Enantioselective extraction of unprotected amino acids coupled with racemization

Scalable and economical methods for the production of optically pure amino acids, both natural and unnatural, are essential for their use as synthetic building blocks. Currently, enzymatic dynamic kinetic resolution (DKR) underpins some of the most effective processes. Here we report the development of enantioselective extraction coupled with racemization (EECR) for the chirality conversion of underivatized amino acids. In this process, the catalytic racemization of amino acids in a basic aqueous solution is coupled with the selective extraction of one enantiomer into an organic layer. Back-extraction from the organic layer to an acidic aqueous solution then completes the deracemization of the amino acid. The automation of the EECR process in a recycling flow reactor is also demonstrated. Continuous EECR is made possible by the sterically hindered chiral ketone extractant 5, which prevents the coextraction of the copper racemization catalyst because of its nonplanar geometry. Furthermore, the extractant 5 unexpectedly forms imines with amino acids faster and with greater enantioselectivity than less bulky derivatives, even though 5 cannot participate in intramolecular resonance-assisted hydrogen bonding. These features may allow EECR to challenge the preponderance of enzymatic DKR in the production of enantiomerically enriched amino acids.


General method for the preparation of extractant (R)-5 and its analogues
(R)-5: A solution of (R)-3-tert-butyl-2,2′-dihydroxy-1,1′-binaphthyl ketone (3.7 g, 10 mmol) in DMF (50 mL) was added dropwise to a stirred slurry of NaH (60% in mineral oil, 0.44 g, 11 mmol) in DMF (50 mL) at 0˚C. After 2 h of stirring at this temperature, 3-phenyluryl-benzyl bromide (3.36 g, 11 mmol) was added, and the stirring was continued at room temperature overnight. The reaction was then quenched with a saturated NH4Cl solution and extracted with ethyl acetate and water. After evaporation of the volatiles from the combined organic layers, the crude product was purified by column chromatography on silica gel using EA/hexane (1:7) as the eluent to furnish compound (R)-5 as a yellow solid ( (R)-3 and (R)-4 were similarly prepared using the above procedure.

Extraction with (R)-5 for representative amino acids
The experiments were carried out following the typical ELLE procedure (see Supplementary Section 2.1) using the chiral extractants (R)-5. After completion, the organic layers were analyzed by 1 H-NMR. Then, the imines were hydrolyzed by stirring the organic layer in the presence of aqueous 2N HCl (1.5 mL), and the purity of the back-extracted amino acids was assayed by chiral HPLC.

Robustness and recyclability of the extractant (R)-5
An experiment was performed to establish the robustness and recyclability of the chiral extractant (R)-5 during the course of ELLE experiments. The ELLE of DL-Phe was performed, recycling the organic layer containing (R)-5 and Aliquat 336 after acid hydrolysis, over 20 cycles. The composition of the organic layer was analyzed by 1 H-NMR, as shown in Supplementary Figure 19 shown below. No sign of degradation or decomposition of (R)-5 could be detected over the course of the 20 ELLE cycles. Figure 19. 1 H NMR spectra of the organic layer in the ELLE of DL-Phe with (R)-5: before the first extraction cycle (top), following the first extraction and acid hydrolysis cycle (middle), and following the completion of 20 ELLE cycles (bottom). S27

Hydrolysis (back-extraction) kinetics
The kinetics of back-extraction of Phe was tested using either 1 eq 2 N HCl, or 2 eq  The successful racemization of amino acids in the aqueous layer is a critical factor to achieve EECR with acceptable performance and timescale. We found that small catalytic amounts of PLP and Cu 2+ ions were sufficient to catalyze the racemization of amino acids in the aqueous layer at acceptable rates. Supplementary Figure 21 shows representative HPLC kinetics data for the racemization of phenylalanine over

Comparison of the transfer of Cu 2+ from organic layer to aqueous layer in the EECR with (R)-1 and (R)-5 as extractors
As shown in the Supplementary Figure 23, when an EECR process was attempted with (R)-1 as the extractant, the Cu 2+ ions were transferred to organic layer, as evidenced by the change of the color of the organic layer from yellow to dark blue. The sequestration of the Cu ions in the organic layer was also found to shut down the catalysis of the racemization of the amino acid. Figure 23. Photographs of the EECR process using (R)-1, illustrating the sequestration of the Cu ions in the organic layer (top). Proposed rationale for the complexation of Cu ions by the coplanar, chelating ligand formed between (R)-1 and the amino acid, responsible for the transfer of the Cu ions into the organic layer. S32 As shown in the Supplementary Figure 24, by contrast, when the EECR process was carried out with (R)-5 as the extractant, no color change was perceived in the organic layer over the course of 5 EECR cycles. No leeching of Cu ions was detected by in the organic layer by atomic absorption spectroscopy ([Cu] ≤ 1 ppm). Furthermore, the racemization catalyst remained active in the aqueous layer. Figure 24. Photographs of the EECR process using (R)-5, illustrating the lack of phase-transfer of the Cu ions from the aqueous to the organic layer (top). Proposed rationale for the absence of complexation of Cu ions by the non-coplanar, imine formed between (R)-5 and the amino acid. S33

EECR of 1.1 equiv DL-Phe with (R)-5, single extraction.
The organic layer was prepared by dissolving (R)-5 (0.60 g, 1.0 mmol) and 1.2 eq of Aliquat 336 in CDCl3 (6 mL). The aqueous layer was prepared by dissolving DL-phenylalanine (0.18 g, 1.1 mmol), NaOH (0.048 g, 1.2 mmol), PLP (0.0027 g, 0.011 mmol) and CuSO4 (0.0018 g, 0.011 mmol) in H2O (1 mL). The two solutions were vigorously stirred in a 10-mL vial, and 1 H NMR spectra of the separated organic layer were acquired after 1h, 2h, 4h, 12h and 24h. The results are shown in Supplementary Figure 25. The extraction yield after 24h, as estimated by NMR was of 68% based on the extractor (R)-5 and 62% based on the amino acid. The NMR study shows that the mismatched imine (R)-5-D-Phe is present as a minor constituent at the beginning of the EECR experiment, but later disappears. The qualitative kinetics for the formation and extraction of the imine (R)-5-L-Phe (and therefore (S)-5-D-Phe) may be estimated on the basis of these results.

EECR of 4 equiv D-Phe (4 equiv) with (R)-5, single extraction
The organic layer was prepared by dissolving (R)-5 (0.90 g, 1.5 mmol) and 1.2 eq of Aliquat 336 in CDCl3 (10 mL). The aqueous layer was prepared by dissolving D-phenylalanine (1.0 g, 6.0 mmol), NaOH (0.26 g, 6.6 mmol), PLP (0.016 g, 0.060 mmol) and CuSO4 (0.010 g, 0.060 mmol) in H2O (5 mL). The two solutions were vigorously stirred in a 20-mL vial, and 1 H NMR spectra of the separated organic layer were acquired after 0.5, 1h, 2h, 3h, 6h, and 24h. The results are shown in Supplementary Figure 26. The results indicate that imine formation is complete within 6h. The NMR study also shows that the mismatched imine (R)-5-D-Phe is present as a minor constituent at the beginning of the EECR experiment, but later disappears. On the basis of these results, the latter EECR experiments were carried out over ≥ 6h to maximize the selectivity of process.
Hydrolysis (back-extraction) Stage: The separated organic layer was then hydrolyzed by vigorously stirring in the presence of 2.0 N aqueous HCl (2.5 mL) for 4 h. After separation of the layers, the decanted organic layer containing (R)-5 and Aliquat 336 was recombined with the previous aqueous layer containing the amino acid to carry out a second EECR cycle.
The two stages -extraction and hydrolysis -were repeated for a total of 4 cycles without further addition of DL-phenylalanine. Neutralization of the combined aqueous layer from the acidic hydrolysis with 2N NaOH induced precipitation of the amino acid, which was filtered, washed with cold water and ethanol, and finally dried to obtain L-Phe (0.72 g, 4.4 mmol) with an enantiopurity greater than 98% e.r. The isolated yield of L-Phe is 73% based on the amino acid, and 293% with respect to the extractant (R)-5.

EECR of 4 equiv DL-Met (4 equiv) with (R)-5, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated using DL-Met (0.91 g, 6.1 mmol) in place of DL-Phe. Over 4 extraction and hydrolysis cycles, L-Met (0.63 g, 4.2 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 70% based on the amino acid, and 280% with respect to the extractant (R)-5. Figure 28. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Met (4 equiv) with (R)-5 (no DL-Met added in cycles #2-4) upon each EECR cycle. Note that the NMR yield of the extracted imine (R)-5-L-Met decreases after each cycle as the total [Met] remaining in the aqueous layer decreases. S39

EECR of DL-naphthylalanine (Nal) (4 equiv) with (R)-5, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated using DL-Nal (1.31 g, 6.1 mmol) in place of DL-Phe. Over 4 extraction and hydrolysis cycles, L-Nal (0.94 g, 4.2 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 73% based on the amino acid, and 293% with respect to the extractant (R)-5. Figure 29. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Nal (4 equiv) with (R)-5 (no DL-Nal added in cycles #2-4) upon each EECR cycle. Note that the NMR yield of the extracted imine (R)-5-L-Nal decreases after each cycle as the total [Nal] remaining in the aqueous layer decreases. S40

EECR of DL-isoleucine (Ile) (4 equiv) with (R)-5, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated using DL-Ile in place of DL-Phe. Over 4 extraction and hydrolysis cycles, L-Ile (0.57 g, 4.4 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 72.1% based on the amino acid, and 288% with respect to the extractant (R)-5. Figure 30. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Ile (4 equiv) with (R)-5 upon each EECR cycle (no DL-Ile was added in cycles #2-4). Note that the NMR yield of the extracted imine (R)-5-L-Ile decreases after each cycle as the total [Ile] remaining in the aqueous layer decreases. S41

EECR of DL-leucine (Leu) (4 equiv) with (R)-5, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated using DL-Leu in place of DL-Phe. Over 4 extraction and hydrolysis cycles, L-Leu (0.57 g, 4.4 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 72.1% based on the amino acid, and 288% with respect to the extractant (R)-5. Figure 31. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Leu (4 equiv) with (R)-5 upon each EECR cycle (no DL-Leu was added in cycles #2-4). Note that the NMR yield of the extracted imine (R)-5-L-Leu decreases after each cycle as the total [Leu] remaining in the aqueous layer decreases. S42

EECR of DL-valine (Val) (4 equiv) with (R)-5, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated using DL-Val in place of DL-Phe. Over 4 extraction and hydrolysis cycles, L-Val (0.54 g, 4.6 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 75.1% based on the amino acid, and 300% with respect to the extractant (R)-5. Figure 32. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Val (4 equiv) with (R)-5 upon each EECR cycle (no DL-Val was added in cycles #2-4). Note that the NMR yield of the extracted imine (R)-5-L-Val decreases after each cycle as the total [Val] remaining in the aqueous layer decreases. S43

EECR of DL-tryptophan (Trp) (4 equiv) with (R)-5, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated using DL-Trp in place of DL-Phe. Over 4 extraction and hydrolysis cycles, L-Trp (0.93 g, 4.5 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 74.5% based on the amino acid, and 298% with respect to the extractant (R)-5. Figure 33. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Trp (4 equiv) with (R)-5 upon each EECR cycle (no DL-Trp was added in cycles #2-4). Note that the NMR yield of the extracted imine (R)-5-L-Trp decreases after each cycle as the total [Trp] remaining in the aqueous layer decreases. S44

EECR of DL-Phe (4 equiv) with (R)-5 with toluene solvent, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated for DL-Phe using toluene as the solvent in place of CDCl3. Over 4 extraction and hydrolysis cycles, L-Phe (0.76 g, 4.6 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 75.3% based on the amino acid, and 301% with respect to the extractant (R)-5. Figure 34. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Phe (4 equiv) with (R)-5 upon each EECR cycle (no DL-Phe was added in cycles #2-4). The organic solvent toluene was used. Note that the NMR yield of the extracted imine (R)-5-L-Phe decreases after each cycle as the total [Phe] remaining in the aqueous layer decreases. S45

EECR of DL-Phe (4 equiv) with (R)-5 using tert-butyl methyl ether (MTBE) as the solvent, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated for DL-Phe using tert-butyl methyl ether as the solvent in place of CDCl3. Over 4 extraction and hydrolysis cycles, L-Phe (0.77 g, 4.7 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 76.3% based on the amino acid, and 305% with respect to the extractant (R)-5. Figure 35. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Phe (4 equiv) with (R)-5 upon each EECR cycle (no DL-Phe was added in cycles #2-4). The organic solvent MTBE was used. Note that the NMR yield of the extracted imine (R)-5-L-Phe decreases after each cycle as the total [Phe] remaining in the aqueous layer decreases. S46

EECR of DL-Phe (4 equiv) with (R)-5 with 2-methyltetrahydrofuran (MeTHF) solvent, 4 repeated extractions and hydrolyses
The procedure employed for the EECR experiment described in section 6.3.3 was repeated for DL-Phe using 2-methyltetrahydrofuran as the solvent in place of CDCl3. Over 4 extraction and hydrolysis cycles, L-Phe (0.74 g, 4.5 mmol) with an enantiopurity greater than 98% e.r. was isolated, corresponding to a yield 73.2% based on the amino acid, and 293% with respect to the extractant (R)-5. Figure 36. 1 H NMR monitoring of the separated organic layer in the EECR of DL-Phe (4 equiv) with (R)-5 upon each EECR cycle (no DL-Phe was added in cycles #2-4). The organic solvent MeTHF was used. Note that the NMR yield of the extracted imine (R)-5-L-Phe decreases after each cycle as the total [Phe] remaining in the aqueous layer decreases. S47

Hydrolysis (back-extraction) Stage:
The separated organic layer was then hydrolyzed by vigorously stirring in the presence of 2.0 N aqueous HCl (2.0 mL) for 2 h at room temperature. After separation of the layers, the decanted organic layer containing the cleanly recovered (R)-5 and Aliquat 336 was recombined with the previous aqueous layer containing the amino acid to carry out a second EECR cycle.
The two stages -extraction and hydrolysis -were repeated for a total of 5 cycles with the addition of one equivalent of the sodium salt of the amino acid [e.g. D-Phe (0.165 g, 1.0 mmol) and NaOH (0.04 g, 1.0 mmol)] to replenish the aqueous layer after each cycle. Neutralization of the combined aqueous layer from the acidic hydrolysis with 2N NaOH induced precipitation of the amino acid, which was filtered, washed with cold water and ethanol, and finally dried to obtain L-Phe (0.68 g, 4.1 mmol) with an enantiopurity greater than 98% e.r. The isolated yield of L-Phe is 51% based on the total amino acid, and 410% with respect to the extractant (R)-5.  Figure 38-(c). HPLC chromatogram of the residual aqueous layer after 5 cycles of extraction and hydrolysis in the EECR of D-Phe with (R)-5 (1 equiv D-Phe was added in cycles #2-5), indicating that the racemization catalyst remains active over the duration of the experiment.

EECR of other amino acids showing the conversion of D-AAs to L-AAs
The procedure employed for EECR experiment described in section 6.3.13 was repeated using D-Ala, D-Leu, D-Ile, D-Val, D-Met and D-Trp, respectively, in place of D-Phe. The 1 H NMR for the organic layer after first cycle of EECR are shown in Supplementary Figure 39. These results demonstrate that D-AAs are successfully converted to L-AAs during the EECR experiment. Table 3 lists the imine formation yields and enantioselectivities assessed from the 1 H NMR spectra shown in Supplementary Figure 38

Continuous EECR of L-Phe with (S)-5.
The stepwise EECR approach described in Supplementary Sections 6.4 and 6.5 requires tedious manual work and induces losses during each operation. To address these issues was designed the continuous reactor shown in Figure 4 of the manuscript.

Preparation of EECR solutions
Extraction (EECR) stage: The aqueous layer was prepared by dissolving L-Phe Transfer rate of pumps 1-5: 0.15 mL/min

Continuously recycling EECR reactor operation
After continuous operation of the reactor for 5h, one additional equivalent of L-phenylalanine (1.25 g, 7.6 mmol) and NaOH (0.3 g, 7.6 mmol) were added to extraction (EECR) stage. The addition was repeated again after 5h additional operation. The reactor was finally operated continuously for 38 h more without further addition of L-Phe/NaOH. At the end of the run, the collected aqueous layer in the hydrolysis stage was neutralized with dilute aq. NaOH, and the precipitated amino acid was washed with cold water, ethanol and dried to yield D-Phe (4.9 g, 29.7 mmol) with an enantiopurity >98%. The total yield was 390% based on (S)-5, and 65% based on the phenylalanine used.

Extraction efficiency and stereoselectivity during the continuous EECR.
Aliquots of the organic layer within the continuous EECR reactor were taken at regular intervals and analyzed by 1 H-NMR to assess the efficiency and selectivity of the formation of the (S)-5-D-Phe imine (Supplementary Figure 40). Extraction yields were found to be maintained in the 90-95% range with respect to the total (S)-5, and the stereoselectivity was found to be exclusive (>99:1 D:L). These results further confirm the efficient racemization of Phe in the aqueous layer.
No significant changes in the extraction yields or stereoselectivity were observed over the ca. 40h run, suggesting that degradation of the amino acids, extractant or racemization catalyst must be negligible over this timescale. Moreover, 1  The level of computation to obtain the final geometries and energies for all conformers in this study is the density functional theory supplemented with the Grimme's dispersion (namely D3) correction [3]. The BLYP functional, which uses the Becke's exchange [4] and the Lee-Yang-Parr's correlation [5], and the Pople's 6-311G(d) basis set were employed for the final optimization. We will refer this method to as BLYP-D3/6-311G*. We chose this method since the BLYP-D3 gives a minimal mean absolute deviation when tested for the S22 benchmark dataset of DNA base pairs, amino acid pairs and small model complexes [6].
The input geometries for the final optimization with the BLYP-D3/6-311G* were prepared as follows. In the first step, the crystallographic geometry of (R)-5-L-Phe anion, which is shown in Fig. 3 of the main article, was modified by replacing each one of the tert-butyl CH3's with one H to provide each seed geometry for (R)-4-L-Phe.
In similar ways, the seed geometries for (R)-3-L-Phe, (R)-2-L-Phe, and (R)-1-L-Phe were prepared. The seed geometries for (R)-5-L-Val and (R)-5-L-Ala were prepared similarly but these times the phenyl group in the amino acids were replaced by CH3 and H, respectively. The seed geometries for the D-amino acid counterpart were prepared by exchanging -H and the side group of the corresponding amino acid moiety. The seed geometries for the (R)-n, (n=1-5) were prepared after removing amino acid atoms from the geometries of the corresponding imine anion and adding one carbonyl oxygen.
In the second step, each seed geometry mentioned above was relaxed under molecular dynamics (MD) run for 10 ps at the density functional tight binding (DFTB) level of theory [4]. During the MD run the bath temperature was varied between 0.1 K and 290 K. This temperature range was needed to cover a wide range of geometry variation including the cis  trans conversion in the uryl moiety.
Occasionally, the (R)  (S) interconversion was induced. However, as (S)-L-amino acid is a mirror image of (R)-D-amino amino acid, the result was shared by the two S58 mirror images. Typically, one seed geometry was run for 30 cycles of 10 ps MD with updated initial geometries. Then, several low energy geometries from 30 local geometries were selected for further optimization in the third step.
The third step was the pre-optimization with the BLYP-D3/6-31G level of theory.
About 15 to 50 geometries per each imine anion or (R)-n were tried for the pre-optimization and those geometries whose energy was within about 10 kcal/mol to the global minimum value underwent the final optimization.

Geometries and energies for (R)-n
The geometries for the (R)-n (n=1-5) conformers are shown in Supplementary   Figures 43 through 47 The relative values of the electronic part of the energy (Ee) and its zero-point energy (ZPE) correction (E0 = Ee +ZPE) in kcal/mol are shown in parenthesis and square brackets, respectively. These relative values will be denoted with Ee and E0, respectively. Only those conformers whose Ee or E0 is less than 2 kcal/mol are shown and labelled with the alphabetic order.  In general, the energies of the (R)-5-L-amino acids are much lower than those of (R)-5-D-amino acids. This reflects the steric effect due to the bulky side group of the amino acid. This effect is reduced as the size of the side group is reduced. In other words, the energy difference between the imine anion with L-and D-amino acids is reduced in the order Phe > Val > Ala. S62 Supplementary Figure 48. Geometries of the two lowest energy conformers of (R)-5-L-Phe and the two lowest energy conformer of (R)-5-D-Phe optimized at the BLYP-D3/6-311G* level of theory. The third lowest conformer of (R)-5-L-Phe has Ee and E0 values higher by 3.69 and 3.37 kcal/mol than those of (R)-5-L-Phe-A and is not shown. The third lowest conformer of (R)-5-D-Phe has Ee and E0 values higher by 6.43 and 6.93 kcal/mol than those of (R)-5-L-Phe-A and is not shown. Figure 49. Geometries of the three lowest energy conformers of (R)-5-L-Ala and the two lowest energy conformer of (R)-5-D-Ala. In addition, the values of the two conformers labelled as X are included in Supplementary Table 4. The label X means that the corresponding conformer is in category I, while the other low energy ones are all in category II, and is not ranked S64 since the pre-optimized geometry is derived by replacing three H's in a stable geometry of (R)-1-Phe. The energies of the X conformers are very high because of the steric hindrance between the tert-butyl group of (R)-5 and the side group of the amino acid moiety.

Geometries and energies for (R)-n-Phe (n=1-4)
From Supplementary Figures 51 through 54 Especially interesting are the  values since they may be related to the product yield.
The  values for n=5 and n=1 are much lower than those for others. Here we also list the values for the hypothetical unranked conformers X and Y. They are the category I and II conformers, respectively, optimized from nonstandard routes in which preoptimization geometries were obtained by replacing hydrogens of (R)-1-Phe with methyl groups successively.