From Enzyme to Preparative Cascade Reactions with Immobilized Enzymes: Tuning Fe(II)/ α -Ketoglutarate-Dependent Lysine Hydroxylases for Application in Biotransformations

: Fe(II)/ α -ketoglutarate-dependent dioxygenases (KDOs) catalyze a broad range of selective C–H oxidation reactions. However, the difﬁcult production of KDOs in recombinant E. coli strains and their instability in puriﬁed form have so far limited their application in preparative biotransformations. Here, we investigated the immobilization of three KDOs ( Ca KDO, Cp KDO, Fj KDO) that catalyze the stereoselective hydroxylation of the L-lysine side chain using two one-step immobilization techniques (HaloTag ® , EziG™). The HaloTag ® -based immobilisates reached the best results with respect to residual activity and stability. In preparative lab-scale experiments, we achieved product titers of 16 g L − 1 (3 S )-hydroxy-L-lysine ( Ca KDO) and (4 R )-hydroxy-L-lysine ( Fj KDO), respectively, starting from 100 mM L-lysine. Using a HaloTag ® -immobilized lysine decarboxylase from Selenomonas ruminantium ( Sr LDC), the (3 S )-hydroxy-L-lysine from the Ca KDO-catalyzed reaction was successfully converted to (2 S )-hydroxy-cadaverine without intermediate product puriﬁcation, yielding a product titer of 11.6 g L − 1 in a 15 mL consecutive batch reaction. We propose that covalent in situ immobilization is an appropriate tool to access the preparative potential of many other KDOs.


Introduction
C-H functionalization is a chemically challenging reaction because carbon hydrogen bonds are relatively inert, making chemo-, regio-, and stereo-selectivity hard to control with conventional chemical catalysts [1]. The most commonly used enzyme class for such reactions up to now is P450 monooxygenases. However, the application of these enzymes on a preparative scale is often limited due to issues with inefficient electron transfer, uncoupling reactions, low activity and stability, and the requirement of expensive redox cofactors [2,3]. Another promising enzyme class for C-H functionalization are non-heme Fe(II)/α-ketoglutarate-dependent dioxygenases (KDOs, EC 1.14.11.), which make up a large superfamily of enzymes utilizing Fe(II) as a cofactor. KDOs catalyze the oxidative decarboxylation of their cosubstrate α-ketoglutarate (α-KG) towards succinate and CO 2 . The enzymatic reaction activates O 2 , which then can be used in a set of different oxidation reactions including hydroxylation, halogenation, ring closure, desaturation, epimerization, ring expansion, and epoxidation [4][5][6][7][8]. A great advantage of KDOs is that they are selfsufficient, as they do not need specific reductases or expensive biological redox cofactors. Many KDOs are associated with natural product biosynthesis pathways in bacteria, fungi, plants, and vertebrates, where the most studied pathways include the biosynthesis of Therefore, most applications of KDOs in biotransformation employ whole recombi nant cells or respective cell-free extracts [11,13,14,24,27]. Since some KDOs, such a CaKDO, CpKDO, and FjKDO, show very low specific activities [23], high concentration of whole cells [24] or cell-free extracts [11] are required. In both cases, side reactions may occur due to the (potentially higher) activity of other enzymes. Mass transfer, oxygen transfer, and product separation are also often hampered by the high concentration o Three major problems tend to occur when working with KDOs in general: (1) The enzyme yield from recombinant E. coli hosts is often low with a large fraction of insoluble non-active inclusion bodies, which can be partly prevented by coexpression of chaperones [11,13,[27][28][29]; (2) Purification and storage of these enzymes is challenging, due to the requirement of Fe(II) as a cofactor, which must be prevented from oxidation and dissociation from the active site [25,29]; (3) Precipitation occurs under oxidative conditions due to the instability of the enzyme [28,29].
Therefore, most applications of KDOs in biotransformation employ whole recombinant cells or respective cell-free extracts [11,13,14,24,27]. Since some KDOs, such as CaKDO, CpKDO, and FjKDO, show very low specific activities [23], high concentrations of whole cells [24] or cell-free extracts [11] are required. In both cases, side reactions may occur due to the (potentially higher) activity of other enzymes. Mass transfer, oxygen transfer, and product separation are also often hampered by the high concentration of cellular components. In comparison, working with isolated enzymes allows a much more flexible process optimization [28,30]. On the other hand, enzyme purification is expensive; soluble enzymes often have low stability; they cannot be recycled and might complicate reaction engineering and product purification, as in the case of whole cells and cell-free extracts. Furthermore, the enzyme is usually the biggest cost factor in enzyme-catalyzed syntheses [31]. Thus, efficient immobilization techniques are crucial to increase process economy specifically for complex enzymes such as KDOs that are difficult to produce and show only low activity [23]. Thus, we tested two different one-step immobilization techniques to concentrate the biocatalyst directly from crude-cell extracts, increase its stability, and enable recycling.
There is hardly any application of immobilized KDOs in the literature, probably because many immobilization techniques require purified enzymes beforehand. During the course of this study, the Kourist group published the application of immobilized N-succinyl-L-amino acid dioxygenase SadA on EziG™ Amber for the production of N-succinyl-β-hydroxy-L-valine on a preparative lab scale [28]. EziG™ consists of a specific controlled pore glass (CPG), coated with an organic polymer layer, and was recently developed as a one-step immobilization from cell-free extracts for proteins with a poly-histidine tag [32]. Instead of nickel or cobalt ions, iron is chelated on the surface of the respectively modified carriers. EziG™ beads are available with three surface modifications with different hydrophobicity: Amber, Coral, and Opal. The advantage of this non-covalent immobilization technique is the recycling of the carrier by removing inactive enzyme with imidazole. On the other hand, a clear disadvantage is possible enzyme leakage. Here, covalent immobilization using HaloTag ® represents a good alternative. HaloTag ® is a mutated dehalogenase, which recognizes terminal chloroalkane residues on any respectively modified carrier material and instantly forms a covalent ester bond between the carrier, e.g., commercially available HaloLink™ resin, and an aspartate residue in the active site of HaloTag ® [33,34]. The advantage of this covalent immobilization technique is the prevention of enzyme leakage and the high residual activity of 35-65% relative to the soluble enzyme [34][35][36]. In addition, HaloTag ® can enhance protein solubility, which is specifically advantageous to prevent the inclusion body formation of respective fusion proteins [37].
In this study, three different KDOs, which catalyze the stereoselective hydroxylation of the L-lysine side chain in the 3-position (CaKDO from Catenulispora acidiphila) and the 4-position (CpKDO from Chitinophaga pinensis or FjKDO from Flavobacterium johnsoniae) [22,23], were investigated for their potential application in a preparative lab scale. First, KDO production and purification were optimized in order to increase the soluble protein production and enzyme stability during purification. Then, we tested two onestep immobilization techniques (HaloTag ® , EziG™), followed by application of all three KDO-HaloTag ® variants immobilized on HaloLink™ resin in repetitive batch experiments. HaloTag ® -immobilized CaKDO and FjKDO were then selected for preparative-scale biotransformations.

KDO Production and Purification
In the present study, a previously described protocol for CaKDO production using coexpressed chaperones (GroEL/GroES) [11] was successfully applied to enhance the soluble production of CaKDO, CpKDO, and especially, FjKDO, as well as for their HaloTag ® fusions (Supplementary Materials, Figure S1). Without coexpression of chaperones, these enzymes were barely active and rapidly precipitated already from the cell-free extracts (data not shown). As can be seen by SDS-PAGE analysis, chaperones are still present even after purification and immobilization (Supplementary Materials, Figures S1 and S2) due to obviously strong binding to the target enzyme, which was described for several proteins before [38].
Initial tests demonstrated that freeze-drying is the best option to maintain the activity of KDOs after immobilized metal affinity chromatography (IMAC), which prevents using HEPES buffer or the addition of 10vol% glycerin, which both stabilize the enzymes in solution for a short time (data not shown). As earlier reported [29], we also observed the loss of activity after elution from IMAC when we tried to purify CpKDO in TRIS buffer (Supplementary Materials, Figure S6B). Since CaKDO and FsKDO showed a higher degree of ordered structure in structural investigations upon binding of Fe(II) and α-KG [25], we presumed a positive effect on the enzyme stability upon addition of these cofactors and optimized the IMAC purification protocol, respectively. We used a combination of sodium phosphate buffer with low concentrations of the cosubstrate α-KG, as well as L-ascorbic acid and dithiothreitol (DTT) as reducing agents. Precipitation and inactivation of all three KDOs was successfully prevented by the addition of the Fe(II) cofactor immediately after IMAC, and the desalting step took place in the presence of α-KG, Fe(II) and the reducing agents, followed by lyophilization of the enzyme from the same mixture (Supplementary Materials, Section S2.1.2).
While we were able to improve the soluble KDO production and purification, purification of these enzymes is laborious and costly, and the enzyme yield is low. Furthermore, all components applied during the desalting step contaminate the lyophilisate, decreasing the protein content to 10-35%. This and the low enzyme yield consequently lead to problems when utilizing the lyophilisate for reactions. Furthermore, precipitation due to the instability of purified CaKDO during biotransformation remains an issue (Supplementary Materials, Figure S7).

Immobilization and Reaction Optimization with KDOs
We tested the immobilization of KDOs as a suitable reaction engineering approach to increase enzyme stability and avoid enzyme purification, simultaneously, and compared two simple one-step immobilization techniques that allow immobilization directly from the cell-free extract: HaloTag ® and EziG™.
In order to investigate the binding capacities of the different carriers, we quantified the enzyme concentration on the beads using the BCA assay (Supplementary Materials, Section S3.1.1) and confirmed the enzyme immobilization qualitatively by SDS PAGE (Supplementary Materials, Figure S8) capacity. This is lower or in the lower range of the binding capacities specified in the manufacturer's information (15-60% w/w) [32,39].
Both immobilization techniques were compared by measuring the specific activities of the immobilisates relative to the free purified enzymes with His-Tag ( Figure 2A). Immobilization of CaKDO via HaloTag ® , EziG™ Amber, and Opal increased the specific activities, with the HaloTag ® immobilisate showing the highest residual activity (280 ± 39%) compared to the free enzyme without HaloTag ® . The EziG™ Coral immobilisate showed similar specific activity compared to the free enzyme (95 ± 0.9%). For CpKDO, all immobilized variants were less active compared to the free enzyme. The highest residual activity was measured with the HaloTag ® immobilisate (70 ± 1.5%). Likewise, all immobilized FjKDO preparations were less active than the free variant, with the highest residual activity (62 ± 7.6%) for the EziG™ Opal variant. Here, the HaloTag ® immobilization resulted in only moderate residual activities of about 43 ± 0.3% (Figure 2A). These results demonstrate again the different performance of immobilization strategies even with highly similar enzymes.
Both immobilization techniques were compared by measuring the specific activities of the immobilisates relative to the free purified enzymes with His-Tag ( Figure 2A). Immobilization of CaKDO via HaloTag ® , EziG™ Amber, and Opal increased the specific activities, with the HaloTag ® immobilisate showing the highest residual activity (280 ± 39%) compared to the free enzyme without HaloTag ® . The EziG™ Coral immobilisate showed similar specific activity compared to the free enzyme (95 ± 0.9%). For CpKDO, all immobilized variants were less active compared to the free enzyme. The highest residual activity was measured with the HaloTag ® immobilisate (70 ± 1.5%). Likewise, all immobilized FjKDO preparations were less active than the free variant, with the highest residual activity (62 ± 7.6%) for the EziG™ Opal variant. Here, the HaloTag ® immobilization resulted in only moderate residual activities of about 43 ± 0.3% (Figure 2A). These results demonstrate again the different performance of immobilization strategies even with highly similar enzymes.  Since the HaloLink™ resin is commercially available, shows better binding capacities, and for two of the three KDOs, the HaloTag ® immobilization worked best, we decided to continue our work with HaloTag ® -immobilized KDOs.
While phosphate buffer was used for the purification of the enzymes, HEPES buffer was found to be better suited for biotransformations (data not shown). This is most likely, because the Fe(II) present in the reaction mixture tends to oxidize in aqueous systems. This reaction triggers a reaction called the Fenton reaction, leading to the generation of reactive oxygen species (ROS), which can attack the enzyme and impair its stability. It was shown in previous studies that the amount of formed ROS correlates with the buffer and the pH used and is lower for HEPES buffer in comparison to other buffers [40][41][42]. Furthermore, buffers such as HEPES and MOPS are more suitable for reaction systems incorporating metal ions due to their lower metal-binding constants compared to other buffers, such as TRIS or phosphate buffers [43]. One way to deal with the generated ROS is to add catalase to the reaction [44]. We tested the addition of catalase exemplarily with both CaKDO preparations, as this enzyme showed the highest activity, but the lowest stability in the free form among the tested L-lysine hydroxylases (see below). As demonstrated in Figure 2B, catalase was beneficial for the biotransformation with free CaKDO, whereas there was only a negligibly higher conversion for the reaction with HaloTag ® -immobilized CaKDO, which does not justify the application of catalase.
Next, we compared the free KDOs with His-Tag to their respective HaloTag ® variants immobilized on the HaloLink™ resin in terms of productivity and stability under the reaction conditions (Figure 3). The stabilizing effect of immobilization was most pronounced for CaKDO, where the CaKDO-HaloTag ® immobilisate outperformed the free variant already after 1 h of reaction time. While conversion with the free variant stopped after 10%, CaKDO-HaloTag ® fully converted 100 mM L-lysine to (3S)-hydroxy-L-lysine in 24 h. We could demonstrate that the higher stability was a result of the immobilization and not of the HaloTag ® fusion (Supplementary Materials, Figure S9). For CpKDO and FjKDO, both variants, the free and the HaloTag ® immobilisate, were stable over the reaction time of 24 h, but reached only 53-79% conversion until the reaction was stopped, which is in line with the lower specific activity of both immobilized enzymes compared to CaKDO-HaloTag ® ( Figure 2A). As these enzymes are still active after 24 h, full conversion could easily be achieved by a higher enzyme concentration or a prolonged reaction time. The slightly faster conversion observed with free CpKDO and FjKDO relative to the immobilized variants was due to the higher molecular mass of 35 kDa of the HaloTag ® -fusions ( Figure 3B,C).
thermore, buffers such as HEPES and MOPS are more suitable for reaction systems incorporating metal ions due to their lower metal-binding constants compared to other buffers, such as TRIS or phosphate buffers [43]. One way to deal with the generated ROS is to add catalase to the reaction [44]. We tested the addition of catalase exemplarily with both CaKDO preparations, as this enzyme showed the highest activity, but the lowest stability in the free form among the tested L-lysine hydroxylases (see below). As demonstrated in Figure 2B, catalase was beneficial for the biotransformation with free CaKDO, whereas there was only a negligibly higher conversion for the reaction with HaloTag ® -immobilized CaKDO, which does not justify the application of catalase.
Next, we compared the free KDOs with His-Tag to their respective HaloTag ® variants immobilized on the HaloLink™ resin in terms of productivity and stability under the reaction conditions ( Figure 3). The stabilizing effect of immobilization was most pronounced for CaKDO, where the CaKDO-HaloTag ® immobilisate outperformed the free variant already after 1 h of reaction time. While conversion with the free variant stopped after 10%, CaKDO-HaloTag ® fully converted 100 mM L-lysine to (3S)-hydroxy-L-lysine in 24 h. We could demonstrate that the higher stability was a result of the immobilization and not of the HaloTag ® fusion (Supplementary Materials, Figure S9). For CpKDO and FjKDO, both variants, the free and the HaloTag ® immobilisate, were stable over the reaction time of 24 h, but reached only 53-79% conversion until the reaction was stopped, which is in line with the lower specific activity of both immobilized enzymes compared to CaKDO-HaloTag ® ( Figure 2A). As these enzymes are still active after 24 h, full conversion could easily be achieved by a higher enzyme concentration or a prolonged reaction time. The slightly faster conversion observed with free CpKDO and FjKDO relative to the immobilized variants was due to the higher molecular mass of 35 kDa of the HaloTag ®fusions ( Figure 3B,C).  To maximize the productivity, the L-lysine concentration was increased from 100 mM to 500 mM. Full conversion of 200 mM L-lysine to the corresponding hydroxy-L-lysines was possible with all three KDOs in a 3-4 h reaction time with respectively higher enzyme concentrations of 6.5-7.5 mg mL −1 (Figure 4). It is obvious that reactions starting from 200 mM L-lysine proceeded slightly more slowly relative to those with 100 mM, which was probably caused by oxygen limitation, substrate inhibition, or other kinetic effects. The conversion of 500 mM L-lysine was tested with immobilized CaKDO-HaloTag ® , which yielded 82% conversion within 26 h ( Figure 4A). Aeration in this simple reaction setup using a 5 mL reaction tube attached to an overhead shaker was achieved by opening of the tube every 15 min. The reaction stopped temporarily overnight, due to oxygen depletion, and started again the next day after aeration with the same velocity, as can be deduced from the slope of the conversion curve ( Figure 4A). Yet, full conversion of 500 mM L-lysine within about a 12 h reaction time is most probably possible in a reaction setup with continuous aeration, e.g., by performing the reaction in a shaking flask [11,24]. yielded 82% conversion within 26 h ( Figure 4A). Aeration in this simple reaction setup using a 5 mL reaction tube attached to an overhead shaker was achieved by opening of the tube every 15 min. The reaction stopped temporarily overnight, due to oxygen depletion, and started again the next day after aeration with the same velocity, as can be deduced from the slope of the conversion curve ( Figure 4A). Yet, full conversion of 500 mM L-lysine within about a 12 h reaction time is most probably possible in a reaction setup with continuous aeration, e.g., by performing the reaction in a shaking flask [11,24].

Repetitive Batch Studies
In addition to the benefits immobilization offers on enzyme stabilization, it also enables recycling of the catalyst, which is decisive for the process economy. Recyclability of CaKDO, CpKDO, and FjKDO immobilized via HaloTag ® was tested in repetitive batch studies ( Figure 5).

Repetitive Batch Studies
In addition to the benefits immobilization offers on enzyme stabilization, it also enables recycling of the catalyst, which is decisive for the process economy. Recyclability of CaKDO, CpKDO, and FjKDO immobilized via HaloTag ® was tested in repetitive batch studies ( Figure 5). yielded 82% conversion within 26 h ( Figure 4A). Aeration in this simple reaction setup using a 5 mL reaction tube attached to an overhead shaker was achieved by opening of the tube every 15 min. The reaction stopped temporarily overnight, due to oxygen depletion, and started again the next day after aeration with the same velocity, as can be deduced from the slope of the conversion curve ( Figure 4A). Yet, full conversion of 500 mM L-lysine within about a 12 h reaction time is most probably possible in a reaction setup with continuous aeration, e.g., by performing the reaction in a shaking flask [11,24].

Repetitive Batch Studies
In addition to the benefits immobilization offers on enzyme stabilization, it also enables recycling of the catalyst, which is decisive for the process economy. Recyclability of CaKDO, CpKDO, and FjKDO immobilized via HaloTag ® was tested in repetitive batch studies ( Figure 5). After four batches, CpKDO-HaloTag ® still gave 84% conversion in 4 h, while FjKDO-HaloTag ® converted 100% in 3 h ( Figure 5B,C). Even after seven batches, FjKDO-HaloTag ® catalyzed the conversion by 27% in 4 h (data not shown). This corresponds to a specific space-time yield of 2333 g product L −1 h −1 per g immobilized CpKDO and 4803 g product L −1 h −1 per g immobilized FjKDO. By contrast the single batch reactions gave a space-time yield of 795 g product L −1 h −1 per g immobilized CpKDO and 1081 g product L −1 h −1 per g immobilized FjKDO , showing that a recycling approach can effectively increase the reaction efficiency. Factors such as constant shaking, which can lead to friction between the beads and enzyme inactivation, as well as the partial loss of the immobilisate during the washing steps between batches might lead to the loss of active enzyme. Since the stability of KDOs is a major concern anyway, the little loss of activity after four batches for CpKDO-HaloTag ® and after seven batches for FjKDO-HaloTag ® exceeded our expectations, especially since previous experiments with SadA immobilized on EziG Amber showed only 10% of the initial reaction rate after the first reaction cycle [28].
Unfortunately, recycling of CaKDO-HaloTag ® was not that easy, as the catalyst was already almost inactive after the first batch ( Figure 5A). Remarkably, the reaction mix showed a blue color after the first reaction, which occurred after L-lysine was fully consumed (Supplementary Materials, Figure S11). Similar findings were already reported for the 2,4-dichlorophenoxyacetate oxygenase (TfdA) [29,45]. MS-analyses suggested that hydroxylation of a tryptophan residue close to the iron binding site of TfdA occurs in absence of the primary substrate. The tryptophan residue can then chelate the Fe(III) ion located in the active site, which was assumed to be the origin of the blue color. Upon treatment with dithionite, dialysis with EDTA, and reconstitution of Fe(II) in the active site, 81% activity could be restored, most likely due to a displacement of the Fe(III) from the oxidized tryptophan [29,45]. However, in the case of CaKDO, no aromatic residue is close enough to the active site to explain the blue color by an analogous mechanism (Supplementary Materials, Figure S13). Furthermore, in our case, only the reaction mix, not the immobilisate, appeared blue. Since the goal of this work was the application of KDOs in a preparative lab scale, this aspect was not further investigated, but we tested if treatment with dithionite and EDTA could regenerate the activity of the immobilized CaKDO after the first batch. Indeed, it was possible to regain activity, and the regenerated immobilisate was only slightly less active compared to the first batch (Supplementary Materials, Figure S12). These results represent a good basis to regenerate immobilized CaKDO more frequently.

Preparative Lab-Scale Reactions with CaKDO-HaloTag ® and FjKDO-HaloTag ®
Next, CaKDO-HaloTag ® and FjKDO-HaloTag ® were chosen for a reaction on a preparative lab scale for the synthesis of (3S)-hydroxy-L-lysine and (4R)-hydroxy-L-lysine, respectively. Reaction optimization with free CaKDO at a small scale (1 mL) yielded the optimal reaction parameters (pH, temperature, Fe(II) concentration, use of additives), which were mostly in line with the results already published by Baud et al. [23], for the analytical scale. Only the optimal reaction temperature of the CaKDO reaction was found at 20 • C, whereas the optimal reaction temperature of FjKDO was at 25 • C (data not shown). The optimal reaction parameters were used for both immobilized enzymes to convert 100 mM L-lysine in a 15 mL reaction in non-baffled shaking flasks ( Figure 6).
For both reactions, full conversion was reached in less than 24 h, with product titers of 16 g L −1 and a total product amount of 240 mg (Figure 6), corresponding to specific space-time yields of 73.4 g product L −1 h −1 per g immobilized CaKDO and 133.65 g product L −1 h −1 per g immobilized FjKDO .
In order to increase the scale further, different reaction setups were tested with the immobilized CaKDO-HaloTag ® . In a 10 mL shaking flask reaction, full conversion of 200 mM L-lysine was reached in 20 h (Supplementary Materials, Figure S15A) corresponding to a product titer of 32.4 g L −1 and a specific space-time yield of 100 g L −1 h −1 per g immobilized CaKDO . Continuous aeration was also tested using a synthesis workstation at a 50 mL scale, where full conversion of 100 mM L-lysine was successfully achieved in 70 h (Supplementary Materials, Figure S15B). This corresponds to a product titer of 16.2 g L −1 , but a specific space-time yield of only 4.63 g L −1 h −1 per g immobilized CaKDO due to the longer reaction time compared to the shaking flask experiments. Since the aeration rate could not be controlled in our synthesis workstation, we suspected that not only the increased scale, but also an oxygen limitation prolonged the reaction time. We figured out that when working with isolated and immobilized KDOs, the aeration must be carefully balanced. Too little oxygen limits the reaction, but too much oxygen can increase the oxidation of the Fe(II) cofactor, making it either unavailable for the enzyme and/or decreasing the enzyme stability due to the presence of ROS. Often, a simpler setup in shaking flasks can already be sufficient [11,24]. Here, the filling volume and mixing speed must be assessed to provide adequate oxygen supply. Our results demonstrate that an increase in scale and substrate concentration for KDOs is in general possible using immobilized enzymes in combination with an open reaction system for oxygen supply.  Figure S15B). This corresponds to a product titer of 16.2 g L −1 , but a specific space-time yield of only 4.63 g L -1 h -1 per gimmobilized CaKDO due to the longer reaction time compared to the shaking flask experiments. Since the aeration rate could not be controlled in our synthesis workstation, we suspected that not only the increased scale, but also an oxygen limitation prolonged the reaction time. We figured out that when working with isolated and immobilized KDOs, the aeration must be carefully balanced. Too little oxygen limits the reaction, but too much oxygen can increase the oxidation of the Fe(II) cofactor, making it either unavailable for the enzyme and/or decreasing the enzyme stability due to the presence of ROS. Often, a simpler setup in shaking flasks can already be sufficient [11,24]. Here, the filling volume and mixing speed must be assessed to provide adequate oxygen supply. Our results demonstrate that an increase in scale and substrate concentration for KDOs is in general possible using immobilized enzymes in combination with an open reaction system for oxygen supply.
Different other groups have already worked on the production of hydroxy-L-lysines via a KDO-catalyzed reaction, as summarized in Table 1. Apart from Baud et al. [22,23], who used IMAC-purified enzymes, most groups applied cell-free extracts or whole cells. Working with isolated enzymes resulted in low product titers and total yields (1.6 g L −1 or 0.016 g total yields) [22,23]. We were able to increase these titers 10-20-times by increasing the substrate concentration, which was possible due to the optimized production, increased stability, and recyclability by immobilization. Remarkably, our product titers of 32 g L −1 (3S)-hydroxy-L-lysine are comparable to the product titers of 32.43-43 g L −1 obtained with whole cells on a 40 mL scale, as reported by Hara et al. [24]. Different other groups have already worked on the production of hydroxy-L-lysines via a KDO-catalyzed reaction, as summarized in Table 1. Apart from Baud et al. [22,23], who used IMAC-purified enzymes, most groups applied cell-free extracts or whole cells. Working with isolated enzymes resulted in low product titers and total yields (1.6 g L −1 or 0.016 g total yields) [22,23]. We were able to increase these titers 10-20-times by increasing the substrate concentration, which was possible due to the optimized production, increased stability, and recyclability by immobilization. Remarkably, our product titers of 32 g L −1 (3S)-hydroxy-L-lysine are comparable to the product titers of 32.43-43 g L −1 obtained with whole cells on a 40 mL scale, as reported by Hara et al. [24].  In the context of the preparative synthesis of hydroxy-L-lysines, product isolation must also be considered. The isolation of the target product is easier from less-complex reaction mixtures, which preferably only contain the target product without residual substrate or side products. The heterogeneity of reaction mixtures is definitely lower in reactions with isolated enzymes instead of cell-free extracts or whole-cell biocatalysts. For the present lysine hydroxylation, separation of hydroxy-L-lysines from residual L-lysine is specifically challenging due to their chemical and physical similarity. Thus, for integrated processes aiming for isolated hydroxy-L-lysines, only processes with full conversion can be considered. In our case, HPLC and GC-ToF-MS analyses demonstrated that the L-lysine was completely converted to the respective hydroxy-L-lysines and contained, besides α-KG, succinate, and HEPES, no further side products (Supplementary Materials, Section S12). With a two-step chromatographic purification [46], the organic acids were fully removed, although traces of HEPES buffer remained, as can be deduced from the NMR-spectra ( Supplementary Materials, Figures S27A,B and S32A,B).

Cascade Reaction towards (2S)-Hydroxy-Cadaverine
As previously shown by Baud et al. [22], coupling of the KDO reaction with a second step incorporating a (lysine) decarboxylase provides access to valuable hydroxycadaverines hydroxylated in the 2-and 3-position depending on the combination of the respective KDOs and lysine decarboxylases ( Figure 1). However, the reaction was performed with cell-free extract and limited to a substrate concentration of 10 mM at a 10 mL scale, with the KDO reaction being the limiting step [22]. Because (2S)-hydroxy-cadaverine is harder to produce chemically than 3-hydroxy-cadaverine, due to its chiral center, we concentrated on this cascade starting from 100 mM L-lysine in a 15 mL scale with immobilized CaKDO-HaloTag ® in the first step and a lysine decarboxylase from Selenomonas ruminantium (SrLDC) [22,47,48] in the second reaction step. This pyridoxal phosphate-(PLP)-dependent enzyme accepts besides L-lysine and L-ornithine [49] also (3S)-hydroxy-L-lysine as a substrate [22].
Likewise, for SrLDC, we compared the two immobilization techniques: HaloTag ® and EziG™, also with the goal of enzyme recycling. Since (3S)-hydroxy-L-lysine is not commercially available, all experiments concerning the immobilization, optimization of the reaction conditions, and repetitive batch experiments were carried out with L-lysine as a substrate.
As demonstrated in Figure 7A, soluble and HaloTag ® -immobilized SrLDC showed the same performance in the conversion of L-lysine to cadaverine, whereas respective immobilisates on EziG™ beads were less active. The enzyme load of the carrier was higher for the HaloLink™ resin (7.14 mg SrLDC-HaloTag ® per mL resin), while the enzyme load of the EziG™ beads was between 0.082 mg per mg EziG™ Opal beads and 0.126 mg per mg EziG™ Coral beads. Similar to the KDOs, SrLDC binds better to the HaloLink™ resin, showing a similar specific activity as the free enzyme, as can be deduced from the conversion curve ( Figure 7A).
Next, the HaloTag ® -immobilized SrLDC was tested for its activity towards the different substrates (L-lysine, L-ornithine, (3S)-hydroxy-L-lysine). After a 5 h reaction time, almost full conversion of the substrate L-lysine (100 mM) was achieved, whereas the conversion of L-ornithine and (3S)-hydroxy-L-lysine occurred significantly more slowly ( Figure 7B). Yet, full conversion of L-ornithine to putrescin was achieved after 72 h. At this point, only 11% of (3S)-hydroxy-L-lysine was converted to (2S)-hydroxy-cadaverine, showing the low activity of SrLDC for this substrate. However, it has to be considered that (3S)-hydroxy-L-lysine was applied in the form of a supernatant taken from a previous KDO reaction. Therefore, other components in the reaction mixture might also lead to a decrease in activity.
Simultaneously, important reaction parameters for the SrLDC-HaloTag ® reaction from L-lysine to cadaverine were investigated. The influence of pH, substrate concentration, temperature, and the concentration of the cofactor PLP on the reaction was tested (Supplementary Materials, Section S7). the same performance in the conversion of L-lysine to cadaverine, whereas respective immobilisates on EziG™ beads were less active. The enzyme load of the carrier was higher for the HaloLink resin (7.14 mg SrLDC-HaloTag  per mL resin), while the enzyme load of the EziG™ beads was between 0.082 mg per mg EziG™ Opal beads and 0.126 mg per mg EziG™ Coral beads. Similar to the KDOs, SrLDC binds better to the HaloLink™ resin, showing a similar specific activity as the free enzyme, as can be deduced from the conversion curve ( Figure 7A). Next, the HaloTag ® -immobilized SrLDC was tested for its activity towards the different substrates (L-lysine, L-ornithine, (3S)-hydroxy-L-lysine). After a 5 h reaction time, almost full conversion of the substrate L-lysine (100 mM) was achieved, whereas the conversion of L-ornithine and (3S)-hydroxy-L-lysine occurred significantly more slowly (Figure 7B). Yet, full conversion of L-ornithine to putrescin was achieved after 72 h. At this point, only 11% of (3S)-hydroxy-L-lysine was converted to (2S)-hydroxy-cadaverine, showing the low activity of SrLDC for this substrate. However, it has to be considered that (3S)-hydroxy-L-lysine was applied in the form of a supernatant taken from a previous KDO reaction. Therefore, other components in the reaction mixture might also lead to a decrease in activity.
Simultaneously, important reaction parameters for the SrLDC-HaloTag ® reaction from L-lysine to cadaverine were investigated. The influence of pH, substrate concentration, temperature, and the concentration of the cofactor PLP on the reaction was tested (Supplementary Materials, Section S7).
The highest cadaverine concentration after 5 h (67.87 ± 2.72%) was obtained starting from 100 mM L-lysine, whereas L-lysine concentrations > 100 mM resulted in lower conversion, which could be explained by possible substrate inhibition of the enzyme (Supplementary Materials, Figure S16C). In short-term experiments (20 min), the tested PLP concentrations in the range of 0.05-2 mM gave identical results (Supplementary Materials, The highest cadaverine concentration after 5 h (67.87 ± 2.72%) was obtained starting from 100 mM L-lysine, whereas L-lysine concentrations > 100 mM resulted in lower conversion, which could be explained by possible substrate inhibition of the enzyme (Supplementary Materials, Figure S16C). In short-term experiments (20 min), the tested PLP concentrations in the range of 0.05-2 mM gave identical results (Supplementary Materials, Figure S16B). Since supplementation of PLP is known to be beneficial for LDCs and the cofactor is unstable at room temperature and towards light exposure [50], 1 mM PLP was used for further experiments. Additionally, a pH of 7 and a reaction temperature of 35 • C were found to be optimal (Supplementary Materials, Section S7, Figure S16A,D).
Under optimized reaction conditions, the sequential cascade reaction was performed without intermediate purification ( Figure 8A). Full conversion of 100 mM L-lysine in the first reaction step was achieved after approximately a 10 h reaction time in a shaking flask. After 25 h, the supernatant containing the (3S)-hydroxy-L-lysine was transferred to a falcon tube, and 2.5 mg mL −1 of immobilized SrLDC was added. After 47 h, a conversion of 97% was reached, corresponding to a specific space-time yield of 6.5 g L −1 h −1 per g immobilized SrLDC and a product titer of 11.6 g L −1 (2S)-hydroxy cadaverine. The results demonstrate that the SrLDC reaction is not impaired by components from the KDO-catalyzed step, since increasing the enzyme concentration from 0.1 mg mL −1 to 2.5 mg mL −1 led to full conversion of (3S)-hydroxy-L-lysine to (2S)-hydroxy-cadaverine. Increased enzyme concentrations could principally decrease the reaction time for both steps further.
HPLC and GC-ToF-MS analyses demonstrated full conversion of the (3S)-hydroxy-L-lysine to (2S)-hydroxy-cadaverine (Supplementary Materials, Section S12), which enabled us to successfully purify the (2S)-hydroxy-cadaverine by a protocol from Fossey-Jouenne et al. [46] for NMR analysis (Supplementary Materials, Section S12) without any remaining impurities ( Figure S41A,B). This is in contrast to in vivo approaches with a Corynebacterium glutamicum strain overexpressing recombinant KDO genes and three different LDC genes for the production of (4R)-hydroxy-L-lysine and 3-hydroxy-cadaverine from L-lysine [21]. con tube, and 2.5 mg mL of immobilized SrLDC was added. After 47 h, a conversion of 97% was reached, corresponding to a specific space-time yield of 6.5 g L −1 h −1 per gimmobilized SrLDC and a product titer of 11.6 g L −1 (2S)-hydroxy cadaverine. The results demonstrate that the SrLDC reaction is not impaired by components from the KDO-catalyzed step, since increasing the enzyme concentration from 0.1 mg mL -1 to 2.5 mg mL -1 led to full conversion of (3S)-hydroxy-L-lysine to (2S)-hydroxy-cadaverine. Increased enzyme concentrations could principally decrease the reaction time for both steps further. HPLC and GC-ToF-MS analyses demonstrated full conversion of the (3S)-hydroxy-L-lysine to (2S)-hydroxy-cadaverine (Supplementary Materials, Section S12), which enabled us to successfully purify the (2S)-hydroxy-cadaverine by a protocol from Fossey-Jouenne et al. [46] for NMR analysis (Supplementary Materials, Section S12) without any remaining impurities ( Figure S41A,B). This is in contrast to in vivo approaches with a Similar product titers for 3-hydroxy-cadaverine (11.4 g/L) to our approach for the production of (2S)-hydroxy-cadaverine could be achieved, with a strain containing FjKDO and LDCc. However, the amount of by-product (cadaverine titer 39 g L −1 ) and intermediates ((4R)-hydroxy-L-lysine titer 4.1 g L −1 ) was high, most likely due to the substrate preference of the lysine decarboxylases for L-lysine and the generally lower activity of LDCs towards the hydroxy-L-lysines. While the constant supply of L-lysine and α-KG provided by the cellular metabolism in vivo is certainly advantageous, the pH inside living cells is hard to control relative to isolated enzymes, especially when pH-active compounds (lysine and cadaverine derivatives) are involved and cascade reactions include enzymes that are highly pH dependent [48,51]. Further, isolation of the target product 3-hydroxy-cadaverine from a mixture of substrate (L-lysine), intermediate ((4R)-hydroxy-L-lysine), by-product (cadaverine), and other cellular components is definitely challenging, due to the close physical and chemical properties of the molecules. Thus, in the case of cascades containing KDOs and LDCs, a sequential approach including isolated immobilized enzymes seems to be advantageous and simpler compared to in vivo approaches.
Next, we investigated if immobilized SrLDC-HaloTag ® could be recycled in repetitive batch experiments using L-lysine (100 mM) as a substrate ( Figure 8B). It can be clearly seen that the first reaction is the fastest and the turnover rates decrease less between the second and the last reaction (cycles 2-6) than between cycle 1 and 2. Still, a conversion of ≥ 94% could be achieved in the last cycle after 1 h. So far, immobilized SrLDC-HaloTag ® showed a reusability of at least six cycles in a 15 mL scale with little loss of activity and potentially even a higher number of cycles. This is a good basis for further reaction engineering of the cascade towards (2S)-hydroxy cadaverine. Because the enzyme activity towards (3S)-hydroxy-L-lysine is much lower than towards L-lysine, prolonged reaction times are necessary to achieve full conversion.
Besides, the product of the decarboxylation of L-lysine, cadaverine, is also an interesting industrial compound for the production of fully biobased polymers [15,52]. Biological production of cadaverine is commonly performed by fermentative microbial production [52], (immobilized) whole recombinant E. coli cells [15,53], or immobilized LDCs. Immobilization of LDCs was previously performed on poly(3-hydroxybutyrate) (P(3HB) biopolymers [54], chitin [55], or via different carrier-free immobilization methods, such as catalytically active inclusion bodies (CatIBs) [56], or in the form of crosslinked enzyme aggregates (CLEASs) [57]. Currently, most processes using immobilized enzymes use the constitutive (EcLDCc) or inducible (CadA) LDCs from E. coli. While CadA is active in a pH range between 5 and 6, it is rapidly inhibited at pH higher that 8.0 [58]. Furthermore, it is inhibited at higher concentrations of lysine [59] and cadaverine [60]. In contrast, the constitutive EcLDCc has a broader pH range [51] and is hardly inhibited by L-lysine [61]. To our knowledge, an application of SrLDC for the production of cadaverine has not yet been tested. As was recently extensively reviewed by Huang et al. [15], biological cadaverine production using fermentation, whole cells, and biotransformations with immobilized enzymes led to space-time yields between 10 g L −1 h −1 and 204 g L −1 h −1 . With HaloTag ® -immobilized SrLDC, we achieved a product titer of 58.4 g L −1 , corresponding to a specific space-time yield of 655 g L −1 h −1 per g immobilized SrLDC . Furthermore, HaloTag ®immobilized SrLDC was able to catalyze the full conversion of 100 mM L-ornithine on a 15 mL preparative lab reaction (Supplementary Materials, Figure S17), giving access to 8.8 g L −1 1,4-diaminobutane (putrescine), which is another interesting building block for the production of biobased polyamides [16].
Considering all these factors, SrLDC is an interesting enzyme for further investigation of its potential for the synthesis of cadaverine, putrescine, and respective hydroxylated derivatives.

Initial Rate Activity
For initial activity measurements, enzyme concentrations of 0.5-1 mg mL −1 for the free and immobilized variants of CaKDO, CpKDO, and FjKDO were used. Conversions were measured up to a maximum of 10% to ensure initial rate conditions. Due to the different residual activities of the immobilized enzymes, the assay conditions were respectively adapted to the enzyme preparation ( Table 2). The reaction was mixed at 25 • C in an overhead shaker for sufficient mixing of the beads and the reaction mixture

Analytical Scale Reactions
For analytical-scale reactions and initial rate activity measurements, the reaction was started by adding 1 mL reaction mix (3.3) to a 2 mL reaction tube containing either the lyophilized free enzyme or the immobilized enzyme. The reaction was mixed at 25 • C in an overhead shaker or vertically attached on a thermomixer to guarantee sufficient mixing of the beads and the reaction mixture.

Repetitive Batch Experiments
For repetitive experiments, 1 mL of the reaction mixture (3.3) was added to the immobilized enzyme (5 mg mL −1 CaKDO-HaloTag ® , CpKDO-HaloTag ® , FjKDO-HaloTag ® ) using 5 mL reaction tubes. In order to guarantee sufficient oxygen supply, the tubes were opened every 15 min. Samples (10 µL) were taken every 30 min over a period of 4 h for subsequent HPLC analysis. Afterwards the tubes were centrifuged, the beads were washed 4 times with 1 mL 50 mM HEPES buffer, pH 7.5, and stored overnight at 4 • C. The next day, the beads were washed once with 200 mM HEPES buffer, pH 7.5, before the reaction was started again with a freshly prepared reaction mixture. The procedure was repeated for three to seven batches, depending on the enzyme.

Regeneration of Immobilized CaKDO-HaloTag ®
After the first batch (3.3, 3.3.3), the beads were washed 4 times with 50 mM HEPES buffer, pH 7.5. After the addition of different concentrations of dithionite (1 or 10 mM) and 100 mM EDTA, the beads were shaken in an overhead shaker for 1 h at room temperature. Afterwards, the tubes were centrifuged; the beads were washed 4 times with 50 mM HEPES buffer, pH 7.5, to remove EDTA and dithionite and stored overnight at 4 • C. The next day, the beads were washed once with 200 mM HEPES buffer, pH 7.5, and the reaction was started again with a freshly prepared reaction mixture including 1 mM Fe(NH 4 ) 2 (SO 4 ) 2 , thereby restoring the cofactor for the reaction. As a control, one of the reactions was incubated only with 50 mM HEPES (Supplementary Materials, Figure S12D).

Reactions in Preparative Lab Scale
Conversion of 100 mM L-lysine in 15 mL: The reaction was started by adding 15 mL of reaction mix (3.3) with 100 mM L-lysine to a 50 mL Erlenmeyer flask without baffles containing 1.2 mg mL −1 and 1.3 mg mL −1 HaloTag ® -immobilized CaKDO and FjKDO, respectively. The reaction was mixed by orbital shaking at 150 rpm at 20 • C and 25 • C for CaKDO and FjKDO, respectively. Each reaction was performed as a technical duplicate. Samples (10 µL) were taken every hour over a period of 24 h. The reaction was quenched by heat inactivation at 80 • C for 5 min. Substrate and product concentrations were measured by HPLC (3.6).
Conversion of 200 mM L-lysine in 10 mL: The reaction was started by adding 10 mL of reaction mix (3.3) with 200 mM L-lysine and 1 mg mL −1 catalase (Sigma Aldrich) to a 25 mL Erlenmeyer flask without baffles containing 1.35 mg mL −1 HaloTag ® -immobilized CaKDO. The reaction was mixed by orbital shaking at 150 rpm at 20 • C. Samples were taken every hour over a period of 52 h, excluding night hours. For sample workup and analysis, see above.
Conversion of 100 mM L-lysine in 50 mL using an EasyMax 402 Thermostat system (Mettler Toledo): The reaction was started by adding 30 mL of reaction mix (3.3) to the EasyMax reaction vessel (100 mL) containing 20 mL immobilized CaKDO HaloTag ® slurry (1 mg/mL enzyme) in 200 mM HEPES, pH 7.5. The reaction was stirred at 150 rpm, 20 • C, and the pH was continuously controlled with 0.5 M NaOH, while filtered purge gas was introduced to the surface of the reaction and incorporated into the reaction mixture by stirring (the aeration rate cannot be controlled with this device). Samples were taken every hour over a period of 70 h, excluding night hours. For sample workup and analysis, see above.

Biotransformations with SrLDC
All reactions were mixed at 35 • C in an overhead shaker to guarantee sufficient mixing of the beads and the reaction mixture. Reactions were performed as technical duplicates (same enzyme batch). The 10 µL samples were diluted 1:50 with 50 mM HEPES buffer, and the reaction was stopped by incubation at 80 • C for 5 min in a thermo shaker (Eppendorf). Substrate and product concentrations were determined by HPLC (3.6).

Analytical Scale
The reaction was started by adding 1 mL reaction mix containing 100 mM L-lysine, 1 mM PLP, and 100 mM HEPES buffer, pH 7.0, to a 2 mL reaction tube containing either the lyophilized free enzyme or immobilized enzyme. In the case of experiments with (3S)-hydroxy-L-lysine, the reaction supernatant from the previous KDO reaction was taken, and 1 mM PLP was added and titrated to pH 7.0.

Repetitive Batch Experiments in a Preparative Lab Scale
For repetitive experiments, 15 mL of the reaction mixture (100 mM L-lysine, 1 mM PLP in 100 mM HEPES, pH 7.0) was added to the 1 mg mL −1 immobilized enzyme. The reaction was performed in 50 mL falcon tubes. Samples were taken every 10 min over a period of 1 h. Afterwards, the tubes were centrifuged, washed 2× with 100 mM HEPES buffer, pH 7.0, and used for the next batch. The procedure was repeated for 6 batches.

HPLC Analyses
All biotransformations were monitored by HPLC using a diode array detector (DAD) or a fluorescence detector (FLD), with the DAD detector giving the best results. For the analysis of amino acid derivatives, diamines, and (2S)-hydroxy-cadaverine, a pre-column derivatization step with ortho-phthaldialdehyde (OPA, Sigma-Aldrich) was performed. Approximate retention times were 5.6 min for L-histidine (internal standard), 8.8 min for 5-hydroxy-(D,L)-lysine, 8.9 min for (4S)-hydroxy-L-lysine, 9.0 min for (3S)-hydroxy-L-lysine, 9.1 min for L-lysine, 9.4 min for (2S)-hydroxy-cadaverine, and 10.0 min for cadaverine. Concentrations were derived from the linear calibration of five reference solutions (0.1-2.5 mM) containing L-histidine, 5-hydroxy-(D,L)-lysine, L-lysine, and cadaverine. Calibration was performed at least once per week or prior to every HPLC run. For details and chromatograms, see Supplementary Materials, Sections S9 and S12.

GC-ToF-MS Analysis
Components of the reaction mixture and mass information to identify the different hydroxy-L-lysines and the (2S)-hydroxy-cadaverine were analyzed by GC-ToF-MS according to a previously described protocol [62]. For details, see Supplementary Materials, Section S11.

NMR Analysis
After product purification (3.5) of the respective hydroxy-L-lysines and the (2S)-hydroxy-cadaverine, the 1D and 2D NMR spectra were recorded. For both components, the NMR signals were successfully assigned to the molecular structure (Supplementary Materials, Section S12). While no major impurities were visible in the (2S)-cadaverine spectrum (Supplementary Materials, Figure S42A,B), some impurities remained in the hydroxyl-lysine samples, probably due to the high concentration of HEPES buffer present in the reaction supernatant ( Supplementary Materials, Figures S27A,B and S32A,B). For both hydroxyl-L-lysine derivatives, the position of the hydroxyl group was assigned indirectly through the CH-group, as hydroxyl groups are not visible in the NMR spectrum in deuterated water (Supplementary Materials, Figures S27A,B and S32A,B).

Conclusions
Here, we demonstrate that covalent in situ immobilization is an appropriate tool to access the preparative potential of KDOs. Immobilization via the HaloTag ® solved almost all the problems that hamper the application of KDOs besides the analytical scale. The one-step immobilization rapidly concentrated the enzyme from cell-free extracts on the carrier with high residual activity and improved stability, specifically in the case of CaKDO, which showed the lowest stability among the tested KDOs. Upon KDO immobilization, the increase in the stability enabled a substrate conversion of >200 mM L-lysine, without the generation of any side products. Further, enzyme recycling was demonstrated, which was simple for immobilized CpKDO-HaloTag ® and FjKDO-HaloTag ® , but required treatment with dithionite and EDTA in the case of CaKDO. We were able to apply the immobilized CaKDO-HaloTag ® and FjKDO-HaloTag ® in a preparative lab scale (15 mL) and could show that a further increase in scale (up to 50 mL) or substrate concentration (200 mM L-lysine) was in general possible.
This generally led to a decrease in process costs and an increase in process sustainability, meeting the requirements of processes that will become increasingly important within the next few years.
Especially in the cascade reactions of KDOs and LDCs towards hydroxy-cadaverine derivatives, the immobilization approach seems to be superior to systems using in vivo two-phase fermentation approaches [21]. In the case of cascade reactions where the second enzyme has a higher activity towards the substrate of the first reaction (L-lysine) than the intermediate (hydroxyl-L-lysine), full conversion in the KDO-catalyzed step is mandatory, before the LDC comes into play to avoid the loss of L-lysine by the production of cadaverine as a main side product and related purification problems. Using immobilized enzymes allows for an easy separation of the enzyme in a simple sequential reaction setup, where the reaction parameters of the different reaction steps can easily be adjusted to the respective optimal parameters (temperature, pH, reactor design, aeration, and mixing of immobilized enzymes) and successful product purification.
We propose that covalent in situ immobilization is an appropriate tool to access the preparative potential of many other KDOs.