Lukas
Schober
,
Philippe
Dreier
and
Theo
Peschke
*
Chemical and Analytical Development, Novartis Pharma AG, 4056-CH Basel, Switzerland. E-mail: theo.peschke@novartis.com
First published on 22nd February 2024
Despite known advantages of immobilized enzymes, their broad application in pharmaceutical drug manufacturing has not yet caught up. Limited access to commercially available immobilized enzymes on one side and time – and cost – consuming development of new processes on the other side have hampered their broader implementation so far. Herein, we present a walk-away high throughput workflow with state of the art robotic equipment that enables the rapid optimization and development of immobilized enzymes – ready to be scaled - and thus matching the need for speed in modern drug development. Following this workflow, we were able to identify an engineered ketoreductase which was >2-fold more active than the wildtype from L. kefir. Further optimization of the reaction and immobilization conditions lead to an additional 29-fold productivity increase from 0.36 to 10.6 gproduct genzyme−1 for the selected enzyme variant.
We therefore developed an automated μL-scale workflow allowing parallel processing of several experiments in a 96-well plate. For maximum compatibility and transferability to other lab activities, such as enzyme screening or protein engineering, a standard Biomek i7 liquid handling robot was used. For an automated protein content measurement and to calculate the immobilization yield, the Biomek i7 workstation was connected to a sealing device and a UV/Vis photometer via an integrated robotic arm (Fig. 2B).
Essentially, this automated workflow is conducting the same steps as if the immobilization procedure would be done by manual pipetting. The workflow is divided into 4 fully automated processing steps and 3 incubation steps (see Fig. 2C). A full integration of the complete workflow without any interruption was not envisioned to allow alternative usage of the liquid handlers during the incubation steps and to maximize flexibility in the workflows. Detailed experimental procedures of all the processing steps can be found in the ESI† chapter 3.2. to 3.5. including deck layouts for the developed methods. In brief, for NH2– or metal-chelating iminodiacetic acid (IDA)-functionalized resins a pre-activation step via cross-linking with glutaraldehyde or metal-loading was needed to be done in advance prior to the immobilization step (see Fig. 2C, step A0). As some pre-loaded metal-, epoxy-functionalized or hydrophobic porous particles do not need such a pre-activation step, an alternative processing step called B1 was programmed starting similar to the A0 step with an automated resin distribution and immediately continue with the washing and lysate addition. The automated resin distribution step avoids the time-consuming and error-prone mg-scale powder handling of the resins. Instead, resins were resuspended in one reservoir with water and then 25 mg were automatically transferred with the liquid handler to the individual wells (see Fig. 2C, step A0/B1). This approach was validated for amounts between 2.5 and 35 mg resin per well with excellent linearity across the whole range (see ESI†). Subsequently, the particles get automatically washed for 4-times. The washing steps are utilizing the fact that the particles quickly settle on the bottom of the well allowing them to be separated from the supernatant before the latter is removed during each washing cycle. Typically, particle diameter of <200 μm are suitable and the particles should settle at the bottom of the wells within 30 seconds. Any potential settling issue could be addressed by implementing centrifugation steps in between the washing steps, which was not necessary in our case. After the washing of the particles, fresh cell free lysate or resuspended enzyme powder was added for the immobilization step and incubated for 18 h at 25 °C in a shaker (see Fig. 2B, step A1/B1). To be able to calculate the recovered activity of each individual enzyme resin combination afterwards, the same enzyme solution was also transferred into a benchmark plate, which was then stored until the activity test at 4 °C. Prior to each incubation step, the plates were automatically sealed (see Fig. 2A). In the follow up step 2, the unbound enzyme was removed, and substrate mixture was added before the incubation step. The last step 3 was the automatic quench of the reaction and preparation of an analytical dilution plate which was then analysed by supercritical fluid chromatography coupled with UV/vis detection (SFC-UV).
To challenge our workflow, the described procedures were tested to identify the best enzyme candidate for the enantioselective reduction of acetophenone (1) towards (S)- or (R)-1-phenylethanol (S)-2 and (R)-2. In total, 283 unique KRED mutants from 4 Codexis KRED panel plates were screened in stage 1. The KREDs of Plate 1, 2 and 4 originated from Lactobacillus kefir,16,17 and therefore needed MgSO4 supplementation during induction, while KRED Plate 3 were supplemented with ZnSO4.18 Following our workflow, we evaluated all 283 KREDs for their conversion, their selectivity and their recovered activity after cross-linking them on an amino-functionalized carrier. All KREDs were tested as free and as immobilized enzymes in the same reaction setup to minimize experimental fluctuation. As shown in Fig. 3, KREDs with conversions up to 80%, high selectivity ranges from highly (S)- to (R)-selective (>99% ee) and a recovered activity from 0–120% could be identified. We observed two sub-populations of enzymes in the Codexis KRED panels unrelated to their plate origin: One showing excellent (R)-selectivity but low recovered activity (see yellow dots in Fig. 3) and a second one with higher recovered activity but mostly unselective or (S)-selective product formation. We adjusted the lysate concentration to 4% and achieved good differentiation amongst the mutants in terms of conversion. Subsequently, we also screened all plates at a 10-fold higher enzyme concentration of 40% lysate to ensure that the KREDs are capable to high product-related conversion >90% (see Fig. S1 and Table S5†). A list of the conversion, selectivity and recovered activity of all enzymes at 4% and 40% lysate, including those of the Codexis screening kits, can be found in the Table S7.† For further investigation we selected five top performing KREDs with excellent (R)-selectivity, in the following called KRED 1–5, and three non-selective variants with a high recovered activity called KRED 6–8. Notably, the best candidates already showed a >2-fold improvement in conversion compared to the wildtype enzyme (Fig. 3).
For the second stage of our workflow, the 8 KREDs were produced as enzyme shake flask powder (SFP) to evaluate their performance depending on the carrier type. As none of the KREDs has a His-Tag, we omitted testing any kind of metal loaded particles for affinity immobilization. However, given suitable particle properties, the workflow is applicable to other immobilization strategies such as adsorption on ionic or hydrophobic carriers or affinity immobilization for His-tagged enzymes (see Fig. 2A). A list of the tested resins in this work with additional information on supplier, carrier type, size, porosity, and linker-length are summarized in Table S1.† We focused our selection on 8 different amino- and 6 epoxy-functionalized carriers which also differ in porosity, size, and linker-length. The irreversible covalent enzyme binding of these carriers is favoured from a regulatory standpoint compared to non-covalent methodologies as this ensures the prevention of any potential carry over to the product.
As shown in Fig. 4, rather independent from the carrier type, the highest conversion was detected among KRED 1–5, while the highest recovered activities amongst KRED 6–8. All KREDs showed a higher activity when immobilized on an amino carrier compared to an epoxy functionalized one (see also Fig. S5†). Amongst the tested resins, short chain linkers (ECR83XXF line) performed better than the corresponding long chain linkers (ECR84XXF line and HAXXX/S line). Regarding the porosity – higher conversions were obtained for higher porosities. However, due to their lower physical stability, the higher porosity of the particles also limits their applications to packed bed reactions in continuous flow. To remain flexible in terms of the applied process technology and as batch process development is still the more straightforward manufacturing option, we decided to continue with KRED 2 on process stable ECR8304F resins. Unfortunately, we have used this resin type already for the step 1 enzyme screening based on our pre-knowledge on L. kefir KREDs, thus the stage 2 resin screening did not lead to any changes and therefore improvement of our biocatalytic process. In the future, we would therefore skip this development step for highly related KRED variants. However, it should be kept for novel enzymes or KREDs with a lower sequence similarity or in case more diverse approaches, such as affinity immobilization, are added.
Fig. 4 Stage 2 resin screening – what resin and which corresponding immobilization approach is well suited for the KREDs showing the highest reduction of acetophenone (1) (A) Schematic flow chart of the different automated steps conducted. (B) Conversion and (C) recovered activity of immobilized KRED candidates 1–8 on amino-functionalized carriers for the enantioselective reduction. Recovered activity is calculated by comparison of product-related conversion of immobilized enzyme vs. free enzyme and does not take immobilization efficiency into account. KRED 1–5 are highly R-selective, while KRED 6–8 are non-selective (see also Fig. 3). The identified best KRED resin combination for a flexible process development is highlighted with a green dotted circle around the well. |
The third stage of the workflow was used to optimize the immobilization procedure in regard of the glutardialdehyde (GA) concentration, enzyme loading, buffer types and the pH. Usually, several of these parameters are interdependent and identifying the best combination can be time consuming. To address this challenge, we used a 2D parameter screening approach. Thereby 2D gradients were generated across the 96-well plates by serial dilutions of the stock solutions (see ESI for detailed methodology). In the first set of experiments, the buffer system was fixed to 50 mM TEoA, pH 7.5 whereas the GA gradient was varied along the x-axis from 0.0063 and 0.8 g gresin−1 and the enzyme gradient along the y-axis from 0 to 0.4 gSFP gresin−1. Despite no further conversion increase was observed above 0.05 gSFP gresin−1, the enzyme loading sweet spots could be narrowed to a concentration ≥0.023 gSFP gresin−1 and around ∼0.0125 gGA gresin−1 (see Fig. 4A, conditions based on the recommendations of the supplier are highlighted with a red dotted line around the well; the optimized with a green one). To circumvent the stalling of the reaction, the plates were re-screened with a shorter reaction time of 20 minutes instead of 2 h and in a narrower enzyme loading range (see Fig. 5B). The experiment confirmed that a further increase of the enzyme loading above 0.053 gSFP gresin−1 does not further improve the conversion. In a third experiment, different buffer systems and pH values had been screened at varying enzyme concentration (see Fig. 3C). Despite only minor differences amongst them, the enzyme immobilization buffer 50 mM Tris HCl pH 7.5 performed best and was set as the new standard.
As costs are an important factor in the decision making to find the ideal process condition, we calculated the predicted costs for each condition based on the material costs of the consumables (2D-plot shown in Fig. S7†). While this purely cost driven view of one step is useful to take a more informed decision, it is also simplifying the real cost as low conversion values normally go along with a more tedious and costly workup of the reaction mixture. Therefore, the chosen conditions are a compromise of all these aspects. The new conditions allowed significantly lower material costs due to an 8-fold reduction of glutaraldehyde and >2.5-fold regarding the KRED while keeping high product related conversion.
In the second part of the condition screening, the improved immobilization conditions were used to produce immobilized KRED 2 which was then tested under different reaction conditions. Despite mimicking substrate feeding or in situ co-product removal on small scale is unrealistic, parameters, like varying (co)-substrate or cofactor concentrations, buffer-pH or types are well suited and due to the high throughput even preferentially tested in small scale. In a first experimental setup, we varied two highly interdependent parameters, the substrate concentration of acetophenone (1) and the co-substrate/co-solvent isopropanol. Making an educated prediction of the optimal concentration is not trivial. High concentration of isopropanol will push the reaction towards the desired product but also potentially decrease the enzyme stability and thus influence the overall productivity of the process. As the substrate had to be dissolved in isopropanol as well, the isopropanol gradient was diagonal over the plate from A1–H12 (see Fig. 6A). As presented in Fig. 6B and C the immobilized KRED 2 showed the highest product titer with a decent conversion (49.4%) after only 2 h at a loading level of 150 g L−1 acetophenone and 55 vol% (7.1 eq.) isopropanol.
The cofactor NADPH can be a significant cost contributor to a biocatalytic step. Therefore, stage 3 of the workflow was also used to investigate whether the necessary amount of cofactor in correlation with the corresponding buffer system can be further decreased (see results in Fig. S6†). The investigation revealed that the addition of NADP to the reaction buffer did not lead to higher conversion values. This indicates not only that sufficient cofactor being present in the crude shake flask powder but also that it remains firmly bound to the enzyme during immobilization and subsequent washing. The buffer screening of triethanolamine (TEoA), potassium phosphate (KPi), 3-(N-morpholino)propanesulfonic acid (MOPS), tris(hydroxymethyl)aminomethane (Tris) buffers at pHs ranging from 7–9 resulted in a similar productivity for all tested buffer systems (see Fig. S5†). To simplify the process control and sourcing, it was decided to use the same buffer system as in the immobilization step, Tris HCl pH 7.5.
In the final stage 4, we investigated whether the findings based on our small-scale high throughput method are scalable. Therefore, we scaled up the plate-based immobilization procedure 80-fold to a 2 g-scale using Solid phase extraction (SPE) tubes. Two different batches of immobilized KRED2 were prepared, one following typical literature conditions,12 the other our optimized immobilization conditions based on stage 3A (see Table 1). Both immobilization batches were then tested monitoring the conversion for 24 hours under typical reaction conditions based on literature12 as well as with our optimized reaction conditions identified in step 3B. Indeed, the findings from stage 3A were confirmed, and very similar conversion were observed for both conditions despite the optimized immobilization conditions are applying 3 times less enzyme and 6 times less glutaraldehyde (ESI† Table S12). Comparing the literature conditions for immobilization and the reaction conditions with those identified in stage 3A and 3B (higher substrate loading, no cofactor supplementation and higher isopropanol concentration; see Table 1) most key performance indicator (KPI) in process development such as catalyst productivity, product titer, costs and sustainability improved (see Table 1). The conversion was the only KPI that deteriorated, which however is a known phenomenon at increased substrate loadings. The inherent equilibrium problem associated with the coupled-substrate approach can be simply overcome by acetone removal with a lower atmospheric pressure in the reaction vessel.19–21 These results clearly show that our workflow allows the rapid optimization of scalable biocatalytic processes comprising immobilized enzymes.
Parameter | Literature conditions12 | Optimized conditions | |
---|---|---|---|
a The inherent equilibrium problem associated with the coupled-substrate approach can be simply overcome by acetone removal with a lower atmospheric pressure in the reaction vessel.19–21 b A more detailed time dependent plot of the catalyst productivity can be seen in Fig. S8.† c Absolute cost calculation was based on prices from October 2022 and heavily depend on market situation and scale. The numbers should be only used for relative orientation. Further details for the underlying calculation can be found in Table S13.† d Process mass intensity (PMI) is a mass-based metric to evaluate the green credentials of an individual reaction. The PMI was calculated only for the biocatalytic reaction without considering any downstream processing/isolation. | |||
Conditions | Immobilization buffer | 75 mM TEoA pH 7.0 | 75 mM Tris HCl pH 7.5 |
Gluteraldehyde loading | 0.08 g gResin−1 | 0.0125 g gResin−1 | |
Enzyme loading | 0.156 g gResin−1 | 0.053 g gResin−1 | |
Acetophenone concentration | 12 g L−1 (0.1 M) | 150 g L−1 (1.25 M) | |
NADP concentration | 0.5 g L−1 | 0 g L−1 | |
Isopropanol concentration | 20 vol% | 55 vol% | |
Reaction buffer | 75 mM TEoA pH 7.0 | 75 mM Tris HCl pH 7.5 | |
Key Perfomance indicator (KPI) | Conversion | 92% | 73%a |
ee | >99% | >99% | |
Catalyst productivityb | 0.35 gproduct genzyme−1 | 10.2 gproduct genzyme−1 (29-fold increase) | |
Product titer | 11 g L−1 | 111 g L−1 (10-fold increase) | |
Reagent costsc | High | Low (19-fold decrease) | |
Sustainability based on PMId | 102 kg kg−1 | 9.7 kg kg−1 (11-fold decrease) |
Footnote |
† Electronic supplementary information (ESI) available. See DOI: https://doi.org/10.1039/d3re00704a |
This journal is © The Royal Society of Chemistry 2024 |