Steven K. Maa, John Grubera, Chris Davisa, Lisa Newmana, David Graya, Alica Wanga, John Gratea, Gjalt W. Huisman*a and Roger A. Sheldon*b
aCodexis, Inc, 200 Penobscot Drive, Redwood City, CA 94603, USA. E-mail: gjalt.huisman@codexis.com
bDelft University of Technology, Department of Biotechnology, Julianalaan 136, 2628, BL, Delft, Netherlands. E-mail: r.a.sheldon@tudelft.nl
First published on 23rd October 2009
The development of a green-by-design, two-step, three-enzyme process for the synthesis of a key intermediate in the manufacture of atorvastatin, the active ingredient of the cholesterol lowering drug Lipitor®, is described. The first step involves the biocatalytic reduction of ethyl-4-chloroacetoacetate using a ketoreductase (KRED) in combination with glucose and a NADP-dependent glucose dehydrogenase (GDH) for cofactor regeneration. The (S) ethyl-4-chloro-3-hydroxybutyrate product is obtained in 96% isolated yield and >99.5% e.e. In the second step, a halohydrin dehalogenase (HHDH) is employed to catalyse the replacement of the chloro substituent with cyano by reaction with HCN at neutral pH and ambient temperature. The natural enzymes were highly selective but exhibited productivities that were insufficient for large scale application. Consequently, in vitro enzyme evolution using gene shuffling technologies was employed to optimise their performance according to predefined criteria and process parameters. In the case of the HHDH reaction, this afforded a 2500-fold improvement in the volumetric productivity per biocatalyst loading. This enabled the economical and environmentally attractive production of the key hydroxynitrile intermediate. The overall process has an E factor (kg waste per kg product) of 5.8 when process water is not included, and 18 if included.
Fig. 1 Structure of atorvastatin calcium. |
Atorvastatin's high volume demand coupled with the requirement for high chemical and optical purity has led to extensive efforts towards more economic production of its chirality-setting intermediates.1 The key chiral building block in all commercialized syntheses of atorvastatin is ethyl (R)-4-cyano-3-hydroxybutyrate 1, a.k.a. “hydroxynitrile” (see Fig. 2). In the innovator's original commercial process for atorvastatin, the second stereogenic center in atorvastatin (at the 3-hydroxy group) is set by diastereomeric induction, using cryogenic borohydride reduction of a boronate derivative of the 5-hydroxy-3-keto intermediate 2 derived from 1.2
Fig. 2 Synthetic route for atorvastatin calcium. |
Previous routes that have been used for the industrial production of 1 are depicted in Fig. 3. Early routes involved kinetic resolutions using microbes, and the use of (S)-hydroxy butyrolactone produced from chiral pool raw materials, lactose or malic acid. Later routes have involved asymmetric reduction of ethyl 4-chloroacetoacetate, produced from diketene, using an asymmetric hydrogenation catalyst, microbial cells, or an enzyme. Alternative chemoenzymatic routes have been described, using nitrilases3 or lipases4 but they have not, to our knowledge, been commercialized. These processes tend to require high enzyme loadings,1 making product recovery difficult5 and adding significant costs to the process.
Fig. 3 Previous routes to the hydroxynitrile intermediate 1. |
The final step to 1 in all of the previous commercial processes involves reaction of an ethyl 3-hydroxy-4-halobutyrate (a.k.a. “halohydrin”) with a cyanide ion in alkaline solution at elevated temperatures to form the hydroxynitrile 1. Alkaline conditions are necessary to form the nucleophilic cyanide anion (the pKa of HCN is about 9). However, the substrate and product are base-sensitive compounds, resulting in extensive by-product formation. In one report, the reaction with the chlorohydrin was conducted at around 80 °C and pH 10 with a reaction yield of ∼85%.6 The product is a high-boiling oil and the by-products include many which are close in boiling point.6 As a result, a troublesome high-vacuum fractional distillation is required to recover product of acceptable quality, which results in still further yield loss and waste.6
Annual demand for 1 is estimated to be in excess of 100 mT, making it highly desirable to reduce the wastes and hazards involved in its manufacture, while reducing its cost and maintaining or, preferably, improving its quality.
It is difficult to imagine a practical process for 1 that does not involve using cyanide, and we are not aware of any that has been proposed. We reasoned that the key to providing a superior process for 1 would be to find and develop a catalyst capable of accomplishing the cyanation reaction under mild conditions at neutral pH, so that by-product formation would be minimized. Enzymes are known that catalyse the enantioselective ring-closing elimination of halohydrins to the corresponding epoxide in a reversible reaction.7 These enzymes are often referred to as halohydrin dehalogenases (HHDHs). It is also known, from the work of Nakamura and co-workers8,9 that these HHDHs accept cyanide as a non-natural nucleophile leading to the irreversible enantioselective formation of β-hydroxynitriles, under neutral, ambient conditions. More recently, this finding was further elaborated and extended to other non-natural nucleophiles by Janssen and co-workers.10–12 The availability of this enzyme led us to propose the two-step, three-enzyme process for the production of 1 shown in Fig. 4. The key step is the HHDH catalysed reaction of the halohydrin 4 with HCN at neutral pH to produce 1.
Fig. 4 Two-step, three-enzyme process for hydroxynitrile 1. |
In addition to being highly efficient catalysts that typically exhibit exquisite chemo-, regio-, and stereoselectivities, enzymes have many benefits from the viewpoint of developing green, sustainable processes.13 They operate under mild conditions (ambient temperature and pressure as well as neutral pH), with water as the preferred reaction medium. They are produced from renewable raw materials and are nontoxic and biodegradable. Enzymes often provide the opportunity to reduce the number of chemical steps in a synthetic process as there is no need for functional group protection and deprotection. Finally, enzymatic reactions are typically performed in standard multi-purpose manufacturing plants. Notwithstanding all these advantages, the use of enzymes as catalysts in the large-scale production of fine chemicals and pharmaceuticals had been hampered by perceived intrinsic limitations of natural enzymes.14,15 Such limitations include insufficient activity towards non-natural substrates, insufficient activity at high substrate loadings due to substrate and/or product inhibition, and low operational stability under such economically viable conditions. While process engineering solutions might alleviate some of these problems,16 the optimal solution is to employ in vitro directed evolution technologies to generate enzymes that operate effectively with the desired substrate and under the desired, more practical process conditions;17 instead of following the traditional path of compromising the process to fit the available catalyst, the catalyst is evolved to fit the desired process. This enables the development of a process that is ‘green by design’.
Herein, we describe a novel, economically viable and environmentally attractive process for the large-scale synthesis of 1 that includes the laboratory evolution of three enzymes to meet pre-defined process parameters. Details of the enzyme evolution aspects of the HHDH enzyme have been reported elsewhere.18
The activities of natural KRED and GDH for the reduction of 2 to 3 were low (Table 1). With 100 g L−1 of 3, a total of 9 g L−1 natural, recombinantly produced KRED (6 g L−1) and GDH (3 g L−1) were required to complete the reaction in 15 h. However, due to emulsion formation, recovery of 4 was problematic. Although the analytical yield of 4 was >99%, the recovered yield, after an hour to allow for some emulsion separation, was only 85%. To enable a practical large-scale process, the enzyme loadings needed to be drastically reduced.
Parameter | Process design | Initial process | Final process |
---|---|---|---|
Substrate loading/g L−1 | 160 | 80 | 160 |
Reaction time/h | <10 | 24 | 8 |
Biocatalyst loading/g L−1 | <1 | 9 | 0.9 |
Isolated yield/% | >90 | 85 | 95 |
Chemical purity/% | >98 (GC) | >98 | >98 |
E.e. of 4/% | >99.5 | >99.5 | >99.9 |
Phase separation of organic product phase from aqueous phase containing enzyme/min | <10 | >60 | <1 |
Space–time yield/gproduct L−1 d−1 | >384 | 80 | 480 |
Catalyst yield (gproduct/gcat) | >160 | 9 | 178 |
DNA shuffling technology20 was used to improve the activity and stability of KRED and GDH while maintaining the nearly perfect enantioselectivity exhibited by the natural KRED. DNA shuffling involves the mutation of a gene encoding the enzyme of interest to generate “libraries” of mutants. These libraries are screened in high throughput, under conditions that approximate the desired manufacturing process, and improved mutants are selected for further evolution. Their genes are recombined in vitro to create the next generation of libraries to be screened for further improved progeny. The gene libraries are transformed and expressed in a host organism, such as Escherichia coli, and screened in high throughput for enzyme variants that exhibit the desired improved properties. DNA shuffling is becoming a proven method to efficiently improve properties of enzymes, such as specific activity, stereoselectivity, optimum pH, organic solvent tolerance, and stability.21
Through several generations of such DNA shuffling, GDH activity was improved by a factor of 13 and KRED activity by a factor of 7. The enantioselectivity of the improved KRED remained >99.5%. With the improved enzymes, the reaction was complete in 8 h with an increased loading of 3 to 160 g L−1, reduced KRED loading to 0.57 g L−1, and lowered GDH loading to 0.38 g L−1 (Table 1). With a 9.5× lowering of the enzyme loading there were no emulsion problems. Phase separation required less than one minute and provided 4 in >95% recovered yield of >99.9% e.e.
The activity improvement for the ketone reduction biocatalysts over the course of evolution compared with the wild type enzyme is shown in Fig. 5.
Fig. 5 Improvements in the biocatalysts’ performances for the reduction of 3. The numbers identifying the enzymes represent the number of generations of evolution. The KREDs were exposed to thermal challenges before their assay as indicated in the figure. |
Similarly, the activity of natural HHDH for cyanation of 4 to 1 (Fig. 6) was extremely low and the enzyme showed poor stability in the presence of the substrate and product. With 20 g L−1 of 4 and 30 g L−1 of recombinantly-produced natural HHDH, the rate of reaction had virtually approached 0 by 72 h. Product recovery was difficult due to work-up challenges in the filtration and phase separation steps caused by the large amount of enzyme. In addition to the extremely low activity and poor stability of the natural HHDH for the cyanation of 4, the enzyme was also strongly inhibited by the product. Since 1 is more water-soluble than the substrate 4, attempts to overcome the inhibition by extraction of product into an organic solvent such as EtOAc, n-BuOAc, or MTBE were unsuccessful. However, after many iterative rounds of DNA shuffling, with screening in the presence of iteratively higher concentrations of product, the inhibition was largely overcome and the HHDH activity was increased >2500-fold compared to the wild-type enzyme.
Fig. 6 HHDH-catalyzed cyanation of 4. |
Fig. 7 shows the progress of cyanation reactions, using HHDH catalysts spanning multiple generations of directed evolution. With the improved HHDH, the reaction of 140 g L−14 using 1.2 g L−1 enzyme was complete in 5 h (Table 2). The product was isolated in 92% yield.
Fig. 7 Progress of cyanation reactions using HHDH catalysts. Substrate:biocatalyst = 100:1 (w/w). Rd 1 = expression mutant of wild type enzyme. Rd 22 gave 130 g L−1 product. |
The dramatic improvement in the activities of three different enzymes demonstrates that DNA shuffling technologies can enable large-scale enzymatic processes that otherwise would not have been economically viable. Our completely enzymatic process has been scaled-up to 2000 L reactors. In addition to this two-step process, running the reactions as a one-pot process, sequentially or concertedly, was demonstrated on a laboratory scale.19
Principle 1: waste prevention. The highly selective biocatalytic reactions afforded a substantial reduction in waste. In the final process, raw material is converted to product with >90% isolated yield affording product that is more than 98% chemically pure with an enantiomeric excess of > 99.9%. Furthermore, the avoidance of alkaline-induced by-products obviates the need for further yield-sacrificing fractional distillation. The highly active evolved biocatalysts are used at such low loadings that countercurrent extraction can be used to minimize solvent volumes. Moreover, the butyl acetate solvent is recycled with an efficiency of 85%. The E factor (kg waste per kg product)23 for the overall process is 5.8 if process water is excluded (2.3 for the reduction and 3.5 for the cyanation). If process water is included, the E factor for the whole process is 18 (6.6 for the reduction and 11.4 for the cyanation). The main contributors to the E Factor, as shown in Table 3, are solvent (EtOAc and BuOAc) losses which constitute 51% of the waste, sodium gluconate (25%), NaCl and Na2SO4 (combined ca. 22%). The three enzymes and the NADP cofactor accounted for < 1% of the waste. The main waste streams are aqueous and directly biodegradable.
Parameter | Process design | Initial process | Final process |
---|---|---|---|
Substrate loading/g L−1 | ≥120 | 20 | 140 |
Reaction time/h | 8 | 72 | 5 |
Biocatalyst loading/g L−1 | 1.5 | 30 | 1.2 |
Isolated yield/% | >90 | 67 | 92 |
Chemical purity/% | >98 | >98 | >98 |
E.e. of 1/% | >99.5 | >99.5 | >99.5 |
Phase separation of organic product phase from aqueous phase containing enzyme/min | <10 | >60 | <1 |
Space–time yield/gproduct L−1 d−1 | >360 | 7 | 672 |
Catalyst yield (gproduct/gcat) | 80 | 0.7 | 117 |
Waste | Quantity (kg per kg HN) | % Contribution to E (excluding water) | % Contribution to E (including water) |
---|---|---|---|
ECAA losses (8%) | 0.09 | <2 | <1 |
Triethanolamine | 0.04 | <1 | <1 |
NaCl and Na2SO4 | 1.29 | 22 | ca. 7 |
Na-gluconate | 1.43 | ca. 25 | ca. 9 |
BuOAc (85% recycle) | 0.46 | ca. 8 | ca.3 |
EtOAc (85% recycle) | 2.50 | ca. 43 | ca. 14 |
Enzymes | 0.023 | <1 | <1 |
NADP | 0.005 | 0.1 | <0.1 |
Water | 12.250 | — | 67 |
E factor | 5.8 (18) |
Principle 2: atom economy. The atom economy24 is only 45%. The use of glucose as the reductant for cofactor regeneration is cost effective but not particularly atom efficient. However, glucose is a renewable resource and the gluconate co-product is fully biodegradable.
Principle 3: less hazardous chemical syntheses. The reduction reaction uses starting materials that pose no toxicity to human health or the environment. It avoids the use of potentially hazardous hydrogen and heavy metal catalysts throughout the process thus obviating concern for their removal from waste streams and/or contamination of the product. While cyanide must be used in the second step, as in all practical routes to HN, it is used more efficiently (higher yield) and under less harsh conditions compared to previous processes.
Principle 4: design safer chemicals. This principle is not applicable as the hydroxynitrile product is the commercial starting material for atorvastatin.
Principle 5: safer solvents and auxiliaries. Safe and environmentally acceptable butyl acetate is used, together with water, as the solvent in the biocatalytic reduction reaction and extraction of the hydroxynitrile product; no auxiliaries are used. Solvent use is minimized by employing countercurrent extraction.
Principles 6 and 9: design for energy efficiency, and catalysis. In contrast with previous processes, which employ elevated temperatures for the cyanation step and high pressure hydrogenation for the reduction step, both steps in our process are very efficient biocatalytic transformations. The reactions are run at or close to ambient temperature and pressure and pH 7 and the very high energy demands of high vacuum distillation are dispensed with altogether, resulting in substantial energy savings. The turnover numbers for the different enzymes are >105 for KRED and GDH and >5 × 104 for HHDH. Because of the low enzyme concentration used, immobilization of the biocatalyst to make it recyclable is neither practical, nor economic.
Principles 7 and 10: use of renewable feedstocks, and design for degradation. The enzyme catalysts and the glucose co-substrate are derived from renewable raw materials and are completely biodegradable. The by-products of the reaction are gluconate, NADP (the cofactor that shuttles reducing equivalents from GDH to KRED) and residual glucose, enzyme, and minerals and the waste water is directly suitable for biotreatment.
Principle 8: reduce derivatization. The process avoids derivatization steps, i.e. it is step-economic25 and involves fewer unit operations than earlier processes, most notably by obviating the trouble-prone product distillation or the bisulfite-mediated separation of dehydrated by-products.26
Principles 11 and 12: real-time analysis for pollution prevention, and inherently safer chemistry. The reactions are run in pH-stat mode at neutral pH by computer-controlled addition of base. Gluconic acid generated in the first reaction is neutralized with aq. NaOH and HCl generated in the second step is neutralized with aq. NaCN, regenerating HCN (pKa∼9) in situ. The pH and the cumulative volume of added base are recorded in real time. Feeding NaCN on demand minimizes the overall concentration of HCN affording an inherently safer process.
This journal is © The Royal Society of Chemistry 2010 |