Considerations for the Scale‐up of in vitro Transaminase‐Catalyzed Asymmetric Synthesis of Chiral Amines

In vitro transaminase catalyzed asymmetric synthesis is not a new phenomenon in the biocatalytic toolbox. Our group published a review on process consideration for these reactions in 2011, in which process metrics and the unfavorable reaction equilibrium was deduced to be the main bottlenecks at the time. Now, a decade later, this has been solved by different process methods presented in this review, with the development of enzymatic cascades one of the main solutions. Today, the new bottleneck is the downstream processing, due to the complexity of transaminase reactions, in which a ketone and an amine as substrates and a new ketone and a new amine as products. How to recover the produced amine from the resulting product mixture is considered and will be the key challenge for scale‐up of such systems in the near future.

In vitro transaminase catalyzed asymmetric synthesis is not a new phenomenon in the biocatalytic toolbox. Our group published a review on process consideration for these reactions in 2011, in which process metrics and the unfavorable reaction equilibrium was deduced to be the main bottlenecks at the time. Now, a decade later, this has been solved by different process methods presented in this review, with the develop-ment of enzymatic cascades one of the main solutions. Today, the new bottleneck is the downstream processing, due to the complexity of transaminase reactions, in which a ketone and an amine as substrates and a new ketone and a new amine as products. How to recover the produced amine from the resulting product mixture is considered and will be the key challenge for scale-up of such systems in the near future.

Introduction
Chiral amines play an important role in modern chemical synthesis given they are incorporated as key building blocks in pharmaceuticals, fine chemicals, and agrochemicals, amongst others. [1] According to Yin and co-workers (2020), approximately 35 % of the top 200 small molecule drugs sold in 2018 contained at least one chiral amine subunit. [2] Nevertheless, the production of chiral amines by common chemical synthesis methods is not as straightforward as one would hope, and often requires harsh reaction conditions as well as potentially toxic intermediates and transition metal catalysts. [1,3] Thus, conversion to an alternative and greener production method is necessary, if the chemical industry wants to have a positive effect on the rapidly deteriorating environmental issues seen today.
One such method is biocatalytic synthesis, in which enzymes are used to facilitate the desired reaction under mild reaction conditions in aqueous solution. [4] Thus, to ensure an overall environmentally friendly production of enantiomerically pure amines, biocatalytic synthesis has been thoroughly researched, with more and more enzyme types joining the toolbox in recent years due to the advances in protein engineering.
This review will focus solely on the enzyme class transaminases (TAs), albeit with a brief description of the various other enzymes suitable for chiral amine synthesis in the following section. However, in contrast to the many other reviews on transaminases that have been published over the last decade, [3,[5][6][7][8][9] this review will focus on the process aspects of transaminase systems in a similar manner as our group did a decade ago; [10] How would one scale-up such a system, and what challenges could arise in the process? Here, a review of transaminase research published within the last decade will be presented from a process perspective, with emphasis on the reactor technology and scale used.

Chiral Amine Synthesis -Enzyme Possibilities
Currently, the most well-known and established enzymes for chiral amine synthesis are lipases, amine dehydrogenases (AmDHs), imine reductases (IREDs), amine oxidases (AOs) and transaminases (TAs), but also newly engineered cytochromes P450s (for example P411 CHA ) have been developed for this specific purpose. Herein, the differences between these and their suitability for industrial scale synthesis will briefly be described.
Lipases (EC 3.1.1) were the first enzyme type to be used for chiral synthesis and are therefore also the most available and thoroughly investigated class of enzymes. Commonly, in the context of chiral amines, these are used for kinetic resolution of either a racemic amine by selective acylation of one enantiomer to the corresponding amide, a method used at BASF in~multiton-scale production, or a racemic amide by hydrolysis of a one enantiomer to the corresponding amine as seen in Scheme 1a) and 1b), respectively. [11,12] However, like lipase-catalyzed reactions in alcohol-acid/ester systems, the primary reaction in the presence of even small amounts of water is hydrolysis, which makes kinetic resolution by selective acylation a challenge. This is especially valid, given lipases are interfacially activated and thus require a small amount of water to facilitate the reaction which inevitably leads to a compromise of the final yield. Moreover, the substrate scope of most lipases is limited to small and relatively simple molecules, and thus unsuitable for pharmaceutical syntheses. Finally, lipases must be combined with an additional reaction step to achieve acceptable yields by dynamic kinetic resolution. [13] An alternative synthesis uses amine dehydrogenases (AmDHs, EC 1.4.1), which catalyze the reductive amination of ketones by means of ammonia as shown in Scheme 2.
However, as amine dehydrogenases require the cofactor NADH for electron transport, a cofactor regeneration system is necessary to ensure an adequate total turnover number of NADH to compensate for its high price. Commonly, glucose dehydrogenase (GDH) with glucose as co-substrate is used for simple and cheap NADH-regeneration. [13] Nevertheless, due to the production of gluconic acid in this reaction, meticulous pH control is required, which can be an issue at an industrial setting. Moreover, only primary amines are possible with amine dehydrogenases, limiting their overall product scope. [14] If secondary-or tertiary amines are desired, then either amine oxidases or imine reductases could be used instead.
Here, amine oxidases (AO, EC 1. 4.3.4) catalyze the oxidation of amines to imines by the concomitant conversion of molecular oxygen to hydrogen peroxide, for example by kinetic resolution as shown in Scheme 3.
The most well-known amine oxidase is monoamine oxidase (MAO), a type II amine oxidase which requires flavin as cofactor. [13] Due to the great effort in enzyme engineering by the Turner group, the substrate scope of monoamine oxidases has been expanded immensely compared to the corresponding wild type. These can now be used for deracemization of primary, secondary, and tertiary amines. [13] However, as the stability of these is normally poor, their application on an industrial scale is still relatively limited.
Alternatively, imine reductases (IREDs) have gained a lot of attention recently, as these catalyze the asymmetric reduction of imines, making the production of secondary and tertiary amines possible as shown in Scheme 4.
Unfortunately, imine reductases require the cofactor NADPH to facilitate the reaction, and thus a similar cofactor regeneration system as with amine dehydrogenases is necessary to increase the cofactor total turnover number. As NADPH is more expensive than NADH, the total turnover number is slightly more critical in the case of imine reductases. Moreover, given these are still in the early stages of development and currently limited to cyclic imines, they still have a way to go before they will be industrially relevant. Nevertheless, the potential for asymmetric reduction could be an interesting new field of biocatalysis in industry.
Finally, the most used enzyme for chiral amine synthesis is transaminase (TAs, EC 2.6.1), also known as aminotransferases (ATA), which this review will consider in detail.
In the meantime, which of these five enzyme classes to choose for production in an industrial setting depends very much on the biocatalytic route required, the type of product and the easiest route to achieve the desired metrics. However, there is no doubt that transaminases currently have a very important role to play, despite the new and more advanced enzyme classes developed in recent years. This is especially the case if the end-product is a more complex molecule such as a pharmaceutical. With the very versatile nature of the current transaminases and the large selection of transaminases readily available in screening kits, scale-up should be possible.
Julie Østerby Madsen studied chemical and biochemical engineering at the Technical University of Denmark and is currently in the last year of her Ph.D. studies. Her research focuses on continuous enzymatic cascade processes for pharmaceutical processes, in which her main enzymatic system includes a transaminase step for chiral amine synthesis. Her work is done under the supervision of Prof. John M. Woodley

A Decade of Transaminase Publications
Transaminases are used for chiral amine synthesis in one of two ways; either by kinetic resolution in a similar manner to lipases, albeit by converting one of the enantiomers to its corresponding carbonyl compound, or by asymmetric synthesis from a carbonyl compound. For this purpose, the cofactor PLP is required, along with an amine acceptor or donor as shown in Scheme 5a) and 5b), respectively. In this review, the focus will lie solely on asymmetric synthesis given its high applicability in industrial production.
Transaminases are often divided into two subgroups: αtransaminases (α-TAs) and ω-transaminases (ω-TAs), the difference being their substrate scope. Here, α-transaminases only react with carbonyl compounds, which have a carboxylic acid group in the α position to the carbonyl, while ω-transaminases could, in principle, accept any carbonyl compound, making them more relevant for industry. [13] With the immense effort in protein engineering of transaminases done by Codexis and Merck for the production of Sitagliptin in 2010, industrially relevant transaminases are now readily available, with both (S)and (R)-enantiomers possible. [15] Nevertheless, these still have difficulties which must be overcome, the main one being the unfavorable equilibrium. Moreover, even though transaminases are not limited to small-and simple molecules (as lipases), these can still only be used for primary amine production. Thus, if secondary or tertiary amines are desired, additional reaction steps would be required. Nevertheless, within primary amine synthesis they are currently the most versatile option available. [11] The last decade has seen an increase in readily available commercial transaminases, often in screening kits with multiple variants (e. g. from Codexis Inc. (CA, USA) and c-LEcta GmbH (Leipzig, Germany)). This has also ensured a huge increase in published biocatalytic transaminase research over the last decade, both as single step reactions and incorporated in cascades. This review will only take into consideration publications using enzymes in vitro.
In vivo systems are often limited by substrate uptake. Coupled with the development of comprehensive toolboxes for retrosynthesis of biocatalytic routes as presented by Souza and co-workers (2017) [16] amongst others and incorporated in practice by for example Huffman and co-workers (2019) in their impressive retrosynthetic approach for cascade development of the production of the investigational HIV treatment Islatravir [17] (albeit without a transaminase step), highlight the endless possibilities for in vitro biocatalytic conversions as an alternative to in vivo systems.
For this review, relevant publications published within the last decade were identified with the search engine in Web of Science TM . Here, articles tagged with both the topic transamina* as well as biocat* were retrieved and read in detail to ensure they conformed with the defined requirement. The process parameters presented in each are reported in Tables 1-4, representing a decade's worth of publications on in vitro asymmetric synthesis of chiral amines. An overview of abbreviations used in the tables can be found in Section 7.
This includes publications which utilized transaminases either as cell-free extract or purified. Where not explicitly written in the publication, especially in which transaminases from home-grown cells were used, the enzyme recovery-and/ or isolation method was considered to ensure a cell disruption step was included, either by sonication, enzymatic lysis or similar. It is further noted that only articles published where a (fedÀ )batch configuration or similar was used are included. Given only a limited number of publications utilizing a continuous flow configuration in comparison to (fedÀ )batch was published, conclusions on proper measures to establish such a system will not be presented in this review.
The majority of the presented publications report singlestep reactions with transaminase either as the sole enzyme (Table 1), with enzymatic amine donor regeneration or -removal (Table 2) or as a single enzymatic step in a chemo-enzymatic reaction scheme (Table 3). Here, reaction schemes with an in vitro transaminase step combined with either prior-or subsequent whole-cell conversions were also included in the chemo-enzymatic table. The remaining articles reported multistep reactions (cascades), where transaminase was combined with other enzymes (Table 4). It is noted that different perceptions of what defines a cascade are found in literature. Here, a cascade will be defined as a multi-step enzymatic reaction, in which at least two enzymes are used sequentially for the primary conversion, as in a linear cascade and similar, and thus single step reactions with enzymatic cofactor regeneration are included in single step reactions. [192] In our previous work, one of the conclusions drawn was the need to combine biocatalytic-and process improvements to ensure correct optimization parameters and development of scalable systems. [10] For this purpose, a distinction between scaled and scalable must be drawn, given a non-scaled system can still be seen as scalable without necessarily being scaled. We define a scalable system as one, which conforms with the 2018 [24] Rac-sec- Bun  2019 [70] 0  [109] set metrics required for an economical feasible industrial scale production. So, although the scales reported (Working Volume, WV) in the last decade are commonly toward the smaller size, both for single-and multi-step reactions in Table XÀ Q, this does not mean that scalability has not been considered within the research community. In fact, it is clear that the focus has switched from finding new and existing transaminases to developing and optimizing conversions that conform with the metrics presented in our previous publications, as well as summarized more recently by Meissner and Woodley (2022). [10,193] Additionally, a distinction between a reaction and a process needs to be drawn, as these two are used interchangeably in the literature, but are definitely not the same if seen from a scale-up perspective. Considerations into downstream processing need to be taken into account to develop a process.
While research at laboratory scale is essential for the continued exploration and optimization of enzymes and reaction mechanisms, the reactors used in the laboratory and the corresponding mixing techniques/profiles simply do not represent what is seen in larger scale reactors, which could result in limited usability of enzymes optimized in the laboratory if seen from an industrial perspective. If these enzymes cannot perform at the larger scale, the use of laboratory results will be somewhat limited. An example could be if the developed enzymes are sensitive to environmental changes, and thus could get stressed in the non-natural zones present in large-scale batch reactors. However, there is no doubt that all laboratory results are fundamental to new knowledge and subsequently can assist in the transfer to industry.
Thus, as this review focuses on reactor technology for transaminases, studies at all scales are listed in the tables, but only published research with a minimum of 1 L production will be highlighted in the following. It is noted that when not explicitly written in the publication, the working volume has been estimated based on the combined quantities added to the reactor as listed in the materials and methods section.

Challenges with (FedÀ )batch Production
As mentioned previously, (fedÀ )batch production is the most used method for transaminase catalyzed reactions -and most all enzymatic reactions -to date. [194] There are also many reasons for this, other than it is the reaction technology that has been used since the first enzymatic conversions were published.
One of the main advantages of the stirred batch reactor is the flexibility it presents, ensuring a range of different molecules can be produced in the same set-up. This is especially important for industries, where small quantities of a lot of different products are produced, such as with pharmaceuticals. Essentially, a stirred batch reactor is a large vessel with a mechanical stirrer as depicted in Figure 1. Figure 1 includes a stirrer with two Rushton impellers as well as baffles on the inner walls to ensure proper mixing and a heating/  [110] α-Amino 2013 [111] n-Bun  2020 [145] 2 mL Eppendorf Tube 2014 [153] AHAS ω-TA 2 steps 1 pot Seq.  [182] ω-TA  [190] UPO ω-TA 2 steps 1 pot, Seq.  [191] ChemCatChem Review doi.org/10.1002/cctc.202300560

D-Ala
cooling jacket with coolant inlet and outlet for temperature control. Generally, a batch reaction is run by addition of substrate and enzyme through the inlet in the top. During the reaction, the cooling/heating jacket is used to ensure the reaction is kept at a specified temperature if above room temperaturecommonly around 30°C for transaminases as seen in the tables. Acid/base is added through a separate inlet (not depicted) if pH control is required. Similar, a sparger can be added at the bottom of the reactor (not depicted) if gas addition is required, such as with nitrogen sparging as described in a later section. After the end of the reaction, the reactor is emptied through the outlet at the bottom and the product stream consisting of product and enzymes is sent to downstream processing for product isolation. The reactor is then cleaned for subsequent conversions.
From the description, it is clear that (fedÀ )batch processes are simple to run, which is a big bonus both at laboratory scale and in an industrial setting. However, (fedÀ )batch reactors also have their disadvantages, especially when the vessels become larger, given that the mixing time increases significantly. If pH control is required due to production of acid/bases or similar, this can result in pH gradients in the vessel, which can result in loss of enzyme activity due to enzyme unfolding in non-optimal pH zones. In addition, the filling-and emptying of the vessels as well as cleaning in between batches requires a significant cost time, which is the major disadvantage for this type of reactor. [194] Nevertheless, as the (fedÀ )batch reactor is still the most commonly used reactor type for laboratory biocatalytic reactions, how to move transaminase catalyzed reactions from the laboratory to an industrial scale (fedÀ )batch reactor will be discussed in the following, with focus on the main difficulties that this could result and practical solutions to circumvent them.

Substrate Solubility and Inhibition
The first step in any enzymatic reaction is the addition of substrates, co-substrates and co-factors to the reaction media as required for the desired reaction. For transaminases, this translates to a ketone as substrate, an amine donor as cosubstrate, PLP as co-factor and the enzyme itself as mentioned previously. Generally, the reaction media for these types of conversions is either a buffer solution or pure water, with a buffer solution the most common laboratory practice. It is noted that given buffers can be expensive, systems based on pure water would be more feasible at scale. However extensive pH regulation would be required, if an alkaline amine donor is used, such as isopropylamine, given this would result in a drop of pH when converted to its corresponding ketone. It is noted, if a large isopropylamine excess is used, this drop in pH should not be observed given the unused isopropylamine still in the reaction media after end reaction. The simplest technique for substrate addition at laboratory scale as well as industrial scale is simply to add everything to the reaction media while stirring with the enzymes as the last component to start the reaction. However, ketones commonly have very low solubility in water, making an otherwise simple substrate addition technique more complex, given low substrate concentrations in large scale production directly translates to an expensive downstream process. Nevertheless, if a higher concentration of ketone is added to a reaction media consisting mainly of water, this would dissolve slowly and result in a hetrerogeneous mixture increasing the power required for mixing. This phenomenon was observed by Kohrt and coworkers (2022), where a 20 L water-jacketed Chemglass reactor with a commercial transaminase from Codexis was used at scale-up to produce Boc-protected-1-oxa-8-azaspiro [4.5]decan-3-amine, a pharmaceutically relevant intermediate. [60] Here, they noted increasing ketone aggregates were seen at increasing scale, resulting in a heterogenous mixture with inefficient stirring and ultimately loss in product formation.
One method to circumvent this issue is running the reaction at higher temperatures, as solubility is known to be enhanced by increasing temperature. [15] However, as transaminases and enzymes in general denature at elevated temperatures, this method can only help to a limited extent.
Instead, what is done in practice is the addition of a cosolvent, as noted in the tables. Different co-solvents have been used in the last decade, including dimethyl sulfoxide (DMSO), acetonitrile (MeCN), methanol (MeOH), dimethylformamide (DMF) and ethanol (EtOH). However, by far the most common is DMSO, as also used by Savile and co-workers (2010) for the production of Sitagliptin, a pharmaceutical used against diabetes. [15] For Kohrt and co-workers, the addition of 8 % DMSO to the larger scale system resulted in a homogenous mixture suitable for conversion at a concentration of approximately 50 g/L ketone. The ketone was dissolved in DMSO at a concentration of around 750 g/L and then added dropwise to the reaction media to avoid spontaneous precipitation and subsequent aggregate formation. Moreover, to ensure complete homogeneity, the mixture was stirred overnight under a nitrogen sweep, prior to enzyme and PLP addition to initiate the reaction.
It is noted that the low solubility depends on the ketone used as substrate, given Girardin and co-workers (2013) did not use any co-solvent in their pilot scale process, for the production of the dual orexin receptor antagonist Suvorexant. [122] For this purpose, a commercial transaminase from Codexis was used in a 100 L Buchi Jacketed reactor with overhead stirring. Here, everything except the ketone -in this case a keto-diester -was added to the reactor and mixed before ketone addition as powder, the final concentration approximately 75 g/L. It is not known if this keto-diester is just much more soluble than the ketone used by Kohrt and coworkers given the higher substrate concentration, or if precipitation was seen but not noted in the publication.
If higher substrate concentrations are required, the percentage of DMSO is commonly increased accordingly, up until the DMSO tolerance for the specific transaminase. However, the practical aspect of how to add the substrate can still be in doubt.
Feng and co-workers (2017) used a commercial transaminase from Syncozymes in their chemoenzymatic process to produce N-Boc-azepan-3-one, a pharmaceutically important intermediate. [141] The DMSO tolerance for the enzyme was tested, and it was noted that the highest conversion was achieved at 50 % DMSO. The reaction itself was run in a vessel of unknown size with a working volume of approximately 3.5 L, in which all except the ketone was added and mixed, including DMSO. The ketone was then added as a powder directly to the mixture in one go and dissolved by vigorous stirring, with a resulting substrate concentration of 100 g/L. Despite the good results published in this study, this approach to substrate addition might not be suitable for larger scales, especially if higher substrate concentrations are desired.
A similar tactic as Kohrt and co-workers was used by Savile and co-workers, in which the ketone was dissolved in DMSO prior to addition. Here, around 40 % DMSO was used, with the ketone dissolved in one-third of this and the rest added directly to the reactor. The ketone-DMSO mixture was then fed slowly to the vessel containing the remaining reaction mixture to avoid precipitation of the ketone. The concentration in the ketone-DMSO feed mixture was 900 g/L, with the final substrate concentration around 125 g/L. This method of substrate addition is much more likely to work at an industrial setting, as an overall homogenous system is achieved and no significant changes in viscosity are likely to hinder the mixing efficiency.
However, it is noted that Savile and co-workers defined 'slowly' as addition over 2-3 hours, in which the ketone-DMSO mixture was fed to their vessel. Contrary to Kohrt and coworkers, the transaminase was added to the vessel prior to the addition of the substrate, and thus the first 2-3 hours of substrate addition will already result in conversion.
The same is true in a more recent study by Feng and coworkers (2019), where N-Boc-azepan-3-one was produced by a commercial transaminase from Enzymetree in a 500 L enamel reactor. [62] Here, around 40 % DMSO was used, of which about half was used to dissolve the ketone and added slowly to the reactor over a period of 3 hours. This technique combines a pure batch reactor, in which all the substrate is added at the start, and a fed-batch reactor, in which substrate is added continuously throughout the reaction. The fed-batch reactor is often used if the reaction is hindered by substrate inhibition, given the constant addition of a substrate results in constant dilute substrate concentrations. Commonly, the substrate is fed at a rate, which ensures a constant optimal substrate concentration.
The difference between a pure batch-and a fed-batch reactions is illustrated in Figures 2 and 3, respectively, along with the resulting concentration profiles, assuming Michaelis-Menten kinetics as common for many enzymatic systems. [194] The concentration profile for the fed-batch was developed based on the assumption that the substrate feeding results in small working volume changes for the reaction media, something that can be achieved if a high feed concentration is used  (although may be difficult in some cases). The combined batchand fed-batch approach as used by Savile and co-workers as well as Korht and co-workers would be a combination of the two.
Savile and co-workers may have benefited from running the reaction as fed-batch the first part of the reaction, as substrate inhibition is often observed in these types of reactions, as noted by Feng and co-workers. Here, they observed decreasing conversion at increasing substrate concentration, indicating substrate inhibition. Nevertheless, they still chose to run the system as pure batch, instead of converting to fed-batch, potentially increasing their overall yield and conversion.
Another example of a reaction run as fed-batch was reported by Zhou and co-workers (2020) in the production of Lphosphinothricin, an active ingredient in common commercial herbicides. [115] Here substrate inhibition was found at concentrations above 35 mM substrate, with decreasing initial reaction rate after this point. For this reason, they chose to run the reaction in a 100 L stainless-steel bioreactor operated as fedbatch, with a start-and end working volume of 40 L and 90 L, respectively. Thus, 50 L substrate solution with a substrate concentration of 200 g/L was added over a period of 3 hours, with a total reaction time of 7 hours. The feed rate was monitored manually, which could be why a spike of substrate concentration was noted in their process, and thus not kept constant as desired for fed-batch systems.
The required substrate concentration depends directly on the value of the resulting product, given lower-value products require a higher final concentration, and by extension a higher start substrate concentration, compared to high-value products as presented by Meissner and Woodley. [193] Thus, substrate inhibition could be the deciding factor in the end, given the limit for co-solvent addition. If a low-value product is produced, the high start-substrate concentration required would often lead to substrate inhibition. In this case a fed-batch system could be the only solution to succeed with such a process.
Moreover, the use of co-solvents is not always possible, as observed by Gu and co-workers (2020) in their production of a chiral sacubitril precursor, a key component in the heart failure drug Entresto. [59,195] In this case, the co-solvents unfortunately diminished the enzyme activity and the process was therefore kept as a heterogenous mixture through the reaction instead.

Reaction Equilibrium
The next step after addition of all reactants to the reactor is the reaction itself, and how to ensure complete conversion to the desired product in an economically feasible production. For this purpose, Tufvesson and co-workers (2010, 2011) and more recently Meissner and Woodley (2022) presented a set of success factors for economic feasibility of a biocatalytic process, including product concentration, specific yield, reaction yield and %ee. [10,193,196] For transaminases specifically, complete conversion and/or high reaction yield to the amine by asymmetric synthesis is not guaranteed as mentioned previously, given the thermodynamically unfavorable reaction equilibrium.
In general, to shift the reaction equilibrium towards the product side, Le Chatelier's principle is used. If the concentration of one of the reactants/products in the reaction equilibrium is changed, the reaction equilibrium is shifted accordingly to counteract this. Ultimately, if the concentration is increased of one of the substrates in the transaminase catalyzed reaction Scheme 5b) or decreased for the products, the reaction will thereby shift toward production as desired. Thus, to achieve a higher yield, either the concentration of the amine donor can be increased, or the produced amine or co-product can be removed during the reaction.

Excess isopropylamine and acetone removal
Addition of excess amine donor was a common practice previously, as shown by Savile and co-workers where a 10-fold excess was used, in this case of isopropylamine. [15] However, as Tufvesson and co-workers emphasize, the addition of amine donor excess is not a feasible strategy much beyond this. Therefore, the equilibrium constant needs to be relatively close to unity. [10] This is mainly due to the high substrate concentrations required, and by extension unrealistic space this occupies in the reactor.
For example, if a DMSO concentration of 50vol % is considered, half of the tank volume would be dedicated to the DMSO/substrate mixture. That would leave half of the tank volume left to the amine donor. However, if pure isopropylamine is used, at least half of the remaining tank volume would be dedicated to pH neutralization if concentrated acid is used due to the alkaline nature of isopropylamine. This is based on the authors own personal experiences with pH neutralization of pure isopropylamine. This would leave one-fourth of the tank volume as a maximum for the amine donor, as illustrated in Figure 4. If a less concentrated acid is used instead, this would require more acid and by extension less tank volume remaining for isopropylamine. Contrarily, if a higher pH than neutral is used, less acid would be needed and thus a higher isopropylamine volume would be possible.
For the plot on the right-hand side of Figure 4, the slopes of the amine excess lines were calculated based on the isopropylamine concentration required for the specified substrate concentration, here for Sitagliptin, converted to tank volume required compared to the total volume of the tank in percentage. It is noted that the calculations were based on several assumptions and simplified to demonstrate the tank volume limit for amine donor excess. This would differ from substrate to substrate depending on the molar weight of the substrate as well as the volume the substrate itself take up in the tank, which would increase with increasing substrate concentration.
However, theoretically, if a 50-fold excess of isopropylamine is used, it will only be possible to increase the substrate concentration to just below 25 g/L, before the tank volume limit is reached. Contrarily, if a 5-fold excess is used instead, the substrate concentration can be as high as 235 g/L, before the tank volume limit is reached.
Overall, at increasing excess, the volume required of isopropylamine as amine donor for Sitagliptin production becomes greater than the set 25 % tank volume at increasing substrate concentration. It is therefore physically impossible based on this to have an amine donor excess higher than 10fold for substrate concentrations higher than around 125 g/L with this system.
Other amine donors, such as alanine for example, are not as alkaline as isopropylamine, and therefore a higher amine donor volume could be envisioned with lower requirements for pH regulation. However, this would also have a limit, since at some point the solubility of the amine donor would be reached, resulting in precipitation of the amine donor itself.
Moreover, with increasing excess of the donor follows increasing amounts of unreacted donor in the product mixture, which could be a problem for the following downstream processing. These difficulties therefore explain why this strategy has not been widely implemented at larger scale in the last decade.
Instead, the removal of co-product is by far the most common method for equilibrium displacement. For this purpose, the correct choice of amine donor for a specified transaminase is essential for the ease of co-product removal. However, this choice is not as straightforward as one might think and is often done by trial and error in the laboratory, if not based on past precedent.
To overcome this, Meier and co-workers (2015) proposed a practical fast-approach method to quickly screen and predict thermodynamic unfavourability of amine donors with specific substrates based solely on a few ketone-amine pairs. [197] While such a tool would be extremely helpful for in silico screening and could potentially cut down significantly on experimental costs overall, it would require a larger database than that presented to be widely applicable. [197] Moreover, even with an accurate prediction prior to process development, additional strategies would most likely still be required to further shift the equilibrium to yield high enough substrate conversion to ensure economic feasibility.
As seen from the tables, the most common amine donor used in literature by far is isopropylamine as seen in Scheme 6. This is a cheap compound with acetone as its corresponding ketone, which is very volatile and can therefore easily be removed from the reaction media by evaporation or similar as noted in the scheme, both at laboratory scale and in an industrial setting. In the laboratory, this is simply done by using a reactor without lid allowing evaporation of the produced acetone to the surroundings. However, this cannot be done at a larger scale 62 due to safety issues from the resulting fumes. Instead, three acetone removal methods are recommended as illustrated in Figure 5.
The first method (left, Figure 5) is carrier-gas flushing, here nitrogen, in which a carrier gas is sparged through the reaction media from the bottom of the reactor. The resulting gas  15 . Half of the tank volume is used for DMSO/substrate (pink), and a maximum of one-fourth of the tank volume is possible for isopropylamine (blue) due to the need for pH neutralization with concentrated acid (green).
bubbles will travel to the top of the reactor along with any volatile compounds, such as acetone. This method is simple, although it has its limitations at a larger scale, as the sparging can result in foam formation if not controlled properly. Given enzymes are known to stick to surfaces such as bubbles, these would be trapped in the foam resulting in a loss of enzyme from the solution, unless immobilized.
The second method is headspace sweeping (middle, Figure 5), where the carrier-gas, here nitrogen, is swept over the surface of the reaction media instead of bubbling through, removing the acetone that automatically evaporate into the headspace during reaction. This method depends solely on acetone's ability to evaporate, which in turn depend on the temperature of the reaction media. The higher the temperature, the more acetone will evaporate and be moved from the reactor by the sweep.
The third method (right, Figure 5) is implementing a vacuum in the reactor, which lowers the temperature required for acetone to evaporate, ensuring higher evaporation than if atmospheric pressure is used.
Despite the three methods illustrated separately in the figure, a combination of the three can be very advantageous, especially if headspace sweeping and vacuum is combined. This ensures that more acetone is evaporated, which is then swept away by the carrier-gas, as done by Savile and co-workers with nitrogen as carrier-gas. Likewise, Kohrt and co-workers used both vacuum and nitrogen sweep, albeit sequentially with vacuum the first 12 hours of the reaction and a nitrogen headspace sweep the last 16 hours. Why these were divided up is unclear.
However, one of the big disadvantages to the easy removal of acetone, which is rarely mentioned anywhere, is that the evaporation is not limited to acetone, but all volatile com-pounds in the reaction mixture. [198] Unfortunately, this includes the amine donor isopropylamine itself, which has a much lower boiling point (32°C) than acetone (56°C) and will therefore be much more susceptible to evaporation and therefore be removed along with acetone. Thus, to retain enough isopropylamine in the reaction media to complete the reaction, this must be added in higher amounts than the reaction requires to circumvent this issue. Otherwise, if acetone is removed isopropylamine/acetone mixture evaporated from the reactor, the purified isopropylamine could be recycled to the reactor after condensation. However, this makes the simple approach to removal of the co-product somewhat more complex.

Amine donor regeneration or reduction
A different approach is by means of another amine donor, which can be either recycled or regenerated enzymatically. A range of different options have been reported in the scientific literature, with a few scaled up to pilot scale. For further information of the remaining possibilities, please refer to the indepth review done by Simon and co-workers (2014). [199] As seen from the tables, the most common method utilized in literature is the lactase dehydrogenase/glucose dehydrogenase (LDH/GDH) system, in which alanine is used as amine donor, either as L-, D-or a racemic configuration as seen in Scheme 7.
Here, the co-product pyruvate is reduced by LDH to lactate, thus shifting the overall equilibrium. However, as the reduction requires the co-factor NADH, glucose is used as electron donor for co-factor regeneration with GDH. This enzymatic co-factor reduction system is commonly used at laboratory scale, however Girardin and co-workers implemented it in their transaminase step in their nine-step chemoenzymatic cascade. [122] Albeit the most common enzymatic system, the LDH/GDH co-factor regeneration system is not the only system used. Another example is the glutamate dehydrogenase/alcohol dehydrogenase (GLDH/ADH) system as reported by Zhou and co-workers and seen in Scheme 8. [115] Unlike the LDH/GDH co-factor regeneration system, the GLDH/ADH system regenerates the amine donor, here Lglutamate, using ammonia, NADH and isopropanol as amine supplement, co-factor and reducing agent, respectively. By utilizing an amine donor recycling system instead of reduction, only stoichiometric amounts of amine donor would be required in theory, ensuring a cheaper overall process. It is noted that additional regeneration/recycling systems are used in laboratory scale, such as alanine dehydrogenase/formate dehydrogenase (AlaDH/FDH) and pyruvate decarboxylase as seen from the tables. However, as these have not been reported at a larger scale, these will not be described in detail here. The application of an enzymatic reduction-or regeneration system at larger scale depends almost exclusively on the cost of the extra enzymes required. If these have high activity and a low cost, this technique could be very relevant for industry, if the alternative amine donor is affordable. If not, these systems will be unlikely to compete with the cheap isopropylamine, unless the specific transaminase cannot use this as an amine donor.
Similarly, the application at large scale of the new range of amine donors, known as smart amine donors, is also not established yet, probably due to their high cost. These include diamines such as o-xylene-diamine, [90][91][92] but-2-ene-1,4diamine, [23] cadaverine, [27,66,88,91] putrescine and spermidine [88] as seen in Table 1, which cyclize spontaneously after amine donation, thus shifting the overall equilibrium towards synthesis as shown in Scheme 9 with cadaverine as an example.
However, if the cost of these can be brought down to a manageable level, and if the system can be scaled up without issues such as precipitation etc., they could prove an interesting alternative to the already established methods. It is noted that the amine produced by Girardin and co-workers spontaneously cyclized in a similar manner as the smart amine donors, further driving the equilibrium along with the LDH/GDH regeneration system.
Another approach is in-situ product removal, which was implemented in the study by Matassa and co-workers (2020), [69] where a three-liquid-phase spinning reactor was used in the production of (S)-1-methyl-3-phenylpropylamine. Here, a 1.5 L double-jacketed glass reactor with nitrogen head space sweep and a fixed inner tubular cylinder was used. The organic solvent top phase consisted of n-heptane and the poorly water-soluble substrate, the aqueous (reaction) bottom phase consisted of the transaminase and the amine donor with an acidic extraction phase in the cylinder. More details on the system itself can be found in the publication. The overall principle of separation was diffusion of the substrate through the reaction phase, with further diffusion to-and trapping in the extraction phase of the produced amine. [69] The amine donor used was the commercial Jeffamine® ED-600, and aliphatic polyether diamine chosen based on its high-molecular weight and resulted in a 2.6-fold improvement from conventional batch conversion. [69] However, difficulties with separation of the three phases after the end of conversion and poor extraction selectivity between substrates and products could limit the proposed set-up at larger scale unless additional optimization is done. This report highlights the difficulties with in situ product removal; the difference between the substrate and product in a transaminase catalyzed reaction is just a single functional group. Thus, the in situ product removal system must be extremely selective towards the amine for such a system to be in any way preferable to a simple batch system. If not, the substrate will also be extracted along with the amine, which will then have to be separated and recycled to the reactor for full conversion. Moreover, the low product concentrations versus substrate further complicate the removal during production. Combined, this may be the reason why in situ product removal has not been used in the last decade, and why this might not be applicable for single-step transaminase catalyzed reactions.
However, it could still be relevant for multi-step cascades, where the difference between the initial substrate and the final product may be more pronounced. This will be discussed in more detail in a later section.

Enzyme efficiency
Besides the challenges and strategies for biocatalytic transamination presented in the previous sections for batch operation at pilot scale, additional considerations must be taken into account prior to further scale-up and implement the process at an industrial scale.
One of the main cost drivers when dealing with biocatalytic conversion is the price of the biocatalyst, which can often contribute to a significant percentage of the overall operating costs of such a process. It is therefore imperative that the cost of the biocatalyst is as low as possible, which corresponds to reducing the enzyme purifications cost as well as maximizing the specific yield. [200] Commonly, enzymes used at an industrial scale will simply be released from the cells by cell disruption and subsequent cell-debris removal, and thus used as cell-free extract,. Based on the presented pilot scale publications, this requirement was fulfilled for all publications except Savile-, [15] Matassa- [69] and Yasuda and co-workers, [140] where purified enzymes were used instead.
To further bring down the cost contribution of the biocatalyst, recycling of the biocatalyst can improve the biocatalytic efficiency (specific yield). However, this can be challenging if soluble enzymes are used, given these can be difficult to separate due to emulsion formation and are sometimes unstable if used for a prolonged time.
Here, immobilization of the biocatalyst is a possibility as presented by Feng and co-workers (2019), where the transaminase used was immobilized on epoxy-amine resin beads by covalent binding. [62] This commonly stabilizes the enzyme and enables multiple recycles.
Thus, immobilization of enzymes could potentially be a key step within biocatalytic processes overall. [200] However, as there is no one-size-fits-all technique developed for enzyme immobilization, significant research is required for each new system, which both require a lot of time and money that could further escalate the price of the biocatalyst. Moreover, additional investigation into the effect of stirring on the immobilized enzymes could be required, since rigid immobilization supports could potentially break when struck by the impellers. Thus, softer and more flexible supports could be required for agitated batch systems, whereas rigid supports are required in packed bed reactors. Therefore, a use-and-discard technique with soluble enzymes is a simpler and cheaper approach at a larger scale to avoid issues with emulsification at separation. [201] 6. Future perspectives 6

.1. Enzymatic Cascades
A newer field within biocatalysis developed in the last decade is the development of whole reaction schemes consisting solely of enzymatic steps, also known as cascades. As highlighted in a previous section, the name cascade has been used interchangeably in literature to describe enzymatic systems with multiple enzymes, such as single-step reactions with enzymatic co-factor regeneration, and multi-step reactions, where enzymes are coupled in a linear (or similar) reaction scheme. In this review, only the latter will be defined as cascade systems, also known as linear cascades, with the former described in the previous section, also known as parallel cascades. [199] The advantages of linear cascades, in which a substrate is converted to an intermediate with one enzyme and then subsequently converted either to another intermediate or a product by another enzyme depending on the number of steps in the cascade are numerous, some of which will be highlighted here.
In general, there are three ways to run a cascade; (1) as a one-pot system, in which all enzymes are added in the beginning of the reaction along with the substrate; (2) with sequential enzyme addition, in which the first enzyme is added in the beginning of the reaction, the second later and so forth; (3) with intermediate removal of previous enzyme, before the following is added. These three possibilities are illustrated in Figure 6,, 7 and 8, respectively, with a two-step reaction from an alcohol (blue) to an ketone (green) and further to and amine (red) as example.
Given the specificity of enzymes in general, intermediate isolation is often not required between steps in cascades, limiting the handling and storage of unstable-or toxic intermediates as these can be quickly converted. [199] Furthermore, due to the similar conditions required for enzymes in general, with close temperature-and pH ranges, entire reaction schemes could potentially by run as one-pot systems, either with all the enzymes added at start as in Figure 6 or sequentially as in Figure 7. [4] If biocatalytic transamination is considered in this regard, the one-pot system is noted as the main method used in batch systems in literature based on Table 4. Over the last decade, several cascades which include a transaminase step, have been published. Of these, a few used three biocatalytic steps in the reaction scheme, the rest with two.
If the amine donors used are considered, over one third of the articles published have used alanine in some form, while another third used isopropylamine and the remaining a mixture of other donors.
Furthermore, it is noted that biocatalytic transamination is most often used as the last step in any given cascade.
Nevertheless, even though an increase in cascade publications with a transaminase step can be seen over the last decade, only very few have been scaled up from laboratory scale, one of which was published by Chen and co-workers (2011). [144] Here, a two-step cascade was used for the production Figure 6. Enzymatic cascade run as a one-pot system, in which all enzymes (orange and purple) are added in the beginning of the reaction (t 1 + 2 ) along with the substrate (blue). The substrate is converted to the intermediate (green) before immediately being converted to the product (red), which is removed by end reaction (t n ).

Figure 7.
Enzymatic cascade with sequential enzyme addition, in which the first enzyme (orange) is added at the start of the reaction (t 1 ) along with the substrate (blue). The intermediate is produced (green) and converted when the next enzyme (purple) is added at a later time (t 2 ). The product (red) is removed by end reaction (t n ). of (S)-2-amino-3-(6-o-tolylpyridin-3-yl)propanoic acid, an intermediate in an antidiabetic drug candidate, from its racemic counterpart. The first step in the cascade used an R-selective amino acid oxidase for the oxidation of the racemic amine to its corresponding keto acid by means of oxygen, with the coproduct hydrogen peroxide further decomposed by catalase. The resulting keto acid was converted to the desired (S)configuration in the second step of the cascade by means of a (S)-transaminase with aspartate as the amine donor. The entire process was operated as a one-pot system in a 100 L vessel of approximately 80 L working volume, with all three enzymes added at the start of the reaction. Here, oxygen was sparged through the reactor for the first 2 hours until oxidation was complete, thereafter the aeration was stopped.
However, despite the specificity of enzymes in general, the one-pot system is not always possible. For example when the substrate used in the first step of the cascade can be used by subsequent enzymes. Such a situation was described by Burns and co-workers (2021), where a two-step cascade was used for the production of tert-butyl ((2R,4R)-2-methyltetrahydro-2Hpyran-4-yl)carbamate, an intermediate in a pharmaceutical under development to treat Parkinson's disease. [167] Here, a keto reductase (KRED) was used to resolve a corresponding racemic ketone by reduction of the unwanted ketone enantiomer, in this case the (2S)-ketone. The remaining (2R)-ketone was then converted to its corresponding amine by a transaminase with isopropylamine as the amine donor. As these two steps were incompatible given their cross reactivity, this cascade was operated as a sequential one-pot system at laboratory scale, where the transaminase was added to the reactor after a specified time to ensure complete reduction by the KRED in the first step. Unfortunately, due to higher enzyme loadings used at large scale, and thus potential issues with emulsion formation, the KRED needed to be removed after the first step resulting in a 2-step-2-pot cascade with intermediate separation similar to that illustrated in Figure 8. Here, a 1,000 L glass-lined reactor was used, in which the first step was run until completion prior to filtration and subsequent KRED removal. The filtrate was added to the same reactor, before transaminase, isopropylamine, etc., were added with a resulting working volume of 800 L. The final conversion was completed with nitrogen headspace sweeping. Overall, this resulted in a multi-kilogram linear cascade with a transaminase step, the size of which does not appear to have been reported previously. [167] In a previous study, the same group published an article with a chemoenzymatic route to chiral intermediates for the synthesis of a gamma secretase inhibitor, where a transaminase and an alcohol dehydrogenase were used. [184] However, unlike the recently published cascade, the transaminase and alcohol dehydrogenase used were not coupled in sequence, but instead used in parallel in separate reaction vessels to create intermediates, which were subsequently combined in additional reactions. The article has been included due to the noteworthy production scale of the transaminase catalyzed reaction. Here, a large tank was used with approximately 1,000 L working volume, which is the largest scale published in the last decade based on current knowledge.
Despite the few reported cascades at a larger than laboratory scale to date, these nevertheless show the potential for such systems in an industrial setting.

Downstream processing
The difficulties now come with the downstream processing of both batch-and cascade systems. If it is not possible to use in situ product removal for such systems, how will product inhibition be dealt with? And if a small excess of amine donor is required to drive the equilibrium, how do you recover the product from a product mixture not only containing the enzyme, PLP, buffer and co-product but also the excess donor? And if enzymatic co-factor regeneration is used, how to deal with the extra substrates/products this would add to the overall product mixture?
A general downstream process for all single-step transaminase catalyzed synthesis presented in Section 4 would include an enzyme removal step immediately after the reaction completion. As written in Section 5, if immobilized transaminases are used, this would be done by simple filtration and subsequent reuse of the enzyme. Contrarily, enzyme reuse is more difficult if soluble enzymes are used, but not impossible given the product mixture could potentially be filtered through a membrane with small enough pores so the soluble enzymes cannot pass. However, the feasibility of this solution in regard to enzyme stability and loss of activity from one batch to another is unclear but expected to be low. Instead, soluble enzymes are commonly denatured and then removed from the product mixture either by filtration or centrifugation and subsequently discarded. In theory, the denaturation can be done either through heating or by change of pH. At larger scale production, change of pH by acid-or base addition is the best solution, given the immense energy cost it would require heating up the reaction solution.
After enzyme removal, the subsequent downstream processing required depends on which method was used to drive the equilibrium as presented in Section 4. Overall, transaminase catalyzed reactions are complex systems, given the substrates are a ketone and an amine and the products are a new ketone and a new amine, which complicates the downstream processing. In addition, the product mixture will also include PLP and potentially DMSO or similar co-solvent used to ensure better substrate solubility.
If isopropylamine is used as amine donor, the resulting product mixture after enzyme removal will consist of the produced amine as well as acetone as co-product, unreacted substrate and excess isopropylamine. Fortunately, given both acetone and isopropylamine has much lower boiling points than water, these can easily be removed from the product mixture by distillation.
Given the produced amine commonly has a higher boiling point than water, this cannot be recovered by distillation as this would require removal of water from the mixture, which is not a feasible solution at large scale due to the excessive energy costs this would require. Instead, amines are commonly recovered either by ion exchange chromatography or crystallization.
Ion exchange chromatography can be done easily with amines, as these are positively charged at low pH and can thus be recovered on a cation exchange column. However, if the excess amine donor is not removed prior to the cation exchange column, these will also be adsorbed on the resin, and thus a large column would be required to recover all the charged amines in the-product mixture.
Crystallization of the produced amine was presented by Hülsewede and co-workers, in which continuous crystallization was as in situ product removal, albeit with a whole-cell system. [202] However, this can also be used as a final recovery step for the produced amine, which can then be filtered off from the product mixture.
Unfortunately, if most other amine donors than isopropylamine is used, the downstream processing is more complex, given the excess amine donor cannot be removed by distillation due to high boiling points. In this case, the excess amine donor would have to be recovered along with the product, and then removed from the product by chromatography or similar methods. This is unless a smart amine donor is used the spontaneously precipitate or crystallize, which can then easily be removed from the product mixture by filtration.
Moreover, if enzymatic regeneration or reduction is used, extra substrates are added which results in additional products, which must be considered. For the LDH/GDH enzymatic system, this includes glucose and gluconic acid for example. In this case, both glucose and gluconic acid are cheap compounds, so these do not necessarily need to be recovered and reused. However, whether it is possible to discard the reaction mixture directly after amine recovery, or if additional steps are required, before is not known.
This is also the case if DMSO is used, given this co-solvent cannot be removed by distillation due to its high boiling point. Thus, given it is assumed high DMSO concentrations cannot be released to the environment with the wastewater, this will have to be removed from the reaction mixture even after amine recovery. How to do this is not known, but it could be an argument to either not use a co-solvent, or switch to another with a lower boiling point that could easily be removed from the reaction media and potentially reused. This could for example be heptane or acetonitrile, however these have higher toxicity than DMSO, so extra precautions at a larger scale production would be required.
Generally, the downstream processing of these type of systems is not something that is researched, which may be an issue for scale-up. Based on publications in the last decade, it is clear that the bottleneck for transaminase catalyzed synthesis has moved from the reaction equilibrium to the downstream processing. Thus, this area of research should be in focus in the next decade.
To summarize, Table 5 lists the potential process solutions for potential scale-up challenges associated with batch conversion as presented in this review.