Maleate salts of bedaquiline

The single-crystal structures of several maleate salt of bedaquiline, a drug used for the treatment of drug-resistant tuberculosis (TB), are described.

Bedaquiline is one of two important new drugs for the treatment of drugresistant tuberculosis (TB). It is marketed in the US as its fumarate salt, but only a few salts of bedaquiline have been structurally described so far. We present here five crystal structures of bedaquilinium maleate {systematic name: [4-(6bromo-2-methoxyquinolin-3-yl)-3-hydroxy-3-(naphthalen-1-yl)-4-phenylbutyl]dimethylazanium 3-carboxyprop-2-enoate}, C 32 H 32 BrN 2 O 2 + ÁC 4 H 3 O 4 À , namely, a hemihydrate, a tetrahydrofuran (THF) solvate, a mixed acetone/hexane solvate, an ethyl acetate solvate, and a solvate-free structure obtained from the acetone/hexane solvate by in situ single-crystal-to-single-crystal desolvation. All salts exhibit a 1:1 cation-to-anion ratio, with the anion present as monoanionic hydromaleate and a singly protonated bedaquilinium cation. The maleate exhibits the strong intramolecular hydrogen bond typical for cis-dicarboxylic acid anions. The conformations of the cations and packing interactions in the maleate salts are compared to those of free base bedaquiline and other bedaquilinium salts.

Chemical context
Bedaquiline is one of two important new drugs for the treatment of drug-resistant tuberculosis (TB). It is marketed in the US as the fumarate salt with the trade name SirturoTM (Brigden et al., 2015) and described in US Patent 8,546,428 (Hegyi et al., 2013). A number of other bedaquilinium salts have been reported since the emergence of its pharmacological relevance, but until recently only free base bedaquiline had been fully structurally described (Petit et al., 2007). To fill this gap, which severely hampers understanding of the chemical, physical and physiological properties of bedaquiline and its derivatives, we have recently reported and analyzed the single-crystal structures of several bedaquilinium salts, including that of the commercially traded fumarate as well as two differently solvated benzoate salts (Okezue et al., 2020). This study revealed that bedaquiline and its salts have a very rich and diverse structural chemistry. Depending on the nature of the anion (fumarate, benzoate or none for free base bedaquiline), different molecular conformations and structural motifs are observed. In free base bedaquiline, the amine moiety is engaged in an intramolecular O-HÁ Á ÁN hydrogen bond, limiting the formation of intermolecular hydrogenbonding interactions. The packing is instead dominated by weaker and less directional interactions such as BrÁ Á ÁBr ISSN 2056-9890 interactions and -stacking (Petit et al., 2007). In the fumarate and benzoate salts, the protonated amine moiety is available as a hydrogen-bond donor and forms bonds with the benzoate or fumarate anions, and these salts are dominated by a multitude of N-HÁ Á ÁO and O-HÁ Á ÁO hydrogen-bonding interactions that connect the cations and anions into strongly hydrogen-bonded ribbon-like structures. The ethane backbone and the malleable ethylamine fragment of the bedaquiline core result in a high degree of flexibility, and molecular conformations vary not only widely between the bedaquiline and bedaquilinium structures, but even between independent molecules within the same structure (both the free base and the fumarate are Z 0 = 2 structures). For a pharmaceutically relevant material, it is essential that a crystalline material can be obtained in a stable and well-defined form. The formation of solvates is generally undesirable, especially if the incorporated solvent molecules are volatile or not generally recognized as safe (GRAS) for human consumption. For the bedaquilinum system, the pronounced conformational flexibility makes any predictions about how a bedaquilinium anion pair might crystallize, and whether solvates are formed and of which kind, extremely difficult. In silico crystal-structure prediction, even if only intended as a screening to narrow down a list of anion and solvent candidates, is not yet a viable option for this system; therefore, the best recourse for the bedaquiline system remains experimental screening of combinations of anions, solvents and crystallization conditions to establish which combinations will yield stable and welldefined crystal forms for potential use in pharmaceutical formulations. To this end, we have investigated the combination of bedaquiline with maleate as the anion. Via screening of solvents and crystallization conditions, we were able to establish the structures of several of its solvates: a hemihydrate, a THF solvate, an acetone/hexane solvate and an ethyl acetate solvate. A solvate-free form obtained by in situ desolvation of the acetone/hexane solvate will also be described. The influence of the incorporated solvents on the crystal structures and their stability will be discussed.

Structural commentary
Probability ellipsoid plots with selected atom labels and solvate molecules (where present) are shown in Figs. 1-5. The atom-naming scheme was adopted from the one used for the fumarate and benzoate structures and used for all solvates. Refinement details, including disorder refinement strategies (where present) are given in the Refinement section.
Similar to the other three bedaquilinum salts reported thus far, the maleate salt features a singly protonated bedaquilinium cation with a 1:1 anion-to-cation ratio. Like in the fumarate and benzoate salts, the protonation site is the dimethyl amine fragment. The second basic site, the quinoline nitrogen atoms, remained unprotonated, in agreement with the second pK a of maleic acid (6.22, European Chemical Agency, 2015), which is not sufficiently basic for a proton transfer to this site. The first pK a of maleic acid (1.94, European Chemical Agency, 2015) should be sufficient to protonate the quinoline site if higher ratios of maleic acid to bedaquiline are used (Okezue et al., 2020). We were, however, unable to identify or isolate any different crystalline materials when increasing the amounts of acid (screening was done by powder XRD).
The four bedaquilinium maleate structures presented here were found to be isomorphous or nearly isomorphous, differing mostly only in the nature of the incorporated solvate molecules. One of the structures, the ethyl acetate solvate, also shows a pronounced modulation of the bedaquilinium maleate, leading to breaking of the crystallographic symmetry observed for the other structures (see the Supramolecular features section for a detailed discussion). Similar isomorphous structures were also found for samples obtained from other solvent systems such as isopropanol or n-propanol, as evidenced by their powder XRD patterns. However, no single crystals of high enough quality for a full structural analysis could be obtained thus far.
The ethane backbone and the malleable ethylamine fragment gives the bedaquilinium cation a high degree of flexibility that allows the cations to respond readily to crystalpacking forces. In the previously reported structures, the conformations did vary widely not only from structure to structure, but even between independent molecules within the same structure (both the free base and the fumarate are Z 0 = 2 structures). For these structures, the torsion angles involving the ethylamine fragment adopted conformations ranging between gauche and trans (Table 1)

Figure 3
Probability ellipsoid plot (50% probability) of the acetone/hexane solvate (see Fig. 1 for cation and anion carbon-atom labels). C-bound H atoms and labels for C and H atoms are omitted for clarity. Acetone molecules are disordered: fourfold around a twofold axis plus general disorder (O8, disorder by the twofold axis not shown for clarity) or with a hexane molecule (O7, the hexane molecule is located on a twofold axis).

Figure 5
Probability ellipsoid plot (50% probability) of the desolvated structure (see Fig. 1 for cation and anion carbon-atom labels). C-bound H atoms and labels for C and H atoms are omitted for clarity.

Figure 4
Probability ellipsoid plot (50% probability) of the ethyl acetate solvate, showing both ion pairs (suffixes A and B) related by pseudo-translation (see Fig. 1 for cation and anion carbon-atom labels). C-bound H atoms, labels for C and H atoms and disorder of ethyl acetate molecules are omitted for clarity.
gauche found mostly for free base bedaquiline, where it was induced by an intramolecular O-HÁ Á ÁN hydrogen bond. However, one of the two molecules in the fumarate salt also featured a single gauche angle (for the C1-C2-C3-C4 torsion angle), and two angles in between gauche and trans [C2-C3-C4-N1 for the two fumarate molecules, with values of 137.2 (2) and 133.7 (2) , respectively]. All other torsion angles involving the ethylamine group adopted trans conformations with various degrees of slight distortions, ranging from 164.04 (15) to 178.8 (3) . The maleate salts follow the same trend. The torsion angles are all slightly distorted trans and range from À164.4 (7) to 176.4 (3) (C17-C1-C2-C3 and C1-C2-C3-C4 angles, extreme values are for each one of the two molecules of the ethyl acetate solvate).
The other free variables that determine the overall molecular structure of the cations are the torsion angles between the rigid planes of bedaquiline, i.e. the 6-bromo-2-methoxyquinoline, phenyl and naphthyl planes ( Table 1). Variations of only a few degrees can be seen between equivalent angles of the various interplanar angles in the cations, as would be expected for mostly isomorphous structures. The 6-bromo-2methoxyquinoline vs phenyl angle ranges from 63.0 (2) to 71.31 (7) , which is slightly smaller but similar to what was observed in the previously reported structures [73.29 (7) to 86.02 (8) ; Petit et al., 2007;Okezue et al., 2020]. Phenyl to naphthyl angles are between 63.7 (1) and 66.60 (7) . Previously reported values span a much wider range, from 44.2 (1) to 89.74 (9) . The bromo-2-methoxyquinoline vs naphthyl angles are between 26.4 (1) and 32.62 (9) , compared to 8.16 (9) to 37.50 (6) for the other known bedaquiline structures.
Numerical variations between the four structures are thus clearly resolved. They are not, however, large enough to substantially alter the overall shape and appearance of the cations, as can be seen in a least-squares overlay based on the atoms C1, C2, C7, C17 and C23 around the center of the cation (Fig. 6). The structures clearly still have the same overall conformation, just slightly modulated by interactions with solvate molecules and small differences in unit cell dimensions. The largest variations in the overlay can be seen for the outer atoms of the 6-bromo-2-methoxyquinoline plane, especially the bromine atom, the methoxy group, the outer atoms of the naphthyl group, and to a lesser degree for the dimethyl ammonium fragment.

Supramolecular features
Packing and intermolecular interactions not involving solvate molecules are essentially identical between the four structures, a virtue of their isomorphous or nearly isomorphous nature. Unless stated otherwise, all distances in the following discussion will be those of the hemihydrate structure.
The main directional forces that are involved in stabilizing crystals of bedaquilinium maleate are hydrogen bonds (Tables  2-7) andstacking interactions. One hydrogen bond is intramolecular and connects the carboxylic acid and carboxylate groups of the hydromaleate anion, which shows the very strong and close to symmetrical hydrogen bonding typical for cis-dicarboxylic acid anions (Fig. 7). The acidic maleate hydrogen atoms are well resolved in all five structures and their positions were freely refined. The position of the H atom varies slightly between the five structures. It is close to symmetric, with a slight deviation towards oxygen atom O5 in all but the THF solvate (see Table 8 for numerical details). The more accurately measured carbon-oxygen bond distances confirm the slight asymmetry for the hydromaleate, with the Least-squares overlay based on the atoms C1, C2, C7, C17 and C23. Color coding: hemihydrate -orange; THF solvate -green; ethyl acetate solvate -red and pink (two independent molecules); acetone/hexane solvateblue; desolvated structure -cyan. Table 2 Hydrogen-bond geometry (Å , ) for the hemihydrate. Symmetry codes: (i) x À 1 2 ; y À 1 2 ; z; (ii) Àx þ 1; y; Àz. Table 3 Hydrogen-bond geometry (Å , ) for the THF solvate. C33-O4 bond being on average 0.02 Å shorter than the C36-O5 bond. This includes the THF solvate, for which the H atom was found slightly closer to O4, indicating that the Hatom position is not measured sufficiently accurately to reliably determine its actual position (among the five structures, the THF solvate has the largest estimated standard deviations for atom positions, and within its s.u., the position of H5 is symmetric between O4 and O5).
The not quite symmetric nature of the hydromaleate anion could be a result of asymmetric intermolecular hydrogen bonding towards the two ends of the anion. Oxygen atom O3 acts as a hydrogen-bond acceptor towards the ammonium cation, while O6 plays the same role for the hydroxyl group of another cation (at 1 2 + x, 1 2 + y, +z). The N-HÁ Á ÁO hydrogen bond, being charge assisted, is slightly shorter and stronger than its O-HÁ Á ÁO equivalent on the other side of the anion, inducing the negative charge of the anion to be localized more on the O3/O4 carboxylate group, and the positive proton being slightly delocalized towards O5.
chain are related to each other via half-unit translations of the C-centered cell (AE 1 2 + x, AE 1 2 + y, +z). This differentiates the maleate structures described here from the other previously described bedaquilinium salt structures, the fumarate and benzoate structures, in which hydrogen-bonding interactions between cations and anions led to formation of layered structures (Okezue et al., 2020). For highly solvated structures such as the maleate salts described here, the formation of 1D rather than 2D structures can be of relevance for the resilience of the lattice upon removal of solvent, or the persistence of the packing motif if a different solvent is used. Interactions within the layers or chains, mediated via hydrogen bonds, are likely to be strong and persistent. The stability of the entire lattice thus depends on how these layers or chains are connected with each other. Are they tightly interwoven or connected in other ways to ensure stability of the lattice after removal or exchange of solvate molecules? Or can layers or chains easily move past each other, thus allowing easy movement of the secondary building units and either collapse or undergo a complete rearrangement of the entire structure?
In the maleate salts, interactions between individual chains is facilitated through effective interlocking of neighboring chains as well as a number of directional interactions, such as C-HÁ Á ÁO, C-HÁ Á ÁN and C-HÁ Á Á interactions. Importantly, neighboring chains that are rotated against each other by the twofold axis are interdigitating with each other viastacking interactions of the bromoquinoline rings, preventing easy slippage of chains against each other. The stacked quinoline rings are thus related to each other through a twofold rotation (1 À x, +y, 1 À z). They are not exactly coplanar but their planes are angled against each other by 19.43 (8) . As a result, no exact interplanar distance can be defined, but the 3.432 (14) Å centroid-to-centroid distance between the quinoline rings (measured for the hemihydrate) indicates an efficient stacking interaction. The closest atom-toatom distance is 3.252 (7) Å for the two atoms C30 related by the twofold axis.
Additional weaker interactions within the 1D chains and between parallel chains as well as chains that are inclined with respect to each other are provided by C-HÁ Á ÁO and C-HÁ Á ÁN interactions involving the quinoline nitrogen and several of the hydromaleate oxygen atoms as well as by several C-HÁ Á Á interactions towards the naphthyl and quinolinesystems. The para C-H group of the phenyl ring forms a C-HÁ Á ÁN hydrogen bond with the quinoline nitrogen atom of a neighboring molecule (at 1 2 + x, À 1 2 + y, +z). C-HÁ Á ÁO bonds towards the maleate O atom O6 are established by both C3 and C6, being the methylene and methyl groups of the dangling dimethyl propylene ammonium group. O6 also acts as the hydrogen-bond acceptor for the hydroxyl O-HÁ Á ÁO bond, and these C-HÁ Á ÁO bonds thus just reinforce this connection within the hydrogen-bonded chains, and do not provide any new connection between chains. Atoms O4 and O5 of the maleate, on the other hand, act as C-HÁ Á ÁO acceptors towards the methyl C6 and naphthyl C10 atoms from cations in neighboring chains, thus providing some stabilization for the overall 3D lattice. Another C-HÁ Á ÁO interaction towards O4, originating from another methyl C6 atom, does provide reinforcement for the N-HÁ Á ÁO bond and no connection between neighboring chains. The methoxy and hydroxyl O atoms do not act as acceptors for intramolecular C-HÁ Á ÁO bonds. Finally, a number of intermolecular C-HÁ Á Á interactions towards the naphthyl and quinolinesystems are observed: from naphthyl C14 and maleate C35 towards quinoline density (these interactions are within the 1D chains and assist in stabilizing the hydrogen bonds), from phenyl C19 towards maleate C35 (this is an inter-chain interaction), and a weaker interaction from methyl C6 towards Packing view of the hemihydrate structure showing the propagation directions of the hydrogen-bonded chains. In the lower half of the unit cell, chains propagate horizontally (right-left, along [110]; in the upper half they propagate longitudinally (forward-backward, along [110] (3) naphthyl C10 (this is an inter-chain interaction, but the geometry of this interaction makes it unlikely to be very stabilizing). The sum of these interactions, especially the interlocking of the stacked quinolones, is likely to prevent slippage of hydrogen-bonded chains against each other, which stabilizes the three-dimensional arrangement against collapse, even upon complete removal of all solvate molecules. Additional intermolecular interactions towards the various solvate molecules are observed. These are generally much weaker than the interactions described so far, with the possible exception of the water molecules in the hemihydrate, which are partially hydrogen bonded to the main lattice. The partial occupancy of the water molecules does, however, indicate that these hydrogen bonds are not essential in any way to sustain the overall structure, despite being individually quite strong. It appears that the water molecules simply occupy the positions most suitable for them, but that they do not influence the overall structure much. This is further substantiated by the fact that the hemihydrate is isomorphous to the other solvates, with no indication that the structure is modulated much by the presence of the water molecules. Their partial occupancy does, for example, not lead to disorder of the cations or anions, but the 1D chains are unfazed by the presence or absence of the water molecules.
When the water molecules are present, then they are located such that they are hydrogen bonded (Fig. 8). One of the molecules, associated with O7 and about one quarter occupied [refined occupancy 0.276 (17)] is located in a general position and is hydrogen bonded to the maleate C O group of O3 (which is also hydrogen bonded to the ammonium cation). The second solvate water molecule, associated with O8, is located on a twofold axis, and is in hydrogen-bonding distance to the other water molecule. It features a higher occupancy rate, 0.40 (4), but less than double that of the first water molecule, indicating that it is hydrogen bonded to either O7 or to its symmetry-related counterpart by the twofold axis, but not to both at the same time. Its large displacement ellipsoid indicates possible unresolved disorder resulting from the varying environments and/or large thermal libration due to the absence of a second hydrogen-bonding partner and the presence of an unoccupied void space instead. No second acceptor site for the first water molecule is present, which indicates that the overall structure is not well suited for inclusion of water in its lattice. A PLATON SQUEEZE analysis (van der Sluis & Spek, 1990;Spek, 2015) revealed 6.9% of additional void space not occupied by any solvate molecules (even partially occupied). Crystals of the hemihydrate were grown from acetonitrile by evaporation with only trace amounts of water available from the solvent and the surrounding atmosphere, and crystals were exposed to atmosphere prior to analysis. Thus presence of additional acetonitrile solvate molecules in the original crystals, which were subsequently lost, is likely. Attempts to grow single crystals from solvents with more available water have so far been unsuccessful, which indicates that the presence of larger amounts of water might result in formation of a different type of maleate salt. Further single crystal and powder XRD experiments are under way to investigate this possibility.
The less-than-ideal nature of the overall structure to host hydrogen-bonded guest solvate molecules is supported by the ready formation of solvate structures with aprotic solvents, such as THF and acetone/hexane resulting in isomorphous structures with little or no modulation. Well-formed crystals could readily be grown from these solvents up to millimeters in size, showing how readily accessible this structural motif is.
THF molecules in that solvate are only loosely bonded to anions and cations. Two sites occupied by THF molecules were found. One located on a twofold axis and intrinsically 1:1 disordered. It is encapsulated between four different naphthyl groups and is weakly hydrogen bonded to all of them via C-HÁ Á ÁO interactions originating from C9 and C12. The other molecule is in a general position and exhibits no directional interactions with any neighboring entities at all, thus simply taking up the space provided by the lattice. The molecule is disordered, in a refined ratio of 0.587 (16) to 0.413 (16), further supporting the absence of any steering interactions with its neighbors in space.
Acetone and hexane molecules also do not strongly interact with the cations and anions in this solvate. Two distinct solvate-occupied sites are present in the lattice. One site is occupied by only acetone. This molecule is located on and disordered around a twofold axis and is additionally disordered by a slight tilt of the molecule. Occupancies refined to two Â 0.230 (11) and two Â 0.270 (11) for this site. The other solvate site is occupied by both acetone and hexane, with either one hexane molecule located on another twofold axis, or two acetone molecules being symmetry equivalent by this axis. The occupancy rates refined to 0.505 (9) and 0.495 (9) in slight favor of the acetone molecules. The acetone molecules of this site are weakly bound via a C-HÁ Á ÁO interaction to the methoxy methyl group and to one of the maleate C-H groups. No other directional interactions of either acetone or hexane with anions or cations are observed.
Interactions with solvate molecules are more pronounced in the ethyl acetate solvate, but the exact nature of the interactions is obscured by substantial disorder, with up to fivefold disorder refined for one solvate cluster. Solvate disorder induces disorder of a cation phenyl group and a cation naphthyl group (see the Refinement section for a more detailed discussion of disorder). Some C-HÁ Á ÁO interactions appear evident for the major disordered moieties though, which will be discussed below. The larger extent of the solvate interaction with the main structure, when compared to the hemihydrate, THF and acetone/hexane solvates, is also supported by the fact that the ethyl acetate solvate is not exactly isomorphous with the other three solvates, but is modulated and crystallizes with lower symmetry than the other structures. C-centered and twofold symmetry are broken, resulting in a structure with a similar unit-cell size and shape, but with a primitive lattice and space group P2 1 . Exact translation and twofold symmetry for the ethyl acetate solvate is broken by ordering of the solvate molecules and by a slight modulation of cations and anions (see Refinement section for more details).
The ethyl acetate molecules are arranged into two clusters with light and severe disorder, refined as twofold and fivefold disorder, with partial overlap between the two clusters. Total occupancy for the severely disordered site refined to less than unity, just above 60% [0.641 (6)], inducing disorder for the surrounding naphthyl and phenyl groups. Additional unresolved disorder cannot be ruled out for this site. Despite the pseudo-translational symmetry, the two solvate sites are clearly distinct from each other, with little to no correlation effects between the two sites, and the differences between the two solvent sites appear to be the main reason for modulation and breaking of the C2 symmetry.
In the less disordered and fully occupied solvate site, the major moiety ethyl acetate [87.4 (3)% occupancy] is hydrogen bonded via its keto group to the methoxy methyl group, C32A. The major moiety molecule of the other site, on the other hand, exhibits C-HÁ Á ÁO bonds originating from methyl ammonium C5A and naphthyl C12B. The same interaction is observed for one of the minor moieties at this site, with a combined occupancy rate of 35.4%, or more than half of the total site occupancy. No C-HÁ Á ÁO interaction originating from the methoxy methyl group C32B is present.

Stability and desolvation, solvent-free salt
The stronger interaction of the ethyl acetate solvate molecules with the framework molecules, when compared to their THF and acetone/hexane analogues, also translates into the stability of the solvates. The THF and acetone/hexane analogues readily loose most of their solvate molecules under ambient conditions, and crystals become opaque within a few hours. Crystals of the ethyl acetate solvate, under the same conditions, do not change in appearance. When taken out of solution and stored overnight, exposed to normal atmosphere, crystals of the ethyl acetate solvate are visually unchanged, and data collected from single crystals are unchanged from data collected from a crystal fresh out of mother liquor. Solvate molecules are still clearly resolved, the disorder pattern is not changed noticeably, and occupancy rates are unchanged. The modulation of the main molecule framework is preserved, unit-cell parameters are virtually unchanged (reduction by 0.3%), and mosaicity is essentially unchanged (0.73 and 0.77 , respectively; see supporting information, Fig. S1).
Crystals of the THF and acetone/hexane solvate behave differently. Crystals of either compound become milky within a few minutes of being taken out of mother liquor, and even the cores of large crystals (up to 1 mm) completely lose transparency within a few hours when stored in air outside the mother liquor. Crystals do, however, retain crystallinity, despite becoming opaque and white in appearance. Singlecrystal data for such a crystal obtained from the acetone/ hexane solvate did diffract well, with little to no loss of diffraction power compared to the solvated crystals or the ethyl acetate solvate (Fig. S1 in the supporting information), and there was only a small increase in mosaicity from 0.71 to 0.85 after storing in air for 14 h.
Hot stage microscopy showed that if the crystals are crushed, solvent loss is rapid for both the acetone/hexane as well as the ethyl acetate samples. When single crystals were crushed on a microscope slide and heated at 10.0 C min À1 to 110.0 C, then at 5.0 C min À1 to 120.0 C and then finally at 2.0 C min À1 to 140.0 C, no loss of solvate molecules was observable for either the acetone/hexane nor the ethyl acetate crystals for either a dry sample or immersed in mineral oil. Onset of melting was observed between 122.1 and 124.5 C, and melting was complete at 128.3 to 133.7 C, with no noticeable difference between the acetone/hexane and the ethyl acetate sample (Table 9, selected figures shown in the supporting information). The comparable melting temperatures of these materials support the finding from XRD that the crystal structures of the maleate solvates are isomorphically related and that the presence of solvate is not required to maintain these structures. This indicates that for smaller particles, solvate molecules are rapidly lost for both solvates, possibly before start of the hot stage microscopy experiment, while larger crystals of the ethyl acetate solvate (> 200 mm 3 such as used for single-crystal diffraction) do not desolvate readily and retain most of their solvate molecules.
For the acetone/hexane sample stored in atmosphere overnight, when analyzed by SC-XRD, a noticeable change of the unit-cell dimensions was observed, accompanied by a decrease in volume by ca 3.5% [from 3733.9 (3) to 3603.0 (5) Å 3 ]. An overlay of the structure before and after desolvation is shown in the supporting information (Fig. S2). Changes of unit-cell parameters and a slight shifting of functional groups are perceptible, but the overall magnitude of those changes is small.
The decrease of the unit-cell volume is, however, substantially less than the 20.9% of the volume taken up by solvate molecules in the acetone/hexane solvate, Fig. 9. Indeed, a PLATON SQUEEZE analysis (van der Sluis & Spek, 1990;Spek, 2015) reveals a residual void space of 16.8% of the unit-cell volume, Fig. 10. This indicates that either a substantial fraction of the solvate molecules is retained, or that the hydrogen-bonded framework is stable enough to withstand collapse, even without any solvate molecules in the void space between the bedaquilinium maleate framework. A  Table 9 Hot stage optical microscopy data for the acetone/hexane and the ethyl acetate crystals. future in-depth analysis of several bedaquilinium salts, including the various solvates of the maleate system, will focus on their thermal stability and physical properties, and will include thermal gravimetric analysis, porosity measurements of the desolvated salts and surface-area measurements. The single-crystal structure of the acetone/hexane crystals stored under ambient conditions does, however, already provide some first insights. Analysis of the data revealed a well-defined bedaquilinium maleate framework, with barely any increased libration, but a completely featureless electron-density difference map for the areas previously taken up by the acetone/hexane molecules. The largest difference-electron peaks inside the void area are less than 0.5 e Å 3 . A solvate SQUEEZE analysis performed using the program PLATON revealed some residual electron density, but substantially less than what would be expected for full occupancy. The SQUEEZE procedure corrected for 66 electrons within the solvent-accessible voids, equivalent to 1.14 molecules of acetone per unit cell, or 0.28 acetone per cation-anion pair. Prior to desolvation, one molecule of acetone and half a molecule of hexane were present per cation-anion pair, equivalent to 202 electrons per unit cell. Thus, there seems to be some retention of solvate molecules within the voids (ca one third based on the SQUEEZE data), but those solvate molecules appear to be completely disordered and equally distributed within the solvate-accessible area. The bedaquilinium maleate framework is not affected by the residual solvate. No disorder is observed for either cation or anion, nor any increased libration, indicating that any residual solvate has negligible interaction with the framework, and that desolvation is homogeneous throughout the whole crystal. Thermal gravimetric analysis and surface measurements, to be reported in an upcoming publication, will provide more insight as to how much or if any residual solvates are indeed present in the void area.

Database survey
Only four structures of bedaquiline or its salts have been previously reported in the literature (Cambridge Structural Database; Groom et al., 2016), viz. free base bedaquiline (Petit et al., 2007), the fumarate salt (Okezue et al., 2020), and two isomorphous solvates of the benzoate salt (Okezue et al., 2020). The structures of the salts are dominated by a multitude of N-HÁ Á ÁO and O-HÁ Á ÁO hydrogen-bonding interactions that connect the cations and anions into strongly hydrogenbonded motifs, while in free base bedaquiline the packing is dominated by weaker and less directional interactions such as BrÁ Á ÁBr interactions and -stacking (Petit et al., 2016). In all structures, the ethane backbone and the malleable ethylamine fragment of the bedaquiline core give the cations a high degree of flexibility, and molecular conformations not only vary widely between the bedaquiline and bedaquilinium structures, but even between independent molecules within the same structure (both the free base and the fumarate are Z 0 = 2 structures). The fumarate salt was solvent free. The benzoate salt formed a hydrate with one strongly bound solvate water molecule, and a second solvate site occupied either partially by water (occupancy 17%) or by disordered acetonitrile. The acetonitrile solvate was prone to desolvation and converted quickly into a simple monohydrate once taken out of solution. In both of the salts, the anions and cations are bridged via hydrogen atoms into 2D ribbons in which the bedaquilinum cations wrap around a single strand of anions (fumarate) or around anions and water molecules (benzoate). This differentiates the fumarate and benzoate salts from the maleates, which exhibit simpler 1D chains of anions and cations. Residual void space in the acetone/hexane structure after artificial removal of solvent molecules. The solvent-accessible volume would be 781 Å 3 [20.9% of the unit-cell volume; probe radius 1.2 Å ; numerical values from PLATON SQUEEZE calculation (van der Sluis & Spek, 1990;Spek, 2015)]. Probability ellipsoids are at the 50% level.

Figure 10
Residual void space in the desolvated structure. A preset heating program ramp was used during each individual analysis using the hot stage system controller programmed with a ramp of 10.0 C min À1 to 110.0 C, at 5.0 C min À1 to 120.0 C, and at 2.0 C min À1 to 140.0 C. The system calibration was verified with meltingpoint standards prior to analyses. Samples were analyzed in triplicate, twice as dry mounts and once in mineral oil, USP (CAS: 8042-47-5), which was allowed to cover the sample by capillarity. Sample and thermomicroscopy information is given in Table 9, selected images are given in the supporting information. Bedaquiline free base (400.3 mg) was weighed into a 20 mL glass scintillation vial and dissolved in 3 mL THF. Maleic acid (85.6 mg) was added and the contents mixed. The solution was allowed to evaporate slowly at ambient conditions. White crystals that appeared dry were evident within two days in the vial. Approximately 16 mg of this dried material was weighed into a 2 dram glass vial, re-dissolved in 1200 mL THF, and wrapped in aluminum foil. An 18 gauge needle was inserted into the top of the aluminum foil to allow for slow evaporation at ambient conditions until solids were evident and the sample appeared dry.

Synthesis and crystallization
Acetone/hexane solvate, C 32 H 32 BrN 2 O 2 ÁC 4 H 3 O 4 Á-C 2 H 6 O 2 Á0.25C 6 H 14 : Bedaquiline maleate (23.1 mg) was weighed into a glass vial and dissolved in acetone (3 mL). A layer of n-hexanes was gently streamed onto the top of the solution. The vial contents were capped and placed under a hood at ambient conditions. After two days, clear crystals were evident in the vial.
Desolvated structure, C 32 H 32 BrN 2 O 2 ÁC 4 H 3 O 4 : Crystals of the solvent-free compound were obtained from the monoacetone quadrant-hexane solvate by drying in air on a microscope slide. Crystals become milky overnight when taken out of mother liquor solution and left to dry in air, but retain crystallinity. Data collection revealed a solvent-free structure. Ethyl

Refinement
Crystal data, data collection and structure refinement details are summarized in Table 10.
The structures of the four solvates and of the solvent-free salt are closely related. The hemihydrate, THF solvate, the acetone/hexane solvate and the solvate-free salt derived from the acetone/hexane solvate are isomorphous in space group C2. In the ethyl acetate solvate, C-centered and twofold symmetry is broken and the salt crystallizes in P2 1 , but the structure is closely related to the other C-centered structures.
The four isomorphous structures were refined using a common model for the non-solvate part of the structures, with the THF solvate, the acetone/hexane solvate and the desolvated structure solved by isomorphous replacement. The ethyl acetate solvate was solved independently, by dual Patterson/ direct methods, but the atom-naming scheme from the other structures was adopted and augmented by suffixes A and B to distinguish between the two cation-anion pairs related by pseudo-translation.
Hydrogen-atom treatment: C-H bond distances were constrained to 0.95 Å for aromatic and alkene C-H moieties, and to 1.00, 0.99 and 0.98 Å for aliphatic C-H, CH 2 and CH 3 moieties, respectively. N-H bond distances were constrained to 1.00 Å for pyramidal (sp 3 hybridized) ammonium R 3 H + groups. O-H distances of alcohols were constrained to 0.84 Å . Methyl CH 3 and hydroxyl H atoms were allowed to rotate but not to tip to best fit the experimental electron density. The positions of hydromaleate acidic hydrogen atoms were freely refined. Water H-atom positions in the hemihydrate were refined and O-H and HÁ Á ÁH distances were restrained to 0.84 (2) and 1.36 (2) Å , respectively. H-atom positions were further restrained based on hydrogen-bonding considerations. A damping factor was applied during refinement. In the final refinement cycles, the damping factor was removed and the water H atoms were constrained to ride on their carrier oxygen atoms. U iso (H) values were set to a multiple of U eq (C/O/N) with 1.5 for CH 3 and OH, and 1.2 for C-H, CH 2 and N-H units, respectively.
Disorder and solvate refinement, handling of void space: In the hemihydrate, two partially occupied water molecules are situated in the asymmetric part of the unit cell. One is in a general position, the other located on a twofold axis. They are hydrogen bonded to each other, and the one in the general position is also hydrogen bonded to atom O3 of the hydromaleate anion. Water H-atom positions were refined as described above. Occupancy rates refined to 0.276 (17) for O7 (in the general position) and 0.40 (4) for O8 (on the twofold axis).
Additional solvent-accessible space is present in the crystal lattice (two Â 123 Å 3 or 6.9% of the unit-cell volume). No electron density was found inside the void space [a PLATON SQUEEZE analysis (van der Sluis & Spek, 1990;Spek, 2015) found eight electrons in the combined void space], and the content of the void space was ignored.
In the THF solvate, two THF molecules were refined as disordered, one in a 1:1 ratio around a twofold rotation axis, the other in a general position. The three disordered moieties were restrained to have similar geometries. U ij components of ADPs for disordered atoms closer to each other than 2.0 Å were restrained to be similar. Subject to these conditions, the occupancy ratio for the molecule in the general position refined to 0.587 (16) to 0.413 (16).
In the acetone/hexane solvate, two symmetry-equivalent acetone molecules are disordered with a hexane molecule located on a twofold axis. Another acetone molecule is located on and disordered around a twofold axis and additionally disordered by a slight tilt of the molecule. All acetone moieties were restrained to have similar geometries and to be close to planar. The two C-C bond distances within the acetone were restrained to be similar to each other. C-C bond distances of the hexane molecule were restrained to target values [1.55 (1) Å ] and hexane C-C-C angles were restrained to be similar. U ij components of ADPs for disordered atoms closer to each other than 2.0 Å were restrained to be similar. Subject to these conditions, the occupancy rates refined to 0.505 (9) and 0.495 (9) for the acetone/hexane disorder, and two Â 0.230 (11) and two Â 0.270 (11) for the disordered acetone.
The structure of the ethyl acetate solvate exhibits pseudo Ccentered symmetry emulating space group C2 as observed for the hemihydrate, THF and acetone/hexane solvates and the solvent-free salt derived from the acetone/hexane solvate. Exact translation and twofold symmetry for the ethyl acetate solvate is broken by the solvate molecules and by a slight modulation of cations and anions. The mean intensity for reflections that should be systematically absent in C2 was 1.8, vs 6.9 for all reflections (2.2 vs 2.9 for mean intensity/).
Ethyl acetate molecules are arranged into two clusters with light and severe disorder. Solvate disorder induces disorder of a cation phenyl and a cation naphthyl group. The site associated with the ethyl acetate molecule of O1/O2 was refined as twofold disordered and as fully occupied. The site associated with the ethyl acetate molecule of O3/O4 was refined as fivefold disordered and only partially occupied. One of the moieties of O3/O4 (suffix F) extends away from the main cluster. It induces the disorder of the O1/O2 ethyl acetate, and for the naphthyl group of cation A. A common occupancy ratio was used for these three entities. Disorder of the phenyl group of cation B is correlated with multiple disordered moieties of the severely disordered ethyl acetate and was refined independently.
All ethyl acetate moieties were restrained to have similar geometries. The acetate sections were restrained to be close to planar. The ethyl C-C bond distances were restrained to a target value [1.55 (2) Å ]. Disordered phenyl and naphthyl groups were restrained to have similar geometries as their not disordered counterparts in the other cation. U ij components of ADPs for disordered atoms closer to each other than 2.0 Å were restrained to be similar. Subject to these conditions, the occupancy rates refined to 0.874 (3) to 0.126 (3) for the twofold-disordered ethyl acetate of O1/O2 (shared with the naphthyl disorder of cation A). The occupancy rates for the partially occupied site refined to 0.171 (7), 0.183 (7), 0.126 (3) (the same as minor moiety of O1/O2 ethyl acetate), 0.074 (6) and 0.087 (6), for a total occupancy of 0.641. The occupancy ratio of the phenyl disorder of cation B refined to 0.573 (17) to 0.427 (17).
Crystals of the solvate-free salt were obtained from the acetone/hexane solvate by drying on a glass slide in air overnight. In the solvated structure, acetone and hexane molecules are located in infinite channels and slowly vacate the crystal lattice. Crystals become milky overnight when taken out of mother liquor solution and left to dry in air, but retain crystallinity.
No substantial electron density was found in the previously solvate-occupied channels (largest void peaks are less than 0.5 electrons per cubic Å ngstrom), and the residual electrondensity peaks are not arranged in an interpretable pattern. The structure was refined both with and without correction of residual electron density, with only marginally different results. In the second approach, the structure factors were augmented via reverse Fourier transform methods using the SQUEEZE routine (van der Sluis & Spek, 1990;Spek, 2015) as implemented in the program PLATON. The resultant FAB file containing the structure-factor contribution from the electron content of the void space was used together with the original hkl file in the further refinement. (The FAB file with details of the SQUEEZE results is appended to the CIF). The SQUEEZE procedure corrected for 66 electrons within the solvent-accessible voids, equivalent to 1.14 molecules of acetone per unit cell, or 0.28 acetone per cation-anion pair. Prior to desolvation, one molecule of acetone and a quarter molecule of hexane were determined per cation-anion pair [the F(000) values of the solvated and unsolvated structures differ by 178 electrons].   For all structures, data collection: APEX3 (Bruker, 2020); cell refinement: SAINT (Bruker, 2020); data reduction: SAINT (Bruker, 2020 (Macrae et al., 2020); software used to prepare material for publication: publCIF (Westrip, 2010).

Special details
Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes. Refinement. The position of the hydromaleate acidic hydrogen atom was freely refined. Two partially occupied water molecules are situated in the asymmetric part of the unit cell. One in a general position, the other located on a twofold axis. They are hydrogen bonded to each other, and the one ion the general position is also Hbonded to O3 of the hydromaleate anion. Water H atom positions were refined and O-H and H···H distances were restrained to 0.84 (2) and 1.36 (2) Angstrom, respectively, and H atom positions were further restrained based on hydrogen bonding considerations. A damping factor was applied during refinement. In the final refinement cycles the damping factor was removed and the water H atoms were set to ride on their carrier oxygen atoms. Subject to these conditions the occupancy rates refined to 0.276 (17) for O7 (general position) and 0.40 (4) for O8 (twofold axis). Additional solvent accessible space is present in the crystal lattice (two times 123 cubic Angstrom, or 6.9% of the unit cell volume). No electron density was found inside the void space (A Platon Squeeze analysis corrected for 8 electrons in the combined void space), and the content of the void space was ignored.

Special details
Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

sup-20
Acta Cryst. (2021). E77, 433-445 Refinement. The structure exhibits pseudo C-centered symmetry emulating space group C2. C2 symmetry was observed for the related hemihydrate and acetone / hexane solvates. Exact translation and two-fold symmetry for the ethyl acetate solvate is broken by the solvate molecules and by a slight modulation of cations and anions. Mean intensity is 1.8 for reflections that should be systematically absent for C2, vs 6.9 for all reflections (2.2 vs 2.9 for mean intensity / sigma). Ethyl acetate molecules are arranged into two clusters with light and severe disorder. Solvate disorder induces disorder of a cation phenyl and a cation naphtyl group. The site associated with the ethyl acetate molecule of O1/O2 was refined as two-fold disordered and as fully occupied. The site associated with the ethyl acetate molecule of O3/O4 was refined as five-fold disordered and only partially occupied. One of the moieties of O3/O4 (suffix F) extends away from the main cluster. It induces the disorder of the O1/O2 ethyl acetate, and for the naphtyl group of cation A. A common occupancy ratio was used for these three entities. Disorder of the phenyl group of cation B is correlated with multiple disordered moieties of the severely disordered ethyl acetate and was refined independently. All ethyl acetate moieties were restrained to have similar geometries. The acetate sections were restrained to be close to planar. The ethyl C-C bond distances were restrained to a target value (1.55 (2) Angstrom). Disordered phenyl and naphtyl groups were restrained to have similar geometries as their not disordered counterparts in the other cation. Uij components of ADPs for disordered atoms closer to each other than 2.0 Angstrom were restrained to be similar. Subject to these conditions the occupancy rates refined to 0.874 (3) to 0.126 (3) for the two-fold disordered ethyl acetate of O1/O2 (shared with the naphtyl disorder of cation A). The occupancy rates for the partially occupied site refined to 0.171 (7), 0.183 (7), 0.126 (3) (same as minor moiety of O1/O2 ethyl acetate), 0.074 (6) and 0.087 (6), for a total occupancy of 0.641. The occupancy ratio of the phenyl disorder of cation B refined to 0.573 (17) to 0.427 (17).

Special details
Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes. Refinement. The structure was solved by isomorphous replacement from its hemi-hydrate analogue (structure code RPI_46_1_2). Two symmetry equivalent acetone molecules are disordered with a hexane molecule located on a two-fold axis. Another acetone molecule is located on and disordered around a two-fold axis and additionally disordered by a slight tilt of the molecule. All acetone moieties were restrained to have similar geometries and to be close to planar. The two C-C bond distances within all acetone were restrained to be similar. C-C bond distances of the hexane molecule were restrained to target values (1.55 (1) Angstrom) and hexane C-C-C angles were restrained to be similar. Uij components of ADPs for disordered atoms closer to each other than 2.0 Angstrom were restrained to be similar. Subject to these conditions the occupancy rates refined to 0.505 (9) and 0.495 (9) for the acetone / hexane disorder, and two times 0.230 (11) and two times 0.270 (11) for the disordered acetone. Acetone and hexane molecules are located in infinite channels and slowly vacate the crystal lattice. Crystals become milky overnight when taken out of mother liquor solution and left to dry in air, but retain crystallinity. Data collection reveals a solvent free structure (see dataset 91_2_dry). [4-(6-bromo-2-methoxyquinolin-3-yl)-3-hydroxy-3-(naphthalen-1-yl)-4-phenylbutyl]dimethylazanium 3-

Special details
Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes. Refinement. Crystals were obtained from the acetone / hexane solvate by drying on a glass slide in air over night. In the solavted structure acetone and hexane molecules are located in infinite channels and slowly vacate the crystal lattice. Crystals become milky overnight when taken out of mother liquor solution and left to dry in air, but retain crystallinity. The structure was solved by isomorphous replacement from the acetone / hexane solvate (structure code 91_2). No substantial electron density was found in the previously solvate occupied channels (largest void peaks are less than 0.5 electrons per cubic Angstrom). Residual electron density was not corrected for. An alternative refinement using the Squeeze algorithm is appended below this cif file.

Special details
Geometry. All esds (except the esd in the dihedral angle between two l.s. planes) are estimated using the full covariance matrix. The cell esds are taken into account individually in the estimation of esds in distances, angles and torsion angles; correlations between esds in cell parameters are only used when they are defined by crystal symmetry. An approximate (isotropic) treatment of cell esds is used for estimating esds involving l.s. planes.

sup-59
Acta Cryst. (2021). E77, 433-445 Refinement. Crystals were obtained from the acetone / hexane solvate by drying on a glass slide in air over night. In the solavted structure acetone and hexane molecules are located in infinite channels and slowly vacate the crystal lattice. Crystals become milky overnight when taken out of mother liquor solution and left to dry in air, but retain crystallinity. The structure was solved by isomorphous replacement from the acetone / hexane solvate (structure code 91_2). The structure contains solvent accessible voids of 607 Ang3 combined. No substantial electron density peaks were found in the solvent accessible voids (less than 0.5 electron per cubic Angstrom) and the residual electron density peaks are not arranged in an interpretable pattern. The structure factors were instead augmented via reverse Fourier transform methods using the SQUEEZE routine (P. van der Sluis & A.L. Spek (1990). Acta Cryst. A46,[194][195][196][197][198][199][200][201] as implemented in the program Platon. The resultant FAB file containing the structure factor contribution from the electron content of the void space was used in together with the original hkl file in the further refinement. (The FAB file with details of the Squeeze results is appended to this cif file). The Squeeze procedure corrected for 66 electrons within the solvent accessible voids, equivalent to 1.14 molecules of acetone per unit cell, or 0.28 acetone per cation / anion pair. Prior to desolvation, 1 molecule of acetone and half a molecule of hexane were determined per cation / anion pair (equivalent to 202 electrons per unit cell).