Separation of Etiracetam Enantiomers Using Enantiospecific Cocrystallization with 2-Chloromandelic Acid.

Chirality plays an important role in the pharmaceutical industry since the two enantiomers of a drug molecule usually display significantly different bioactivities, and hence, most products are produced as pure enantiomers. However, many drug precursors are synthesized as racemates, and hence, enantioseparation has become a significant process in the industry. Cocrystallization is one of the attractive crystallization approaches to obtain the desired enantiomer from racemic compounds. In this work, we propose a chiral resolution route for an antiepileptic drug, S-etiracetam (S-ETI), via enantiospecific cocrystallization with S-2-chloro-S-mandelic acid (CLMA) as a coformer. The experiments indicate that the system is highly enantiospecific; S-2CLMA cocrystallizes only with S-ETI but not with R-ETI or RS-ETI. Therefore, the chiral purification of S-ETI can be achieved efficiently with a 69.1% yield and close to 100% enantiopurity from the racemic solution. Additionally, structural simulations of the S-ETI:S-2CLMA cocrystal reveal that the cocrystal structure has higher thermodynamic stability than that of R-ETI:S-2CLMA by about 5.5 kcal/mol (per cocrystal formula unit), which helps to confirm the favorability of the enantiospecification in this system.

Introduction

Enantioseparation processes are a crucial step in the pharmaceutical industry since new active pharmaceutical ingredients (APIs) are required to have high enantiopurity due to the possible adverse side effects of the undesired enantiomer.[1] Preferential crystallization has been an easy and very effective method to separate the desired enantiomer from the undesired one.[2,3] Unfortunately, this technique requires enantiomers to crystallize as a conglomerate-forming system, which is only found in ∼5–10% of all chiral systems.[4] Thus, diastereomeric salt formation is still widely used in the industry because it can be used with racemates, which are 90–95% of chiral crystalline compounds.[4,5]

An alternative chiral resolution method, cocrystallization using a chiral coformer, has been proposed for nonionizable compounds, especially racemates, that cannot easily form a salt.[6] Enantiospecific cocrystallization was first introduced by Springuel and Leyssens.[7] In this process, the coformer only cocrystallizes with one of the two enantiomers of the API, and the process can be performed with high enantiopurity and yield. However, the separation also can be done by diastereomeric cocrystallization, in which the coformer interacts with both enantiomers. Obviously, enantiospecific cocrystallization is more attractive since it is easier to design the crystallization process as the number of solid phases in the phase diagram is reduced.[8] Even though chiral resolution via cocrystallization has a common principle with the diastereomeric salt formation technique, adding another chiral compound to form a new crystalline product with a specific enantiomer, the mechanism and outcome are very different. The cocrystallization technique relies on intermolecular interactions between the API and coformer, such as hydrogen bonding, halogen bonding, CH−π interactions,[9] and van der Waals forces.[8] In diastereomeric salt formation, the API and resolving agent interact via strong ionic bonds.[8] Therefore, enantiopurification via cocrystallization is a worthwhile approach for resolution of racemic mixtures since the cocrystal formation does not alter the speciation of the compound in the crystallization step, maintaining its biological activity.[6] Another significant difference between these two procedures is that the cocrystallization technique commonly behaves as an enantiospecific system,[7,8,10,11] in contrast with salt formation where both diastereomers can form but one is typically more stable than the other.[8] Moreover, the hydrogen bonding pattern in cocrystals tends to be weaker and far more directional and thus dependent on the stereochemistry of the API than the ionic bond in salt formation.[8] Lastly, the separation process of cocrystals is much easier than that in salt formation because of the weaker bonds involved in cocrystals.

Levetiracetam or S-2-(2-oxopyrrolidin-1-yl)butanamide is an antiepileptic drug sold under the name Keppra. It is used for treating the symptoms of epileptic seizures that can also occur with patients who have primary or secondary tumors;[12] however, there is no report on the pharmaceutical efficacy of its enantiomer, R-etiracetam (R-ETI).[13] Furthermore, S-etiracetam (S-ETI) is also used in combination with other medicines to treat individual types of seizures.[14] Naturally, S-ETI is a chiral drug that crystallizes as a racemic compound rather than as a conglomerate. Consequently, direct preferential crystallization is expected to be difficult to apply to this system unless a convenient conglomerate salt or derivative can be found. Springuel and Leyssens have successfully resolved S-ETI from its racemic mixture with 70% yield using S-mandelic acid as a coformer via enantiospecific cocrystallization.[7] Cocrystallization using S-ibuprofen has also resolved S-ETI from its racemic mixture with 87.6% ee.[10] In another study, the racemate is converted to an ionic cocrystal by adding a suitable amount of ZnCl2 to the etiracetam solution, which then facilitates chiral resolution.[15] In addition, cocrystals of S-ETI with other compounds have also been found, with coformers including R-α-ketoglutaric acid[11] and d-tartaric acid.[16] Recently, enantioseparation has been achieved for a range of halogenated mandelic acids using cocrystallization with levetiracetam as a resolving agent.[17] Although one of the two diastereomers involved in the separation of the halogenated mandelic acid enantiomers will be similar to that used in the separation of the levetiracetam enantiomers, the potential second diastereomeric cocrystal in the two cases is distinct, which means that the ability to use enantioseparation in the first case does not indicate that the second case will also be successful. This led to a question whether halogenated mandelic acids could be a viable resolving agent for the enantiopurification of levetiracetam.

In cocrystallization, although there are heuristic rules for selecting suitable coformers,[18] the screening process is still generally done via a trial-and-error approach. This shows that further understanding of the molecular origins governing the chiral discrimination mechanism in cocrystal systems is needed. Thus, the purposes of this study are discovering a new chiral cocrystal system and performing chiral resolution via cocrystallization. Moreover, the study aims to gain further understanding of the molecular origins governing the chiral discrimination mechanism in this cocrystal system.

In this work, we demonstrate an enantiopurification route for RS-ETI using the chiral coformer 2-chloro-S-mandelic acid (S-2CLMA), an important cardiovascular drug precursor.[19] Characterizations of the new cocrystal and chiral resolution via the cocrystallization technique are illustrated in this paper. In addition, we elucidate for the first time the crystal structure of this chiral cocrystal by X-ray crystallography and analyze the intermolecular interactions via density functional theory (DFT) calculations to gain more insights into the observed enantiospecific behavior. The techniques and insights presented here could be useful to guide the design of enantiospecific cocrystallization and obtain an idea for screening of suitable coformers.

Materials

Starting Materials

2-Chloro-S-mandelic acid (S-2CLMA) and 2-chloro-R-mandelic acid (R-2CLMA) were purchased from TCI with a purity of >98%. Racemic etiracetam (RS-ETI) was obtained by racemization of S-ETI purchased from Acros following a procedure similar to that of Springuel et al.[20] Starting with 2 g of levetiracetam, 0.05 equiv of sodium methoxide was added to 2 mL of methanol. The solution was stirred at 50 °C for 24 h using a condenser to return any solvent that evaporated. Then, the solution was cooled in a bath at 5 °C to let the compound crystallize. The solid phase was filtered and washed with cold methanol before drying in a vacuum oven. HPLC and NMR were used to confirm that complete racemization had occurred with no degradation of the etiracetam. Deionized (DI) water (PURELAB Classic, 15.0 MΩ cm, type II) and acetonitrile (Honeywell, HPLC grade) were used as solvents without further purification.

Screening Experiment

Screening experiments of two API compounds, S-ETI and RS-oxiracetam, were done with various coformers using the neat grinding technique. An equimolar ratio of the mentioned APIs with various coformers was ground for either 2 or 3 h; then, the solid was characterized by XRPD. The pairs that have been examined are given in Table S1.

Liquid-Assisted Grinding for Cocrystal Creation

The cocrystal for characterization of the stable form was prepared using a liquid-assisted grinding technique with an equimolar ratio of S-ETI and S-2CLMA in the presence of a small amount of acetonitrile. In the experiment, 0.268 mmol S-2CLMA and S-ETI were mixed in a grinding tube with an addition of 10 μL of acetonitrile and two milling balls (7 mm diameter). The mixture was then ground using a Retsch MM200 ball mill at a frequency of 25 rounds/s for 60 min. Then, the solid was dried and characterized by XRPD and TGA.

Single Crystal of the Levetiracetam-2-Chloro-S-Mandelic Acid Cocrystal

A total of 0.2 g of S-ETI:S-2CLMA cocrystal from multiple cocrystal preparations was dissolved in 1 mL of chloroform (boiling point of 61.2 °C). After the cocrystal was completely dissolved, 1 mL of hexane (boiling point of 69 °C) was added as an antisolvent. Then, the vial was left under stirring under ambient conditions with the cap loosened to allow the solvents to slowly evaporate. Since the chloroform solvent has a lower boiling point than that of the antisolvent hexane, the molar ratio of the antisolvent to the solvent increased during the evaporation process. Single crystals formed after some of the solvent mixture evaporated. Finally, a single crystal suitable for X-ray crystallography was obtained.

Chiral Resolution

A total of 114 mg of racemic ETI was mixed with S-2CLMA in 1:1 mole ratio. Then, the solid mixture was dissolved in 0.5 mL of acetonitrile at 65 °C for 35 min. After the suspension was completely dissolved, the solution was cooled to −14 °C. The S-ETI:S-2CLMA cocrystal from liquid-assisted grinding was seeded into the vial after the solution reached the crystallization temperature. A cocrystal formed immediately after seeding, but to get a satisfying yield, the crystalline solid was collected 48 h after seeding. Then, the crystalline solid was washed with 2 mL of cooled diethyl ether and dried at 40 °C under vacuum. To analyze the enantiomeric excess and yield of the resolved enantiomer (S-ETI), 1 mg of the cocrystal was stirred for 1 h in 150 μL of diethyl ether, which is an appropriate amount of solvent to dissolve most of the S-2CLMA from the cocrystal while obtaining a good yield of S-ETI (Table S2). Diethyl ether was chosen as the solvent for this process since the solubility of S-ETI in this solvent is very low (less than 0.007 g/g of diethyl ether[21]) compared to the solubility of S-2CLMA (0.369 g/g of diethyl ether[22]), which makes it suitable for separating the components of the cocrystal. The solubilities of the components in other solvents that were considered are shown in the Supporting Information, Table S3. Afterward, the remaining solid was filtered and analyzed by HPLC using a Chiralcel OD column. Details of the HPLC technique are given in the Supporting Information.

Computational Details

For solid-state simulations, we used the experimental crystal structure of S-ETI:S-2CLMA determined in this work as a starting geometry. To explain why R-ETI:S-2CLMA is not observed experimentally, we employed crystal structure prediction (CSP) to scan its energy landscape via a quasi-random search.[23] To reduce the number of trial structures, we used a Monte Carlo procedure to preliminarily screen low-energy interacting pairs (synthons) using the Gmmx[24] module as implemented in PCModel v10.[25] The structures were then optimized using the Merck molecular force field (MMFF94).[26] The search is stopped when the maximum number of randomly generated synthons reaches 10 000 or when at least one pair reaches 50 duplicates. The 10 most stable synthons were used as the initial asymmetric unit. To generate random unit cells with a predefined space group, we used the Pyxtal program.[27] We constrained the initial cell volume per asymmetric unit to be comparable to that of the S-ETI:S-2CLMA cocrystal. To reduce the computational cost, we constrained the search space to the most common space groups for chiral cocrystals,[28] i.e., P1, P-1, P2, P21, P21/c, and P212121. The generated unit cells were then optimized using the semiempirical PM7 Hamiltonian with periodic boundary conditions as implemented in MOPAC v2016,[29,30] which has been shown to have good performance in predicting thermodynamic properties particularly for organic systems.[31,32] The structures of the five most stable candidates were further refined using periodic density functional theory (DFT) calculations as implemented in the CP2K package,[33] which uses a mixed Gaussian/plane-wave (GPW) basis set. The exchange and correlation energies were evaluated using the semilocal Perdew–Burke–Ernzerhof (PBE) form of the generalized gradient approximation (GGA) functional together with Grimme’s D3 van der Waals correction.[34,35] DZVP-MOLOPT basis sets and Goedecker–Teter–Hutter (GTH) pseudopotentials were selected in the calculations.[36,37] An energy cutoff of 500 Ry and relative energy cutoff of 80 Ry were found to be sufficient for energy convergence. Brillouin zone integration was sampled at the Γ point, which is sufficient due to the large size of the supercell (1 × 2 × 1 cell containing 180 atoms). Atomic structures were optimized using the BFGS algorithm[38] until the forces acting on each atom were less than 0.005 eV/Å. For self-consistent total energy calculations, the convergence criterion was set to 10–7 Hartrees for the crystal structure’s full relaxation (lattice and atomic position optimizations). To evaluate the thermal contributions to the enthalpy and free energy, we performed vibrational frequency analysis at the DFT-D3/revPBE/DZVP level of theory under harmonic approximation. The thermodynamic properties were then computed using the postprocessing tool TAMkin.[39]

Results and Discussion

Screening and Characterization

We have screened various resolving agents that could form enantiospecific cocrystals with S-ETI via grinding experiments (details shown in Table S1). We found that S-ETI forms a cocrystal with S-2CLMA but not with R-2CLMA. This is evidenced by the XRPD pattern shown in Figure a. The powder pattern of S-ETI:S-2CLMA is significantly different from that of S-ETI and S-2CLMA, suggesting the formation of a new crystalline phase. On the other hand, the powder pattern of S-ETI:R-2ClMA is comparable to that of S-ETI. Notice that the peaks corresponding to R-2CLMA disappeared. This is likely due to transformation into an amorphous phase due to the excess energy produced or the increase in temperature during the grinding experiment.[40,41] Nevertheless, this confirms that S-ETI:R-2CLMA is an enantiospecific cocrystal. To further confirm this, we performed thermogravimetric analysis whose results are shown in Figure b. Pure S-2CLMA and S-ETI have similar melting temperatures, 119 and 117 °C, respectively, whereas the cocrystal has a melting point of 98 °C, thereby confirming the formation of a new phase.

Figure 1

(a) Powder X-ray diffraction patterns of S-ETI, S-2CLMA, and S-ETI:S-2CLMA cocrystals and S-ETI:R-2CLMA. (b) Melting temperatures of S-ETI, S-2CLMA, and S-ETI:S-2CLMA cocrystals with 5 °C/min heating rate.

Crystal Structure Analysis

To elucidate the crystal structure of S-ETI:S-2CLMA, we performed single-crystal X-ray crystallography. To the best of our knowledge, this is the first time that the crystal structure of this cocrystal has been determined. The pertinent crystallographic data in Table show that S-ETI:R-2CLMA crystallizes in the monoclinic P21 space group. Moreover, the simulated peaks obtained from the single crystal exhibit excellent correlation with the experimental powder pattern (Figure S2). This suggests that the bulk powder obtained by batch crystallization is well-represented by the selected single crystal, thus ruling out the presence of other phases or polymorphs in the system. Furthermore, given that the system is enantiospecific, it is expected that the counter-enantiomer (R-ETI:R-2CLMA) would have a similar structure by virtue of symmetry.

Table 1

Crystallographic Data and Refinement Details of the S-ETI:S-2CLMA Cocrystal

crystal data
chemical formula	C16H21ClN2O5
Mr	356.8
temperature (K)	100
crystal system	monoclinic
space group	P21
a, b, c (Å)	11.6912 (7), 5.7422 (4), 13.8421 (9)
α, β, γ (deg)	90, 113.527 (2), 90
V (Å3)	852.02 (10)
Z	2
F (000)	376
density calculated (g/cm3)	1.391
radiation type	Mo Kα (λ = 0.71073)
μ (mm–1)	0.253
crystal size (mm3)	0.62 × 0.19 × 0.08
data collection
diffractometer	Bruker D8 Venture
absorption correction	multiscan (SADABS; Bruker, 2016)
Tmin, Tmax	0.681, 0.746
Rint	0.0325
refinement
goodness-of-fit on F2	1.057
R1, wR2 [I > = 2σ I]	0.028, 0.0688
R1, wR2 [all data]	0.0322, 0.0707
no. of reflections	17 385
no. of parameters	222
Δρmax, Δρmin (e Å–3)	0.38, −0.21
Flack value	0.041

To understand the underlying intermolecular interactions that stabilize the cocrystal, we further examined its crystal structure. The asymmetric unit in Figure a shows a ring-like R22(8) H-bonding motif between S-ETI and S-2CLMA. This motif is visible in the crystal packing along the ac-plane (Figure b) where S-ETI is shown in red and S-2CLMA is shown in blue. The perspective view along the same plane (Figure c) reveals that this ring-like motif propagates along the b-direction, which is affirmed by the crystal packing along the ab-plane exhibiting an infinite chain of H-bond network. Analysis of the geometric parameters reveals four important intermolecular H-bonding interactions reported in Table S4. The strongest of these is the 1.77 Å bond between the carboxylic hydrogen of S-2CLMA and the amide oxygen of S-ETI (Figure d). This is followed by the 1.99 Å bond between the hydroxyl hydrogen of S-2CLMA and pyrrolidone oxygen of S-ETI. The remaining H-bonds are due to the interaction between the amide of the S-ETI and the carboxylic oxygen of the S-2CLMA.

Figure 2

(a) Asymmetric unit of the S-ETI:S-2CLMA cocrystal; (b) packing along the ac-plane; (c) perspective view along the ac-plane; and (d) packing along the bc-plane.

Molecular Simulations

To understand why this cocrystal system behaves enantiospecifically, we generated and ranked the most probable crystal structure of R-ETI:S-2CLMA (the less-stable diastereomeric cocrystal) via molecular simulations, constrained within the most common space groups for chiral cocrystals (i.e., P1, P1̅, P2, P21, P21/c, and P212121).[28] The resulting energy landscape is reported in Figure . For comparison, the experimental crystal structures of enantiopure ETI, rac-ETI, S-2CLMA, and S-ETI:S-2CLMA are included. From a thermodynamic perspective, a cocrystal would likely form if its lattice energy is larger in magnitude than the sum of the lattice energies of the coformers (the gray line in Figure ). Interestingly, two of the predicted structures satisfy this threshold and are energetically close to the experimental S-ETI:S-2CLMA form (with around 2 and 4 kcal/mol difference, respectively, at the PM7 level of theory).

Figure 3

Energy landscape of the predicted crystal structures of R-ETI:S-2CLMA (in ▲) in comparison with experimental crystal structures (in ●) computed at the semiempirical PM7 level of theory (at 25 °C, 1 bar). The critical lattice energy (gray dashed line) is set as the sum of the conformer’s lattice energy (enantiopure ETI plus S-2CLMA). The color corresponds to the space group.

To refine the structure of the most stable hypothetical structure, we further relaxed its structure at the DFT-D3/revPBE/DZVP level of theory with periodic boundary conditions. Note that the dispersion-corrected DFT is generally considered more accurate than semiempirical PM7 but comes with a greater computational cost. The resulting hypothetical structure in comparison with the DFT-optimized experimental structure of the other stereoisomer is shown in Figure . Moreover, the two crystal structures are remarkably similar as evidenced by the resemblance of their simulated powder pattern (Figure S3).

Figure 4

DFT-optimized crystal structures of (a) hypothetical R-ETI:S-2CLMA and (b) experimental S-ETI:S-2CLMA. The H-bond lengths (hydrogen to acceptor) are shown in angstroms, and the chiral centers are indicated accordingly.

To evaluate the stabilities of the crystal structures at various temperatures, we performed vibrational analysis at the DFT-D3/revPBE/DZVP level of theory together with harmonic approximation. The difference in enthalpy ΔH and free energy ΔG between the predicted hypothetical structure (R-ETI:S-2CLMA) and the experimental structure (S-ETI:S-2CLMA) is reported in Figure S4. Interestingly, although the two crystal structures are remarkably similar, their stabilities are markedly different. At the selected temperature range, ΔH remains relatively constant at around −5.5 kcal/mol, which confirms that the experimentally observed cocrystal has stronger bond energies. The flat variation of ΔH with temperature suggests that the two cocrystals have similar specific heat capacities. On the other hand, ΔG is negative and increases with temperature, implying that the hypothetical structure is entropically favored (more disordered).

Consequently, the two cocrystals would have a closer magnitude of free energy at higher temperatures (below the melting point). This supports the idea that the strength of stereospecific interactions is highly directional; thus, such interactions would diminish when molecular vibration increases.

With such a difference in stability despite having a similar crystal packing, one might expect that the H-bond interaction distances of the predicted R-ETI:S-2CLMA cocrystal must be systematically longer than those of the stable S-ETI:S-2CLMA cocrystal. However, as shown in Figure , this is not the case, as some H-bonds of the unstable cocrystal are shorter. To further investigate this, we employed the Hirshfeld surface analysis, which is a popular approach to quantify differences in intermolecular interactions as implemented in CrystalExplorer.[42] The Hirshfeld surface can be defined as a region of space surrounding a particular molecule where its contribution to the total electron density is equal to that of its surroundings.[43] Intermolecular interactions can then be visualized by mapping the color of the surface with the normalized contact distance, dnorm, which compares the distance between two atoms across the surface to the combined van der Waals radii of the atoms.

The Hirshfeld surfaces of both cocrystals are shown in Figure a,b, and the corresponding fingerprint plots are shown in Figure c,d (Hirshfeld analysis of each atomic interaction is provided in Figure S5). The conspicuous red regions on the surface (close contacts) are due to O···H contacts, which appear as spikes (encircled in red) in the fingerprint plot. One notable difference between the two fingerprints is the occurrence of a small H···H spike in R-ETI:S-2CLMA, which does not occur in S-ETI: S-2CLMA. As a result, the H···H contribution on the Hirshfeld surface is slightly higher in R-ETI: S-2CLMA, as shown in Figure e, while the others (O···H, C···H, Cl···H) are essentially the same. With this, one might speculate that the H···H repulsion is the reason for the lower stability of R-ETI:S-2CLMA. However, the sum of di and de for the closest H···H contact is around 2.4 Å, which is within the range of attractive dispersion interactions for H···H interactions. Thus, this could not explain the observed difference in stability.

Figure 5

Hirshfeld surfaces of S-2CLMA in the cocrystals (a) R-ETI:S-2CLMA and (b) S-ETI:S-2CLMA and the corresponding 2D fingerprint plot of (c) R-ETI:S-2CLMA and (d) S-ETI:S-2CLMA (interior distance di and exterior distance de are in angstroms) and (e) comparison of contribution (in %) of selected close contacts to the Hirshfeld surface.

Upon further inspection of the structure, we observed that the centroid-to-centroid distance between phenyl rings of S-2CLMA is longer in R-ETI:S-2CLMA than in S-ETI:S-2CLMA, as shown in Figure .

Figure 6

T-shape aromatic interactions (magenta) and CH−π interaction (orange) in (a) R-ETI:S-2CLMA and (b) S-ETI:S-2CLMA. The angle between the ring normal is around 71° for both.

Note that the T-shape aromatic interaction (also known as the edge-to-face interaction) occurs when the distance between the ring centroids is around 5 Å and when the angle between the ring normals is greater than 50°.[44,45] On the other hand, the CH−π interaction occurs when a H attached to C has a distance of less than 3 Å from the aromatic ring centroid.[46] These definitions for T-shape aromatic interactions and CH−π interactions are also consistent with those of the PLATON program.[47] To quantify the strength of the T-shape aromatic interaction, we used the CCDC’s Aromatic Analyzer as implemented in Mercury,[48] which is based on a neural network model. From the distances between ring centroids and the relative orientation (angle between the ring normals), it calculates an interaction score ranging from 0 (no interaction) to 10 (strong interaction). Indeed, the aromatic interaction score of the stable S-ETI:S-2CLMA is higher than that of the predicted R-ETI:S-2CLMA cocrystal (Table S5), illustrating the important role of T-shape aromatic interactions and CH−π interactions in the stability of these cocrystals. Thus, we postulate that due to the change in the chirality of ETI in R-ETI:S-2CLMA, S-2CLMA would have to rotate the functional groups attached to its chiral center (−OH and −COOH groups) to maximize the H-bond interactions. However, in the process, the phenyl rings would have to reorient in a way that reduces the strength of T-shape aromatic interactions. This makes the alternative cocrystal less stable as supported by the lower lattice energy. Overall, our simulations reveal molecular-level insights that could rationalize the observed enantiospecific behavior of this cocrystal system.

Since many polymorphs of racemic ETI[49] have been found, the synthesized RS-ETI (form II) that we used in the separation experiment is confirmed by XRPD and NMR, as shown in Figures S6 and S7, respectively. The enantioseparation of S-ETI from its racemate was performed in acetonitrile with S-2CLMA as a coformer. Enantiomeric excess and yield of the product are defined in eqs and 2, respectively.

S-ETI:S-2CLMA is an incongruent cocrystal system due to the huge differences of the solubility of the two species in acetonitrile. Therefore, an equal ratio of S-ETI and S-2CLMA is not suitable for the production of the cocrystal. Varying the initial composition ratio between the API and coformer is then vital to obtain the optimum cocrystallization outcome.

Results show that a 1:1 ratio of rac-ETI and S-2CLMA is the optimal ratio since the resolution was achieved efficiently with an excellent purity (97% ee) and satisfactory yield (70%) of S-ETI, as shown in Table . Note that this is a 2:1 ratio of S-2CLMA:S-ETI. Increasing the amount of S-2CLMA further does not improve the experimental yield of S-ETI since S-2CLMA will crystallize together with the S-ETI:S-2CLMA cocrystal. This could be explained by a schematic ternary phase diagram, as shown in Figure S8, where the operation conditions exceed the region that allows only the cocrystal to crystallize as a stable phase. Alternatively, reducing the amount of S-2CLMA to be 0.5 times by mole of rac-ETI (1:1 ratio of cocrystal) is not favorable since only rac-ETI was acquired as a solid phase at this point. This is because of the differences in the solubility of S-2CLMA that was about 10 times higher than that of S-ETI in acetonitrile (Table S3). Therefore, at this ratio, pure RS-ETI is the most stable solid phase. Finally, the purity of the desired product (S-ETI) in all cases is considerably unchanged because the system is enantiosepecific, which means that only one enantiomer (S-ETI) cocrystallizes together with S-2CLMA.

Table 2

Enantiomeric Excess and Experimental Yield of Chiral Resolution Using S-2CLMA When Varying the Molar Ratio of RS-ETI and S-2CLMA

initial composition				final solid composition (HPLC)
ratio	RS-ETI (mg)	S-2CLMA (mg)	ACN (mL)	CoCry. weight (mg)	% ee (S-ETI)	S-ETIa (mg)	experimental yield (%) (based on S-ETI)
1:1	114	125	0.5	81.5	96.7	39.4	69.1
1:1.5	114	187	0.5	68.0	97.0	36.8	64.6
1:2	114	250	0.5	57.8	98.1	27.0	47.4

The total weight of the cocrystal was measured after washing and drying. The amount of S-ETI, R-ETI, and %ee was calculated based on the total weight of the cocrystal and ETI composition in the cocrystal by chiral HPLC following eqs and 2.

Conclusions

In this work, we demonstrate the enantiospecific cocrystallization of etiracetam (ETI) with 2-chloromandelic acid (2-CLMA). In this system, S-2CLMA forms a cocrystal with S-ETI but not with R-ETI. The crystal structure of the new cocrystal has been determined and analyzed. Exploiting the enantiospecific cocrystallization behavior, we showed that the resolution of S-ETI from rac-ETI can be achieved efficiently with 96.7% purity and 69.1% yield. To understand why the R-ETI:S-2CLMA cocrystal does not form, we employed crystal structure prediction and molecular simulations. Although the predicted hypothetical structure of R-ETI:S-2CLMA is strikingly similar to that of the experimentally observed R-ETI:S-2CLMA, a large lattice energy difference of 5.5 kcal/mol per cocrystal formula unit was obtained. While both cocrystals have relatively similar H-bonding geometries and Hirshfeld fingerprints, we found that the aromatic interactions in S-ETI:S-2CLMA are more favorable than in R-ETI:S-2CLMA, leading to the observed enantiospecific behavior.

For further process design, all compounds in the cocrystallization process could be recycled after the first crystallization stage. In the experiment, S-2CLMA can be recycled practicably by washing from the surface of the solid cocrystal with diethyl ether, drying to remove the solvent, and adding back to the reactor, accordingly. For R-ETI, racemization could be a process to prepare a racemic solution from the remaining enriched solution and then use it further in the next batch of cocrystallization.