Rapid deracemization through solvent cycling: proof-of-concept using a racemizable conglomerate clopidogrel precursor

We demonstrate that a conglomerate-forming clopidogrel precursor undergoing solution phase racemization can be deracemized through cyclic solvent removal and re-addition. We establish that the combination of slow growth and fast dissolution of crystals is ideal for rapid deracemization, which we achieve by repurposing a Soxhlet apparatus to realize the slow removal and fast re-addition of solvent autonomously.


B. HPLC analysis
Sampling and sample preparation Two distinct sampling methods were used. One method for system sampling and one method for solid phase sampling. For both methods, samples are always taken at the end of each solvent cycle (i.e. directly after re-addition, when the solvent in the boiling flask is at maximum).
▪ System sampling A 1 mL syringe fitted with a long needle is used to collect ~ 50 µL of slurry from the boiling flask. This system sample is then dissolved in 7 mL of IPA. After brief ultrasonication (5 minutes), 1 mL is transferred to a 2 mL HPLC vial and directly submitted for HPLC analysis.
▪ Solid phase sampling A 1 mL syringe fitted with a long needle is used to collect ~ 250 µL of slurry from the boiling flask. The collected suspension was cast on top of filter paper laid down on glass filter connected to a vacuum filtration set-up (whilst under active vacuum). Two separate samples of the solids were taken using a Pasteur pipette, dissolved in 1.5 mL of IPA by vortex and ultrasonication (10 minutes, 2 mL HPLC vial) and subsequently analysed by HPLC.
HPLC method HPLC analysis was performed on a chiral column (CHIRALPAK IA (250 x 4.6 mm, 5μm)) with a mobile phase consisting of n-heptane and 1-propanol. For compound 1 (Fig. S-1), the eluent is mixed in a 7:3 ratio (heptane:IPA). The flow rate was 0.7 mL/min, injection volume 4 µL, and detection was performed by UV-detector (wavelength: 220 nm). Each run had a total time of 12 minutes. S-3

C. Solvent Cycling-Induced Deracemization using a Soxhlet-Apparatus
Preliminaries For the solvent cycling experiments, we use a two-neck round bottom flask of 50 mL, to which we attach a Soxhlet-apparatus and a tap-water cooled condenser. The other neck is closed with a septum to allow sampling during the solvent cycling process. The Soxhlet-apparatus has a nominal volume of 30 mL, which is decreased to 15 mL by adding a closed glass vial to the sample compartment. Aluminium foil is used to prevent unwanted cooling of the glassware by air during the experiments. A batch of 250 mL of the binary solvent mixture (225 mL diethyl ether and 25 mL acetone) was prepared for use throughout the various experiments. During the solvent cycling experiments, unless stated otherwise, 28 g of PTFE beads and a stirring bar are present in the round-bottom flask (boiling flask) and stirring is performed on the highest possible setting. Experiments are performed under slight nitrogen pressure (Schlenk line) to reduce the possibility of solvent evaporation out of the system as far as possible.
Slurry preparation Starting material of 1 of a certain enantiomeric composition was prepared by mixing the required amounts of enantiopure 1 and racemic 1 in a pestle and mortar and ground until homogenous. The resulting ee of the starting material was always confirmed by HPLC (an example is shown in Fig. S-2).
To prepare the slurry, a 20 mL vial is charged with 800 mg of compound 1, in the desired enantiomeric composition. Subsequently, 10 mL of the solvent is added and the resulting suspension is sonicated for 30 minutes to obtain a homogenous slurry. The slurry is then transferred to the 50 mL round-bottom flask and an extra 15 mL of solvent is added to the flask. Seminal solvent cycling experiment Starting material of 10% ee in (R)-1 was used to prepare the slurry. The round-bottom flask was connected to the Soxhlet-apparatus and the condenser as described and lowered into a water bath set at 50 o C on a hotplate (maximum stirring setting). Once the solvent cycles attained a consistent time-interval (approximately 4 to 5 minutes per cycle), 40 μL of racemization catalyst (DBU) was added in 2.5 mL of solvent by syringe through the septum. After 18 hours, the solid phase and system were sampled as previously described. The seminal solvent cycling experiment was deemed successful, since virtually all material had converted to the major enantiomer. The final solid phase chromatogram is displayed as Fig. S-3 (>99% ee in (R)-1). After cooling back to 20 o C, the contents of the boiling flask were filtered to afford virtually enantiopure material with 90% yield as a white residue.

Kinetics of Solvent Cycling-Induced Deracemization
Having demonstrated the potential of Solvent Cycling-Induced Deracemization, the experiment was repeated with starting material of 10%, 20% and 50% ee in (R)-1 and 50% in (S)-1. At various points after starting the experiment, solid phase and system samples were taken ( Fig. S-4). These experiments show that full deracemization can be achieved within 2, 3.5, and 6 hours respectively.
N.B. The system phase ee is always lower than the solid phase ee since a number of molecules is in the racemic liquid phases (soluble). This is also a reason why solubility should not be too high while cyclic mass transfer should be optimized. (triangles), 20% (diamonds) and 50% (squares) in R (red) or S (blue). Exponential kinetic fits are represented by solid lines. The left graph shows the solid phase ee and the right graph shows the system ee.

S-5
Control experiments In order to confirm that solvent cycling is the cause of the enantiomeric enrichment observed, and not-for example-attrition by the PTFE beads or stirring or ripening phenomena accelerated at higher temperatures, we conducted two specific control experiments. The kinetic solvent cycling experiment was repeated starting with material 10% ee in (R)-1.
In both experiments, the Soxhlet-apparatus is removed and the condenser is placed directly on top of the boiling flask. In this way, no-or only minimal trough refluxing-solvent cycling is achieved. These experiments were then performed in the presence of either the PTFE spheres or glass beads, since glass beads were used in the first demonstration of attrition-enhanced deracemization [2].
The results are shown in Fig. S-5. Comparing of these experiments clearly demonstrates that solvent cycling, using the Soxhlet-apparatus, is vastly superior in kinetics to the two control experiments. In fact, on the relevant timescales of deracemization through solvent cycling, almost no enrichment is produced by the control experiment with PTFE beads and refluxing alone. This shows that it is the removal and re-addition of the solvent that is responsible for the enantiomeric enrichment reported here. Moreover, attrition by glass beads clearly outperforms the PTFE spheres (but still does not reach the kinetics of solvent cycling by any stretch). This shows that the soft PTFE beads are no strong source of attrition as compared to glass beads, the standard in attrition-enhanced deracemization experiments, and their impact is indeed minimal.

D. X-Ray Powder Diffraction Analyses
In order to exclude the presence of metastable polymorphs or solvates as key intermediates in the deracemization and confirm that no new polymorphs were formed as a starting material or final product during solvent cycling in the designed solvent, X-Ray Powder Diffraction analyses were performed on racemic reference material of 1 (unexposed), the sonicated initial slurry (before starting the solvent cycling experiment, but having been exposed to the solvent) and the final deracemized product (having been exposed to the process). For each sample, 20 -50 mg of powder was placed on a 1 cm 2 aluminium square and pressed into a cake. X-Ray powder diffraction patterns of each sample were measured using a Bruker D2 Phaser with a Cu X-ray source (Cu K-α, λ = 1.5418 Å). The resulting patterns are displayed in Fig. S-6. These patterns are identical for all three samples, indicating that the presence of metastable polymorphs or solvates as key intermediates in the deracemization can be excluded and no new polymorphs were formed as a starting material or final product during solvent cycling in the designed solvent. Fig. S-6. XRPD patterns for the three samples of interest: reference material of (R,S)-1, the sonicated initial slurry (before starting the solvent cycling experiment, but having been exposed to the solvent) and the final deracemized product (having been exposed to the process). We have also included a Single Crystal XRD (simulated XRPD) reference from the existing CSD entry OHODUL [3].

E. Simulations for the Design of the Binary Solvent Mixture
In order to design a solvent with tailor-made composition which has a low boiling point and relatively low solubility at the boiling point, while still providing reasonably fast racemization kinetics, we use a binary solvent mixture of diethyl ether and acetone. To guide our choice in composition, we have simulated solvent cycles for various compositions, boiling points and solvent cycling geometries.
Our model consists of two compartments: the boiling flask, from which solvent is removed by evaporation, and the sample compartment, which adds all solvent condensed back to the boiling flask once a critical volume is reachedanalogous to the Soxhlet-apparatus set-up.

Premise and Assumptions
In our model, we take a range of considerations into account. Not only the properties of the bare solvent are relevant. Since the solvent composition changes during the process, also the properties during the deracemization process should be considered. This means that: ✓ We consider that the composition of the vapour (the evaporated solvent) depends on the composition of the solvent that is boiling (in the boiling flask). ✓ We consider that the composition in the boiling flask changes as a result. ✓ We consider that the actual solubility of compound 1 and the boiling point of the solvent in the boiling flask changes over time as the composition of the solvent changes. ✓ We consider that that changes in solubility and volume mean that crystals have to grow or dissolve, but these processes are not instantaneous and have a mass transfer rate.
The relative amount of cyclic mass transfer is small compared to the total concentration in the solvent. We assume that changes in the boiling point and vapour pressure as a function of solute concentration can thus be neglected for these simulations.

Underlying Datasets
In order to accurately simulate the solvent cycling experiment for the binary diethyl etheracetone system, we required solubility data as a function of solvent composition. Therefore, we determined the solubility for various compositions and temperatures of the solvent using the method previously reported [1]. The data, shown in Fig. S-7, were taken as empirical basis for the simulations. where ρ is the density of the solvent (function of composition), R is the gas constant and T is the temperature in K.

S-9
▪ Removal and re-addition of solvent Material is removed from or re-added to the boiling flask using the following mass transfer equation: dV/dt = rate of mass transfer where the rate is either that of removal or re-addition. The addition or removal in case of the sample compartment follows the negative sign to uphold the mass balance.

▪ Dissolution and Growth of crystals
For crystal dissolution and growth, the supersaturation is first calculated based on solubility: If S < 1, dissolution will take place. If S > 1, growth will take place. These processes are modelled following the approximation dm/dt = rategrowth . where Smax is the so-called 'maximum supersaturation' at which the steady-state growth rate is reached and dm/dt is the mass transfer rate between solid and liquid phase.
▪ Time-marching implementation Separate volumes are tracked for acetone and diethyl ether for each of the two compartments in the implementation. At time-step 0, all parameters in the system are initialized and initial compositions are set. For each new time-step, the procedure is as follows.
First, we calculate the boiling point. If the boiling point is below or equal to the system temperature, we execute a solvent removal step. The amount of solvent to be removed is calculated and the composition of the removed solvent is based on the VLE. The removed solvent is then added to the sample compartment.
Second, we check whether the volume in the sample compartment is above the critical volume or if re-addition was busy in the previous step (if the volume of the sample compartment is at or below the 'zero-volume', we stop re-addition). If re-addition is required, we move part of the solvent from the sample compartment back into the boiling flask.
Third, we recalculate compositions of the boiling flask and the sample compartment. Based on this new data, we calculate the solubilities and execute crystal growth or dissolution.

▪ Simulation parameters
Simulations were nominally performed with the following parameters. Temperature of the system was set at 50 o C. We used a sample compartment of 15 mL, initial solvent volume in the boiling flask of 25 mL and 1 mL of zero-volume (vapour lost in the system). We routinely used a maximum growth rate of 0.1 mg/s, dissolution rate of 1 mg/s, maximum supersaturation of 1.5 (c/c*), 1 mL/s solvent re-addition rate and 0.01 mL/s solvent removal rate. The amount of compound in the simulations was set at 800 mg. For time-marching, we used 15000 timesteps and each timestep was 0.5 seconds.