18.2 GPa were placed in a non-standard setting to match that of the monoclinic phase in order to facilitate comparisons between phases.
Intramolecular bond distances in all high-pressure refinements were restrained to those of the ambient pressure structure. Data sets were modelled with isotropic displacement parameters in order to reduce the number of refined parameters. H-atoms were placed in calculated positions and allowed to ride on their parent atoms. The hydroxyl hydrogen atom was placed on the site forming the most favourable H-bond geometry while also being staggered with respect to O3-C3. The refinement and H-atom placement strategies are discussed below. Selected crystal and refinement data of structures in the different phases are collected in Table 1

Neutron Powder Diffraction at High Pressure
In order to improve the precision of the equation of state (EOS) of L-threonine and to corroborate the X-ray measurements, a series of neutron powder diffraction measurements was collected on compression from 0.00 to 8.77 GPa in steps of ca. 0.55 GPa using the PEARL instrument at the ISIS facility, Rutherford Appleton Labs, Didcot, UK. Deuterated L-threonine (CDN Isotopes, used as received) was loaded into a TiZr capsule gasket with a 4:1 mixture of deuterated methanol and ethanol, and a pellet of lead as a pressure marker. The sample was compressed using a type V3 Paris-Edinburgh cell with WC type anvils. Pressure measurements were obtained from the equation of state of lead. 50 Unit cell parameters were extracted by the Pawley method using TOPAS. 51

PIXEL Energy Calculations
The PIXEL method is a semi-empirical computational technique for the calculation of In this study the cluster radius was 15 Å, and the molecular electron densities were obtained from GAUSSIAN-09 57 with the 6-31G** basis set at the MP2 level of theory. The PIXEL calculations themselves were accomplished with the CLP-PIXEL suite. 54 The electron densities were calculated on a grid of dimensions 0.08 x 0.08 x 0.08 Å 3 , but in order to speed up subsequent energy calculations, blocks of 4 x 4 x 4 pixels were combined into superpixels (i.e. the condensation level was 4).
Individual intermolecular interaction energies obtained using PIXEL (and symmetryadapted perturbation theory at the SAPT2+3 level, see below) are shown in Table 2 (Tables S5 and S6) and plots showing dimers formed within the first coordination spheres at 0, 17.1 and 18.2 GPa in Figs. S1-4)

Periodic Density Functional Theory (DFT) Calculations
Geometry optimisations were carried-out on the ambient pressure crystal structure as well as those at 3.97, 5.91, 6.12, 8.50, 9.82, 11.19, 13.94, 15.20, 15.78, 17.05, 18.20, 20.62 and 22.31 GPa, using the plane-wave pseudopotential method in the CASTEP 58 code as incorporated in Materials Studio. 59 The PBE exchange-correlation functional was used with the 'on the fly' pseudopotentials embedded in the program and the Tkatchenko-Scheffler correction for dispersion. 60 The basis set cut-off energy was 780 eV and Brillouin zone integrations were performed with a Monkhorst-Pack 61 k-point grid spacing of 0.07 Å -1 . These conditions gave a convergence in total energy of better than 1 meV/atom.
The starting coordinates for the optimisations were taken from the single-crystal X-ray diffraction structures with distances to hydrogen normalised to typical neutron values. The cell dimensions were fixed to the experimental values, and the space group symmetry was retained. In geometry optimisations the energy convergence criterion was 5x10 -6 eV/atom, with a maximum force tolerance of 0.01 eV Å -1 and a maximum displacement of 5x10 -4 Å; the SCF convergence criterion was 1x10 -8 eV/atom.

Symmetry-adapted Perturbation Theory (SAPT) Calculations
SAPT calculations were performed at the SAPT2+3 level of theory on dimers taken from the ambient pressure and 17.1 GPa structures using the PSI4 code (version 1.0.54) with the aug-cc-pvdz basis set. 62, 63

Other Programs Used
Geometric parameters were calculated using PLATON. 64 EoSFit7-GUI 65

Validation of computational methods
In The pressure-dependence of the lattice energy of L-threonine was calculated using the PIXEL method and periodic DFT (see Section 3.4.1 and Table S4 in the ESI). The goodness-offit of the PIXEL calculated energies from ambient pressure to 22.3 GPa is 0.85, which improves to 0.93 if the highest-pressure structure is discounted.
Individual intermolecular interaction energies obtained using PIXEL and symmetryadapted perturbation theory at the SAPT2+3 level are shown in Table 2

Phase behaviour
The X-ray crystal structure of L-threonine was first determined by Shoemaker in 1950. 74 It exists as a zwitterion in the solid-state with charged carboxylate and ammonium groups  It will be seen from Figure 2

The Effects of Pressure on the Intramolecular Structure
High pressure data sets almost always suffer from low completeness as a result of the limited scattering geometry of the diamond anvil cell. Accordingly, the data-sets of threonine-I, I' and II collected here had completenesses of between 43 and 73 %, the corresponding figures for the lower symmetry phase III were 32-45 %. As a result, it is usually necessary to place restraints on the structure refinements, and bond distances and angles were restrained to values observed at ambient pressure.
In order to assess the suitability of the restraints applied, the structures of phase I at ambient pressure and phase III at 22.3 GPa were optimised by periodic DFT. The calculations indicated that bond distances change by as much as 0.071 Å, and bond angles by 5.43° (see Table S7 in the ESI), while the root-mean-square deviations between the optimised and experimental molecular structures were 0.0284 and 0.131 Å, respectively (see Tables S9 and   S10 in the ESI, which also contains further comparison data). Use of the optimised model as a freely rotating rigid body in the refinement against the 22.3 GPa data set lowered the Rfactor, but only slightly (0.17 % for 183 reflections), so that it is not possible to state definitively whether the intramolecular bond distances and angles are significantly affected by pressure, as the differences are beyond the resolution of the data obtained in this study.
By contrast, the torsion angles do vary significantly with pressure. Figure 3 shows differences in conformation between the ambient pressure and highest-pressure molecular structures, with differences apparent about the carboxylate, hydroxyl and methyl groups.

Intermolecular interactions at ambient pressure
Intermolecular interaction energies in phase I at ambient pressure are listed in Table 2, where contacts are ordered by interaction energy and grouped in symmetry-equivalent pairs. Validation of the PIXEL results against those of other methods is described in Section 2.8.
Given that the PIXEL method is semi-empirical and developed using ambient-pressure structural and thermodynamic data, the level of agreement with other methods is remarkable.
The strongest contacts comprising four pairs of crystallographically unique The third dimension of the H-bond network is completed by C(6) chains formed by O3H6···O1 hydrogen bonds between the carboxylate and hydroxyl groups along b, see Figure   6 [interaction C/D in Table 2, the hydrogen-acceptor distance is 1.87 Å and the energy is −55.1 kJ mol −1 ]. This is the second most stabilising interaction in the ambient pressure structure. The Coulombic interaction is almost as strong as in the A/B contacts (−82.6 versus −93.9 kJ mol −1 ), but the repulsion term is also much more positive (+71.8 versus 38.7 kJ mol −1 ), so that, paradoxically, the interaction with the shorter H-bond is also the less stabilising.
Interactions E and F are strongly stabilising, with an energy of −53.0 kJ mol −1 but featuring a rather long N1H3···O1 contact of 2.65 Å with a <N1H3···O1 angle of only 110.4°.
This angle is too tight for H-bond, 80

and the component energies show that the interaction is
predominantly Coulombic with a much smaller dispersion contribution than the hydrogen bonds described above. The contact is therefore better regarded as a long-range intra-layer electrostatic contact than a hydrogen bond. This interaction has an important influence on the compression of L-threonine, as described in the following section. The hydrogen bonding scheme described above appears to leave the potentially strong donor N1H2 unbound. There is an additional C(5) chain connected by N1H2···O3 interactions (labelled K and L in Table 2) though the DH···X separation is quite long at 2.31 Å.

The effect of pressure on the lattice energy
The effect of pressure on the lattice energy of L-threonine is shown in Figure 7, the points being calculated using the PIXEL method and periodic DFT.
In phases I, I' and II the lattice energies increase steadily with increasing pressure, which is expected as repulsion contributions increase as molecules are forced into close proximity. There is a discontinuity in the gradient at the II to III phase transition at 18. Based on the trends seen in Figure 7, the lattice energy of phase II at 18.2 GPa would be expected to be approximately −126 kJ mol −1 , compared to −124.9 kJ mol −1 for the observed phase III. Although phase III is less stable than phase II in terms of internal energy, at 18.2 GPa, the difference in molecular volume is 2.3 Å 3 . This figure, though apparently modest, contributes a pΔV term of −25 kJ mol −1 to the free energy change of the transition, which more than compensates for the change in lattice energy. Like most high-pressure phase transitions, therefore, the II to III transition in threonine is driven by the need to fill space efficiently at high pressure.

The effect of pressure on intermolecular interactions in the ac planes
An animation showing the path of compression of the layers formed in the ac planes is available in the ESI, while plots of the selected interaction distances against pressure are given in Figures 9 (a)  simultaneously more negative, the only stabilising change to occur on compression, providing a lower energy pathway for the structure to accommodate compression. Since interactions E and F are generated by a screw axis along c, the c-axis compresses substantially more than either the a or b-axes (Figure 2(a)). The gradual bifurcation of H3 reflects Holanda and coworkers' comment that the I' to II transition is the final result of a long process with affects the carboxylate rocking motion from about ca. 4.8 GPa. 33 Crystallographically, we can see that by this stage the H3···O1 distance is already around 2.3 Å.
While the principal structural effect of the I' to II transition is seen in interactions E/F, effects are also seen in the other interactions formed in the ac plane. The H-bonded dimers A/B are also slightly stabilised by the bifurcation of H3 while in phase I', but the energy begins to increase rapidly after the I' to II phase transition. Similar comments apply to interactions G/H, which actually becomes repulsive in phase III.
The lengths of the a and c axes undergo more rapid compression after the II to III transition. The animation of the pressure series shows that the loss of the 21 axes along a and c, which occurs during the transition, enables neighbouring rows of molecules linked by N1H3···O2 to shift in alternate directions enabling them to approach more closely, and the molecules to pack more efficiently. The lowering of symmetry illustrates Dove's comment that high-pressure phase transitions generally favour distorted phases in order to maximise density. 81

Interactions between the ac planes
An animation showing the path of compression of the layers formed in the ab planes is available in the ESI. The layers which form parallel to the ac planes are connected by O3H6···O1 H-bonds between the hydroxyl and carboxylate groups (interactions C/D), and a plot of the interaction distance against pressure is shown in Figure 9 (c).
The C/D interaction energy does not show any initial stabilisation with pressure, in contrast to interactions A/B and E/F, a consequence, perhaps, of the repulsion term, which is appreciable even at ambient pressure. The energy becomes more positive as the pressure increases and does not exhibit any major discontinuities at the phase transitions.

CONCLUSIONS
The crystal structure of L-threonine has been determined up to 22.3 GPa, one of the highest pressures ever achieved for a complex molecular material, providing amongst the most detailed information beyond 10 GPa ever published for this type of system.
The structure undergoes two isosymmetric phase transitions on compression between 0.0 and 17.1 GPa. The ambient pressure phase I transforms to phase I' between 2.1 and 3.0 GPa, the result of a subtle reorientation of the carboxylate group. The transformation of phase I' to phase II between 8.5 and 9.8 GPa follows a gradual transformation of a longrange Coulombic interaction into one mediated by a bifurcated hydrogen bond. The transition therefore reflects a change in the way the structure absorbs pressure rather than a distinct structural change, and in this regard it is somewhat akin to a second order thermal event such as a glass transition.
Further compression results in the transformation to phase III between 17.1 and 18.2 GPa. The II to III transition is driven by a discontinuous reduction in volume and is characterised by the rotation and bifurcation of the hydroxyl groups in half of the molecules in the unit cell.
The crystallographic results are consistent to those seen in the earlier Raman studies.
Indeed, the interpretation of the crystallographic data has been substantially guided by the Raman results of Holanda et al. 33 While the phase II to III transition involves a discontinuous change in volume and symmetry, phases I, I' and II are all very similar, and the designation of these forms as separate phases is based on discontinuities of the trends seen in Holanda et al.'s Raman spectra with increasing pressure. In particular, the I and I' transition is not at all obvious from examination of the crystallographic data alone, and it would probably have been missed without the Raman data. This said, while Raman spectra have proved to be an extremely sensitive tool for detecting the phase transitions, they are a less definitive guide to the magnitude of the structural changes.
Above 17.1 GPa the crystal structure of L-threonine destabilises rapidly. Holanda's Raman spectra suggest that the material remains crystalline to at least 27 GPa, but the ultimate response of a relatively complex crystal structure such as threonine to very rapid onset of destabilisation is largely unexplored territory. One possibility is amorphisation, as seen for L-alanine at 15 GPa. 18 Cleavage of primary covalent bonds forms another potential route, exemplified by proton transfer in oxalic acid at 5.3 GPa, 82 or even wholescale decomposition into amorphous networks such as is seen for benzene 83,84 and pyridine. 85,86 Although the hydrogen bonds in L-threonine are substantially compressed up to 22.3 GPa, it is remarkable that their distances all find precedents at ambient pressure, and none of them can be described as 'abnormally short' (Figure 10) It has been shown that the strongest intermolecular interactions generally persist across phase transitions. 87,88 However, the drive to reduce volume becomes ever more pressing as pressure is increased, leading to a perturbation in the hierarchy of intermolecular interactions. It will be fascinating to discover the point at which this effect finally wins out and hydrogen bonds give up the ghost.

ACKNOWLEDGEMENTS
This research used resources of the Advanced Light Source, which is a DOE Office of Science

CONFLICTS OF INTEREST
There are no conflicts of interest to declare.