Accessible biocatalyst development by rapid in vitro semi-rational engineering (RISE) of enzymes

Summary

Tailoring natural enzymes to synthetic needs is often associated with high costs and long timelines, hindering the broader adoption of biocatalysis in the chemical and pharmaceutical industries. To address this, we developed the RISE (rapid in vitro semi-rational engineering) workflow that makes enzyme engineering accessible to chemistry laboratories. RISE integrates four key concepts: computational design of focused variant libraries, rapid generation of linear mutant DNA libraries via PCR, cell-free protein synthesis from linear template DNA, and iterative cycles of mutagenesis, expression, and testing to accumulate beneficial mutations. In a proof-of-concept study, we engineered a ketimine reductase from Rattus norvegicus (RnKIRED), achieving stereoselectivity inversion in one reductive amination reaction and a 400-fold activity improvement in another. These engineered variants enabled the gram-scale synthesis of key intermediates for ACE2 inhibitor drugs. RISE bridges the gap between inefficient wild-type enzymes and expensive directed evolution, promoting biocatalysis implementation in early chemical development.

Graphical abstract

Highlights

• •Innovative cell-free DNA manipulation enables enzyme engineering in chemistry labs
• •Fast and accessible biocatalyst development is demonstrated on a ketimine reductase
• •Focused engineering results in activity enhancement and stereoselectivity switch
• •Engineered variants are used in gram-scale synthesis of drug intermediates

Biotechnology; Enzyme engineering; Biocatalysis

Introduction

Biocatalysis is often pictured as a mature technology that has reached industrial viability through advances in molecular biology techniques, bioinformatic tools, and directed protein evolution.^1,2,3 This is well exemplified by numerous industrial enzymatic processes reported in the literature or filed as patents.^4,5 However, developing an industrial biocatalyst still requires significant time and financial investment, which limits the adoption of biocatalysis by a broader chemistry community.^6,7 As an example, in the early chemical development of active pharmaceutical ingredients a biocatalytic route can prove inferior to chemical methods due to the low activity or insufficient stability of the available wild-type enzymes.^8,9 Directed evolution could be used to improve those properties, but its high cost and time demands are often not justified at that stage causing the enzymatic solution to be discarded. As an alternative, focused rational or semi-rational engineering of enzymes can also provide improved variants, but the laborious DNA manipulation and protein expression involved in this process are often still prohibitive for non-specialized laboratories. To make this option more accessible, we developed RISE (rapid in vitro semi-rational engineering), a simple workflow enabling quick development of biocatalysts.

Central to RISE is cell-free protein synthesis (CFPS) that simplifies the expression of enzyme variants by avoiding DNA transformation, culture growth, and cell lysis to obtain the expressed protein.^10,11 E. coli extract based in vitro transcription-translation systems are available from multiple vendors or can be produced in-house as well.¹² Such systems are increasingly used in ultrahigh-throughput microfluidic screenings^13,14 but there are also several reports of medium-throughput microtiter plate-based applications.^15,16 A major advantage of CFPS is its ability to express proteins from linear template DNA. This opens the possibility to directly use PCR products for expression instead of plasmids that are much more laborious to prepare.

Semi-rational enzyme engineering^17,18 involves the creation of focused mutant libraries of the parent enzyme by site-directed mutagenesis (SDM). Well-established PCR-based methods like QuikChange, Gibson-assembly, or Q5 SDM are used to create plasmids containing the prespecified mutation. To date, there have only been scattered reports of methods creating linear template DNA for single-point mutant proteins. RAPPER¹⁹ involves a primer overlap extension PCR strategy while DiRect²⁰ uses both overlap extension and subsequent nested PCR reactions. Both methods suffer from contamination from the original template DNA, which is mitigated by either the use of biotinylated primers or multiple amplification steps. In RISE, we use a self-developed cell-free Q5 SDM protocol that involves a second PCR step instead of transforming the KLD (kinase, ligase, DpnI) reaction into bacteria. This step results in quick and cell-free amplification of the mutated coding sequence without contamination of the wild-type sequence which is digested by DpnI (Figure 1B). Very recently, Landwehr et al. followed a similar strategy to create a single-point mutant linear DNA library using the Gibson assembly.²¹

Figure 1.

The concept of RISE

(A) General workflow of RISE.

(B) Workflow of the cell-free Q5 SDM protocol.

A key step in every semi-rational enzyme engineering effort is the design of the mutations to test. Computational methods have proven efficient in selecting mutational hotspots for engineering.²² Several tools are available to predict activity- or stability-enhancing mutations.²³ Since epistatic effects arising from the combination of mutations are still hard to predict, often experimental strategies need to be used to accumulate beneficial mutations. In the context of low- or medium-throughput enzyme engineering, FRISM²⁴ (focused rational iterative site-specific mutagenesis) has been a widely used strategy. In FRISM, a reduced alphabet of amino acids is introduced in selected positions and the best-performing variant is used as a starting point for iteration.

Putting these considerations together, we have defined the workflow of RISE (Figure 1A) as follows. First, a focused mutant library is designed using computational tools. Then, a linear DNA library encoding for each designed variant is produced by a cell-free Q5 SDM protocol and the variants are expressed by CFPS. The activity of variants is tested in small-scale biocatalytic reactions and the best-performing variants are selected and used as templates for an iterative mutagenesis, expression, and testing cycle.

Results

Validation of the cell-free Q5 SDM protocol

In a proof-of-concept study, first, the cell-free Q5 SDM protocol (Figure 1B) had to be validated since all other steps are well-documented in the literature. We have selected two unrelated enzymes to perform the validation: a ketimine reductase from Rattus norvegicus (RnKIRED)^25,26 and a 2′-deoxyribosyltransferase from Lactobacillus leichmannii (LlNDT).^27,28 HotSpot Wizard²⁹ webserver and CAVER analyst^30,31 were used on snapshots from molecular dynamics simulations to select positions for mutagenesis based on their predicted mutability and location within the active site and access tunnels. The process identified ten positions in RnKIRED and 15 in LlNDT that were selected for mutation. Six amino acid changes were proposed in each, considering their size, predicted stability, and chemical diversity (Figures S1 and S2; Tables S1–S4). This smart library design resulted in a 60-member variant library for RnKIRED and a 90-member variant library for LlNDT. The primers for each mutation have been designed with the widely used NEBaseChanger and NEB T_m calculator tools to allow the use of a single annealing temperature so that the mutagenic PCR can be performed in a single experiment (Table S5). For the second PCR reaction, a general primer pair was designed 100 base-pairs upstream and downstream of the T7 promoter and T7 terminator regions on the plasmid backbone. Then, we could perform the whole cell-free Q5 SDM protocol (Figure 1B) and obtain the linear mutant DNA libraries for both RnKIRED and LlNDT (Figures S3 and S5). Notably, the library preparation process could be completed within 5–8 h. The presence of the designed mutations has been confirmed by sequencing. This demonstrates that our cell-free Q5 SDM protocol can be used to rapidly generate linear single-point mutant DNA libraries.

Validation of the RISE workflow

Next, all subsequent steps of RISE have also been performed on these two libraries to complete the validation of all experimental protocols. Using the commercial NEBExpress kit, the expression of all 60 RnKIRED and 90 LlNDT variants has been achieved (Figures S4 and S6).

To assess the activity of the produced variants, we have selected test reactions where the activity or selectivity of RnKIRED and LlNDT has to be improved to reach synthetic applicability. RnKIRED shows high activity and R-selectivity in the reaction of l-homophenylalanine ethyl ester (1) and pyruvic acid (2) (Reaction 1, Scheme 1), and S-selectivity but low activity in the reaction of l-alanine tert-butyl ester (4) and benzylpyruvic acid (5) (Reaction 2, Scheme 1).²⁶ Since (2S,2′S)-products are key building blocks of several ACE2 inhibitor drugs, both reactions were selected, targeting modified stereoselectivity in the former and improved activity in the latter. For LlNDT, 2′-modified nucleoside analogs are pharmaceutically relevant synthetic targets since they are important building blocks of therapeutic oligonucleotides.³² The wild-type LlNDT has shown only modest activity on 2′-fluoro derivatives,²⁸ thus, we selected the sugar transfer reaction from 2′-fluoro-cytidine (7) to 2-chloro-adenine (8) (Reaction 3, Scheme 1), targeting improvement in conversion to nucleoside product 9.

Scheme 1.

Test reactions and conditions to optimize by enzyme engineering with RISE

In Reaction 1, we have seen significant increase in (2S,2′S)-product formation from <1% to 15% for one of 60 single-point variants, RnKIRED S228G (Figure S7A). In Reaction 2, we have observed a detectable conversion to the product for five of 60 single-point variants with RnKIRED S228G standing out reaching 37% conversion to 6 in two days (Figure S8A). In Reaction 3, only marginally increased conversion was observed for the best four of 90 single-point variants LlNDT F8A, A10G, Y21R, and V41I (Figure S9). The findings in each case have been validated by performing the reactions with bacterially produced and purified enzymes (Figures 2 and S10). These results confirm that the experimental protocols of RISE are suitable for the rapid production of focused enzyme variant libraries in appropriate amounts for activity screening.

Figure 2.

Validation of results with purified variants from RISE

(A) Conversion and product diastereomer distribution achieved with the top hits from four rounds of RISE. Conditions, 25 mM 1, 75 mM 2, 0.4 mM NADP⁺, 62.5 mM glucose, 6 U/mL GDH, 100 mM Na-phosphate, pH = 7.5, 20 v/v% DMF, 0.1 mg/mL enzyme.

(B) Conversions achieved with the top hits from four rounds of RISE. Conditions, 100 mM 4, 50 mM 5, 0.4 mM NADP⁺, 125 mM glucose, 6 U/mL GDH, 100 mM Na-phosphate, pH = 7.5, 20 v/v% DMF, 0.1 mg/mL enzyme.

Engineering RnKIRED with RISE

The contrast between improvements observed in the RnKIRED and LlNDT libraries prompted us to perform the iterative cycles of RISE only on the former. Using the validated protocols, we combined further mutations on RnKIRED S228G to further modify stereoselectivity in Reaction 1 and improve activity in Reaction 2. A second mutation has been introduced in eight positions (six amino acids in each), resulting in a 48-member double mutant library.

Testing this library in Reaction 1 has revealed 13 of 48 mutations increasing the ratio of the (2S,2′S)-3, but RnKIRED H91I/S228G stood out with inverted stereoselectivity of 35% de (2S,2′S)-3 (Figure S7B). Notably, the activity of both RnKIRED S228G and RnKIRED H91I/S228G is significantly decreased in Reaction 1 compared to the wild-type RnKIRED. Next, we produced a 42-membered triple-mutant library with RnKIRED H91I/S228G as the template. From this, we identified two of 42 improved variants, among which RnKIRED V49I/H91I/S228G was superior, displaying further increased stereoselectivity of 80% de (2S,2′S)-3 while also improving activity in Reaction 1 (Figure S7C). In a fourth round of mutagenesis, a 36-membered quadruple-mutant library was produced based on RnKIRED V49I/H91I/S228G. From that, no variant has shown further improvement in stereoselectivity, the activity could be further improved only at the expense of decreased selectivity toward (2S,2′S)-3 (Figure S7D).

In Reaction 2, we found 14 of 48 activity increasing mutations in the initial double mutant library, the two best double mutants RnKIRED V77G/S228G and RnKIRED H91A/S228G displayed over 90% conversion to 6 in one day (Figure S8B). Therefore, we produced a 78-membered triple-mutant library using both RnKIRED V77G/S228G and RnKIRED H91A/S228G as templates. The high activity of the double mutants in reaction 2 allowed us to lower the excess of 4 used in the screening from 50 to 2 equivalents while increasing the substrate loading as well. The ∼20% conversion achieved by the template double mutants under these conditions enabled capturing further increase of activity. With that we have detected 12 of 78 activity-improving mutations and identified RnKIRED V77G/H91A/S228G and RnKIRED V77L/H91A/S228G as top performers showing over 30% conversion to 6 (Figure S8C). In the fourth round, a 60-membered quadruple-mutant library was produced based on both RnKIRED V77G/H91A/S228G and RnKIRED V77L/H91A/S228G. From that, we have identified 3 of 60 improved variants, among which RnKIRED F58Y/V77L/H91A/S228G achieved close to 50% conversion to 6 (Figure S8D).

Synthetic application of RnKIRED variants

While performing the iterative mutagenesis rounds, we have also started the synthetic application of the improved variants. In Reaction 1, RnKIRED V49I/H91I/S228G enabled the synthesis of (2S,2′S)-3 with >70% conversion on preparative scale (543 mg, 39% isolated yield, >99% de). In Reaction 2, RnKIRED H91A/S228G has been used to produce 6 on gram scale with >97% conversion and 63% isolated yield (967 mg, >99% de).

Discussion

Taking these results together, we have validated the whole workflow of RISE by performing consecutive rounds of semi-rational engineering on RnKIRED with two separate aims. When we aimed at changing stereoselectivity, we achieved inversion of >99% R-selectivity to 80% S-selectivity by testing 186 variants in four rounds. The variant RnKIRED V49I/H91I/S228G has also sufficient activity to reach synthetic applicability (Figure 2A). When we aimed at increasing activity, we achieved a 400-fold increase in specific activity (Figure S11) and high conversions with low aminoester excess by testing 258 variants in four rounds (Figure 2B). The variant RnKIRED H91A/S228G reached >97% conversion in a gram-scale transformation.

The effect of mutations on the stereoselectivity and activity of RnKIRED cannot be easily rationalized by structural or mechanistic intuitions. This might be subject to further studies to inform future evolution campaigns. Most importantly, S228G appeared to be a key mutation modifying stereoselectivity and decreasing activity in Reaction 1 while improving activity without changing stereoselectivity in Reaction 2. In the second round, we have observed strong epistatic effects with position 91 in both reactions.

Our results demonstrate that RISE has the potential to enhance the adoption of biocatalysis in a wider chemistry community by lowering both technical and financial barriers for enzyme engineering. On the technical side, the cell-free nature of the workflow comes with two distinct advantages: it significantly reduces the time required for DNA manipulation (Figure 3A) and it also makes RISE accessible to non-specialized laboratories lacking cell manipulation capabilities. It requires only a PCR machine and basic equipment for DNA and protein analysis, while all reagents used are commercially available. Moreover, the cell-free Q5 SDM protocol also enables direct coupling with large-scale expression of top performing variants by transforming KLD reactions into bacteria. On the financial side, in-house production by cell-free Q5 SDM reduces the cost of focused mutant DNA libraries compared to commercial sourcing (Figure 3B). Furthermore, for chemistry laboratories relying on external partners for enzyme production, expression of single hits instead of whole libraries also results in cost reduction.

Figure 3.

Advantages of RISE

(A) Comparison of timelines for DNA manipulation by the standard and cell-free Q5 SDM (for supporting data see Table S6).

(B) Comparison of the cost of a 60-member single-point mutant DNA library obtained from commercial sourcing and in-house cell-free Q5 SDM (for supporting data see Table S6).

(C) Proposed role of focused enzyme development by RISE in chemical development in the pharmaceutical industry.

The significance of RISE is also underlined by a recent pioneering work in the field of cell-free enzyme engineering that was developed in parallel to our efforts. The paper of Landwehr et al. describes a workflow similar to our cell-free Q5 SDM for the production of linear mutant DNA libraries. In contrast to RISE, they use automation to generate large libraries to inform machine learning algorithms.²¹ Our approach aims to provide an enzyme engineering tool for medium and small-scale experiments accessible to chemistry laboratories due to its simplicity and low technical demand.

In summary, our proof-of-concept study demonstrates that synthetically applicable enzyme variants can be obtained from only a few rounds of laboratory evolution by RISE starting from an unfit wild-type enzyme (unsuitable stereoselectivity or low activity). While such a focused engineering campaign might not provide an industrial biocatalyst, the improved variants can help bridge the gap between the use of inefficient wild-type enzymes and the ultimate biocatalyst obtained by full-scale directed evolution (Figure 3C). The simplicity of the protocol and the short turnaround time enable the demonstration of the synthetic utility of an enzyme in early phases of chemical development helping enzymatic processes to progress further toward greener manufacturing routes.

Limitations of the study

Notably, there are some limitations of RISE. Computational design of mutations requires detailed understanding of the structure and mechanism of the biocatalyst of interest and the focused library will ultimately constrain the achievable fitness improvement. The enzyme should be well-expressed in CFPS. which might exclude proteins having more complex folds, requiring prosthetic groups, metal ions, or cofactors for folding. The full workflow cannot be self-sufficiently performed in chemistry laboratories; bacterial expression is still required for hit validation and synthetic application. The FRISM principle applied for the accumulation of beneficial mutations introduces strict path dependency in the sequence space that might lead to evolutionary dead ends. In the future, a more advanced version of RISE might include data-driven prediction of higher order mutations by machine learning or artificial intelligence tools.

Resource availability

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Beáta G. Vértessy (vertessy.beata@ttk.hu).

Materials availability

This study did not generate new unique reagents.

Data and code availability

• •All data reported in this paper are available within the paper and the supplemental information files. Data will be shared by the lead contact upon request.
• •This paper does not report original code.
• •Any additional information required to reanalyze the data reported in this paper is available from the lead contact upon request.

Acknowledgments

Project no. C1580174 (AT) and C2266727 (VZG) have been implemented with the support provided by the Ministry of Culture and Innovation of Hungary from the National Research, Development and Innovation Fund, financed under the NVKDP-2021 and KDP-2023 funding scheme, respectively.

This work was supported by the European Commission, under Horizon Europe’s Research and Innovation Programme (Bluetools project, grant agreement no 101081957). Views and opinions expressed are however those of the authors only and do not necessarily reflect those of the European Union or European Research Executive Agency. Neither the European Union nor the granting authority can be held responsible for them.

This article is based upon work from COST Action COZYME CA21162, supported by COST (European Cooperation in Science and Technology).

BGV was supported by the National Research, Development and Innovation Fund of Hungary (K135231, K146890, NKP-2018-1.2.1-NKP-2018-00005, 2022-1.2.2-TÉT-IPARI-UZ-2022-00003), the TKP2021-EGA-02 grant, implemented with support provided by the Ministry for Innovation and Technology of Hungary from the National Research, Development and Innovation Fund, and the ICGEB Research Grants Programme 2023 (CRP/HUN23-02).

J.M. is supported by the scholarship Brno Ph.D. Talent. The project was supported by the Ministry of Education, Youth and Sports of the Czech Republic - RECETOX RI (LM2023069), INFRA CZ (90254), and Elixir CZ (LM2023055), European Union’s Horizon 2020 Research and Innovation Programme projects CETOCOEN (857560) and CLARA (1011366070), and the Grant Agency of the Czech Republic (25-18233M). The article reflects the author's view, and the Agency and European Commission are not responsible for any use that may be made of the information it contains.

The authors thank the Analytical Department of SRIMC for assisting in the characterization of the compounds.

Graphical figures were created using BioRender.com. Graphical abstract, https://BioRender.com/nrq6oq0 Figure 1A: https://BioRender.com/s75hva4 Figure 1B: https://BioRender.com/vsch3ev.

Author contributions

The manuscript was written through contributions of all authors. A.T.: conceptualization, methodology, investigation, validation, and writing – original draft. V.Z.G.: investigation and validation. G.D.: investigation. J.M.: software and formal analysis. D.B.: software and formal analysis. B.G.V.: conceptualization, methodology, and writing – review & editing. G.T.: conceptualization, investigation, writing – review & editing, and supervision.

Declaration of interests

The authors declare no competing interests.

Declaration of generative AI and AI-assisted technologies in the writing process

During the preparation of this work, the authors used Servier Secure GPT—internal service based on Microsoft’s Azure OpenAI service—in order to rewrite the summary section to fit within the required word limit. After using this tool or service, the authors reviewed and edited the content as needed and take full responsibility for the content of the publication.

STAR★Methods

Key resources table

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Bacterial and virus strains
E. coli XL1Blue	Agilent Technologies	50-125-058
E. coli BL21(DE3)	Sigma-Aldrich	69450-M
E. coli BL21(DE3)	Thermo Scientific	EC0114
Chemicals, peptides, and recombinant proteins
Chemical 1	Combi-Blocks	OR-1599
Chemical 1	Angene Chemical	AG003Q00
Chemical 2	Fluorochem	169520
Chemical 4	BLD Pharmatech	BD8618
Chemical 5	Combi-Blocks	QC-2820
Chemical 7	Angene Chemical	AG0006OP
Chemical 8	Combi-Blocks	QK-5329
RnKIRED and variants (Table S2)	This study	N/A
LlNDT and variants (Table S4)	This study	N/A
Critical commercial assays
NEBExpress® Cell-free E. coli Protein Synthesis System	NEB	E5360L
Oligonucleotides
Primers (Table S5)	Microsynth	N/A
Recombinant DNA
pET14b-RnKIRED	Genscript	N/A
pET14b-RnKIRED variants (Table S2)	This study	N/A
pET14b-LlNDT	Genscript	N/A
pET14b-LlNDT variants (Table S4)	This study	N/A
Software and algorithms
CAVER Analyst	Loschmidt Laboratories	https://caver.cz/index.php
Hotspot Wizard	Loschmidt Laboratories	https://loschmidt.chemi.muni.cz/hotspotwizard/
Amber	N/A	https://ambermd.org/index.php

Experimental model and study participant details

Escherichia coli XL1Blue strain was used to prepare the plasmids. The cells were incubated in Luria-Bertani (LB) medium containing 50-50 mg/mL tetracycline and carbenicillin at 37°C for 16 h. Escherichia coli BL21(DE3) strain was used to express the target proteins. The cells were grown in Luria-Bertani (LB) medium supplemented with carbenicillin (50 mg/mL). The culture was grown to OD = 0.6 and induced by addition of 0.5 mM IPTG and then incubated in at 16°C for 16 h.

Method details

Computational analysis and library design

Acquisition and preparation of structure: RnKIRED (UniProt ID: Q9QYU4) homology model structure was generated by the SWISS-MODEL³³ server using default settings. The template for modeling was chain A of the crystal structure of the ketimine reductase from Mus musculus (PDB: 4BVA)³⁴, having 97.5% sequence identity with RnKIRED and 1.75 Å resolution. Titratable residues were protonated, and orientations of Asn, Gln, and His side chains were optimized using the H⁺⁺³⁵ server at 0.15 M salinity and pH 7.5, with the rest of the parameters left at default values. After processing with H⁺⁺, coordinated crystallographic waters were added from the structure of 4BVA.

Parametrization: Input topologies and coordinates were prepared using the Tleap module of AMBER 14. The protein atoms of the system were parametrized using the ff14SB³⁶ force field. The NADPH cofactor was parametrized using the GAFF³⁷ force field with partial charges and force field parameter modifications taken from the AMBER parameter database.³⁸ The system was solvated in a truncated octahedron of water molecules so that all protein atoms were at least 9 Å from the water box’s surface, so the system is well contained within the periodic box. The TIP3P³⁹ water model was used. The charge of the system was neutralized using 4 Na⁺ ions. The number of ions added for the production simulations was determined using the average volume in the last stage of the equilibration simulation to achieve a final salinity of 0.15 M NaCl. Lastly, masses of protein groups containing hydrogen atoms were repartitioned using the AMBER14 program ParmEd by its command HMassRepartition.⁴⁰

Molecular Dynamics Simulations: Energy minimizations and MD simulations were performed using the PMEMD.CUDA module of AMBER16. Initially, the system was minimized by 10 steps of steepest descent followed by 9990 steps of conjugate gradient with 500 kcal mol⁻¹×Å⁻² restraints on all atoms of the protein. The system was further minimized in four more rounds, each consisting of 2500 steps of steepest descent and 7500 steps of conjugate gradient minimization with decreasing harmonic restraints. The restraints were applied as follows: 500, 125, and 25 kcal mol⁻¹×Å⁻² on all backbone atoms of the protein. Finally, the system was minimized with 5000 steps of steepest descent and 15000 steps of conjugate gradient minimization without any restraints. The subsequent MD simulations employed periodic boundary conditions, the Particle Mesh Ewald method to treat electrostatic interactions with the long-range cut-off of 10 Å and the same cut-off for nonbonded interactions, the SHAKE⁴¹ algorithm to fix all bonds containing hydrogens, and a 4 fs time step. Equilibration simulations consisted of two steps: (I) 500 ps of gradual heating from 0 to 300 K using the Langevin⁴² thermostat with a collision frequency of 5.0 ps⁻¹ and constant volume, (II) 400 ps in the NPT ensemble using the Berendsen barostat⁴³ at constant pressure of 1.0 bar and pressure relaxation time of 1.0 ps⁻¹ and harmonic restraints of 150.0 kcal×mol⁻¹×Å⁻² on the positions of all protein backbone atoms. Then, the system was further equilibrated during 8800 ps at 310 K in the NPT ensemble as in step (II) in 10 rounds of 400 ps each with decreasing harmonic restraints. The restraints were applied as follows: 100.0, 75.0, 50.0, 25, 15.0, 10.0, 5.0, 1.0, 0.5, and 0.0 kcal×mol⁻¹Å⁻² on the backbone atoms of the protein. After equilibration with the final NaCl concentration, 35 ns long production MD simulations were run using the same settings as the last equilibration step. Coordinates were saved at intervals of 4 ps. The simulation was run in 4 replicas, producing 140 ns of aggregated simulation time.

Analysis of molecular dynamics simulations using CAVER Analyst 2: The production MD simulations were visualized in CAVER Analyst 2.0 beta. Tunnels were calculated from the production MD simulation trajectories. The tunnel origin was set on the center point between residues R48 and K75 Cα atoms. The calculation was run using default settings except probe_radius = 1.4 Å and exclude_end_zone = 2 Å.

Assessment using Hotspot Wizard: The structure of RnKIRED was submitted to the Hotspot Wizard web server and was processed using default settings. The conservation and correlation of the individual positions, as well as their mutational landscape and mutational effect predictions, were used to construct the mutational libraries. For RnKIRED, residues labeled as hotspots, both reliably and with unreliable mutational data, were considered for selection into the mutant library (Figure S1 and Table S1). In the case of LlNDT (UniProt ID: Q9R5V5), we focused on residues lining the ligand in the structure (PDB: 1F8X)⁴⁴ which included hotspots, tentative hotspots as well as non-hotspots. Hotspot Wizard was used to select amino acid substitutions based on stability score and chemical diversity (Figure S2 and Table S3).

Site-directed mutagenesis with the cell-free Q5 SDM protocol

Mutagenic PCR reactions were conducted on 96-well PCR plate according to the Q5 Site-Directed Mutagenesis Kit Quick Protocol provided by the manufacturer (12.5 μL reaction volume). A single annealing temperature was used (65°C for RnKIRED libraries, 64°C for the LlNDT library) for 30 cycles. KLD reactions were assembled according to the manufacturer’s protocol and incubated at room temperature for 1h. 2 μL from the KLD reactions was transferred to a new 50 μL Q5 Hot Start PCR reaction containing the primers for amplification of the coding sequence at 0.5 μM concentration. 72°C was used for both annealing and elongation (50s, 35 cycles) to minimize crossreactions with contaminating primers from the first PCR step. KLD reactions were stored at −20°C until eventual use for plasmid production. PCR products were purified using the Mag-Bind TotalPure NGS kit according to the manufacturers protocol and sent for sequencing. DNA concentration was determined by measuring absorbance at 260 nm using Thermo Scientific Multiskan SkyHigh Microplate Spectrophotometer (Thermo Fisher Scientific Inc., Waltham, MA, USA) with μDropDuo plate.

Plasmid production for hit variants

KLD reactions of selected mutants were transformed into XL1Blue chemically competent cells and were selected on LB-TC/CAR agarose plates. Single colonies were picked and used to inoculate 5 mL overnight cultures. From these, plasmids were purified according to the NucleoSpin Plasmid Prep protocol. The purified plasmids were sent to sequencing to verify mutations, and then used for bacterial protein expression.

Cell-free protein synthesis

CFPS was carried out using the NEBExpress Cell-free E. coli Protein Synthesis System according to manufacturer’s protocol (12.5–50 μL reaction volumes, 16h incubation at 29°C). The synthesized variants were eventually immobilized on NEBExpress Ni-NTA Magnetic Beads applying 10 μL slurry per reaction (see Figure S8).

Bacterial enzyme production and purification

RnKIRED and LlNDT genes were codon optimized for E. coli, synthesized by Genescript and cloned into pET14b vector using NdeI and BamHI cloning sites resulting in N-terminal His-tag proteins (see supplemental information for protein sequences).

pET14b-RnKIRED or its variants were transformed into E. coli BL21(DE3) chemically competent cells, and 5 mL overnight cultures were grown in Luria-Bertani (LB) medium supplemented with carbenicillin. Then, 4 mL was added to 500 mL fresh LB medium. The culture was grown at 37°C to OD = 0.6 and induced by addition of 0.5 mM IPTG. The flasks were then incubated in a shaker at 16°C overnight. The cells were harvested by centrifugation at 4000 rpm, 4°C for 30 min. The pellets were resuspended in lysis buffer (50 mM Tris pH = 8.5, 300 mM NaCl, 1 mM benzamidine, 2 mM phenylmethylsulfonyl fluoride (PMSF), 1 mM Tris(2-carboxyethyl)phosphine (TCEP), DNase) and lysed by ultrasonication. The suspension was centrifuged at 11000 rpm, 4°C for 30 min. The supernatant was loaded onto equilibrated Ni-NTA column, washed with 15 mM imidazole in lysis buffer. The bound protein was eluted with 250 mM imidazole in lysis buffer and then dialyzed in 50 mM Tris pH = 8.5. The resulting solution was aliquoted after addition of 10% glycerol and then stored at −20°C until further use. Total protein concentration was determined based on absorption at 280 nm measured on a NanoDrop spectrophotometer.

The transformation of pET14b-LlNDT or its variants and subsequent selection and preparation of the master cell bank (glycerol stock) was carried out according to the protocols provided by the manufacturer of E. coli BL21(DE3) competent cells (Thermo Scientific). Then, seed culture was prepared by inoculating 100 mL of LB medium containing ampicillin (100 mg L⁻¹) with 1000 mL glycerol stock of the producer strains. The flasks were incubated at 28°C, 250 rpm overnight. OD₆₀₀ reached 6.0–6.5 and the homogeneity of the culture was confirmed by microscopic examination. 20 mL of E. coli seed culture was used to inoculate 2000 mL of 2 YT medium containing ampicillin (100 mg L⁻¹) and incubated at 37°C, 250 rpm agitation until the culture reached OD₆₀₀ 0.4–0.5. The expression was induced by addition of IPTG (1 mM final concentration). The fermentation was continued for 3 h at 37°C. The biomass was harvested by centrifugation at 5500g and 4°C for 20 min and cell pellets were stored at −20°C. The cell pellet was resuspended (100 mL end volume) in chromatography load buffer (20 m M NaP_i, 500 mM NaCl, 20 mM imidazole, pH 7.4) and homogenized twice (1000 bar, 4°C–8°C), centrifuged twice (8000 g, 4°C, 10 min) and the supernatant was used for protein purification. The purification of the enzymes was carried out on an AKTA Purifier (UPC100) and a Cytiva’s HisTrap HP 5 mL column according to the manufacturer’s recommendations using load buffer for washing and elution buffer containing 20 mM NaP_i, 500 mM NaCl, 500 mM imidazole, pH: 7.4, Then the purified solutions were dialyzed in a dialysis buffer composed of 50 mM Tris×HCl, 250 mM NaCl, pH:7.0 twice for 12 h using 50× volume dialysis buffer at 4°C with gentle stirring. Then, the protein solution was concentrated using Amicon Ultra Centrifugal Filter, 10 kDa MWCO combined with further buffer exchange (4×). After the purification process, protein samples were stable in the elution buffer at 4°C but during the dialysis, significant precipitation was observed in case of all the enzymes. Precipitated protein was removed by centrifugation, but further precipitation was observed. After the second removal of the precipitated protein, the solutions seemed stable at 4°C (5 days of observation). There was at least 50–75% of protein loss was confirmed by Bradford-assay. The proteins were stored at −20°C.

RnKIRED test reactions

For activity screening of variant libraries, the reaction mixtures (100 μL) contained 0.4 mM NADP+, 25 mM 1 and 1.1–3 equivalents of 2 with 2.5 equivalents of d-glucose and 6 U/mL GDH (Codexis CDX-901, 50 U/mg) in 100 mM Na-phosphate buffer containing 20% v/v DMF, adjusted to pH 7.5 (Reaction 1) or 0.4 mM NADP+, 10–50 mM 4 and 2–50 equivalents of 3 with 2.5 equivalents of d-glucose and 6 U/mL GDH (Codexis CDX-901, 50 U/mg) in 100 mM Na-phosphate buffer containing 10% v/v DMSO, adjusted to pH 7.5 (Reaction 2). The blank mix was added directly to the CFPS reactions or to the magnetic beads containing the immobilized variants (for details see Screening results). The reaction mixtures were incubated at 30°C with shaking at 600 rpm for 16–48 h. Analysis was done on HPLC-UV and HPLC-MS. Conversions (%) were calculated by using the HPLC area of limiting reactant (including hydrolyzed amino acid form when applicable) and expected product, and were given as area percentage of expected product at 210 nm, assuming similar UV responses of substrate and product.

LlNDT test reaction

For activity screening of the variant library, the reaction mixtures (150 μL) contained 5 mM 2-chloroadenine (8) and 50 mM 2′-deoxy-2′-fluorocytidine (7) in 10 mM NaP_i buffer, adjusted to pH 7.0. The reaction mix was added to the magnetic beads containing the immobilized variants or the wild type enzyme and to three empty wells as well. The reaction mixture was incubated at 40°C with shaking at 400 rpm for 18–22 h. Analysis was done on HPLC-UV and HPLC-MS. Transfer rate (%) was calculated as the ratio of HPLC area under the curve, at 254 ± 4 nm, of nucleoside product (2-chloroadenosine, 9) and the sum of 2-chloroadenine and 2-chloroadenosine given as a percentage, assuming similar UV response of substrate and product.

Measurement of specific activity

Enzyme activities were measured spectrophotometrically at 30°C in 96-well half-area plates in a Thermo Scientific Multiskan SkyHigh Microplate Spectrophotometer (Thermo Fisher Scientific Inc., Waltham, MA, USA). Standard assays were carried out by using 20 mM 5 and 0.4 mM NADPH in 100 mM sodium phosphate buffer containing 10% v/v DMSO (pH = 7.5). Depending on the activities of each enzyme, enzyme concentrations were fixed between 0.05 and 0.5 mg/mL. The reaction was started by the addition of 40 mM 4. The decrease in absorbance was monitored at 340 nm for 2.5 h. Initial velocities were calculated from the linear section of the plots using an NADPH calibration and specific activities calculated based on enzyme amounts used in each reaction.

Preparative-scale enzymatic reactions

Scale-up of (2S,2′S)-3. 100 mL reaction mixture contained 50 mM 1, 250 mM (5 eq.) 2, 0.25 mM (0.25 eq.) NADP+ and 2.5 equivalents of d-glucose (125 mM) and 6 U/mL GDH (Codexis CDX-901, 50 U/mg) in 100 mM Na-phosphate buffer containing 20% v/v DMF, adjusted to pH 7.5. The reaction mixture was incubated at 30°C with stirring at 200 rpm in a 250-mL Optimax glass reactor. RnKIRED V49I/H91I/S228G enzyme was added to the reaction in two aliquots (t₀: 10 mg, 25 h: 1 mg). pH was monitored and kept at 7.5 with 2M NaOH solution using an SP-50 Dosing Unit. The reaction was stopped after 69 h, when conversion reached 90%+, using 150 mL MeCN. The 5 G Celite was added to the solution, stirred for another 15 min, then filtered. The filtrate was concentrated in vacuo, then lyophilized. Purification was carried out using preparative HPLC. Fractions containing the (2S,2′S)-3 compound were merged, concentrated and lyophilized resulting in 543 mg pure product (yield = 39%, >99% de).

Scale-up of (2S,2′S)-6. 100 mL reaction mixture contained 50 mM 4, 200 mM (4 eq.) 2, 0.25 mM (0.25 eq.) NADP+ and 2.5 equivalents of d-glucose (125 mM) and 6 U/mL GDH (Codexis CDX-901, 50 U/mg) in 100 mM Na-phosphate buffer containing 10% v/v DMSO, adjusted to pH 7.5. The reaction mixture was incubated at 30°C with stirring at 200 rpm in a 250-mL Optimax glass reactor. RnKIRED H91A/S228G enzyme was added to the reaction in two aliquots (t₀: 10 mg, 24 h: 0.9 mg). pH was monitored and kept at 7.5 with 2M NaOH solution using an SP-50 Dosing Unit. The reaction was stopped after 44 h, when conversion reached 97%+. From the mother liquor 523 mg pure (2S,2′S)-6 could be isolated by filtration. Then, the filtrate was extracted using 2 x 200 mL DCM. The organic phases were merged and dried on MgSO4, followed by filtration of the drying agent and removal of the organic solvent in vacuo. The crude product was redissolved in water, cooled down to 4°C resulting in precipitation of the product. By filtration another 444 mg pure (2S,2′S)-6 could be isolated. (total yield = 63%, >99% de).

Analytical methods

HPLC methods. HPLC measurements were performed on Agilent Technologies 1260 LC system equipped with a DAD detector using Gemini 3 μm NX-C18, 50 mm × 3.00 mm i.d. 110 Å column and 5 mM aqueous NH₄HCO₃ solution and MeCN as eluents in gradient mode. Analytical LC-MS was performed on an Agilent Technologies 1200 LC system equipped with Agilent 6140 quadrupole MS, operating in positive or negative ion electrospray ionization mode (molecular weight scan range was 100–1350 m/z) with parallel UV detection using Gemini 3 μm NX-C18, 50 mm × 3.00 mm i.d. 110 Å column and 5 mM aqueous NH₄HCO₃ solution and MeCN as eluents in gradient mode.

Compound purification. Purifications using preparative HPLC were carried out with Teledyne Isco preparative HPLC using a Gemini 5 μm NX-C18, 250 mm × 50 mm i.d. 110 Å column and 5 mM aqueous NH₄HCO₃ solution and MeCN as eluents in gradient mode. Collected fractions were analyzed using HPLC, the fractions containing the desired product were merged, concentrated by rotavap, then lyophilized.

NMR measurements. ¹H NMR and ¹³C NMR spectra were recorded on a Bruker Avance Ultrashield 400 (100 MHz ¹³C) instrument with Bruker Prodigy Cryo Probe and are internally referenced to residual protium solvent signals (note: DMSO referenced at 2.50 and 39.52 ppm in ¹H and ¹³C NMR measurements, respectively. Samples (3–8 mg) were dissolved in 0.5 mL DMSO-d₆. Data for ¹H NMR are reported as follows: chemical shift (δ ppm), multiplicity (s = singlet, d = doublet, dd = doublet of a doublet, t = triplet, dt = doublet of a triplet, m = multiplet, br is used for broad signals), integration, coupling constant (Hz). Assignments of protons are listed on the individual spectra. Data for ¹³C NMR are reported in terms of chemical shift and no special nomenclature is used for equivalent carbons.

High resolution mass spectrometry. HRMS measurements were carried out on an Agilent 6545 Q-TOF mass spectrometer system, mass resolution: 45.000 FWHM @ m/z 2.722 Da, ion source: AJS-ESI, sheath gas temperature: 300°C, drying gas temperature: 300°C, ionizing voltage: 2500V, nozzle voltage: 1000V.

Quantification and statistical analysis

DNA concentration was determined by measuring absorbance at 260 nm using Thermo Scientific Multiskan SkyHigh Microplate Spectrophotometer (Thermo Fisher Scientific Inc., Waltham, MA, USA) with μDropDuo plate. Protein concentration was determined based on absorption at 280 nm measured on a NanoDrop spectrophotometer. For RnKIRED, conversions (%) were calculated by using the HPLC area of limiting reactant (including hydrolyzed amino acid form when applicable) and expected product, and were given as area percentage of expected product at 210 nm, assuming similar UV responses of substrate and product. For LlNDT, transfer rate (%) was calculated as the ratio of HPLC area under the curve, at 254 ± 4 nm, of nucleoside product (2-chloroadenosine, 9) and the sum of 2-chloroadenine and 2-chloroadenosine given as a percentage, assuming similar UV response of substrate and product. Enzyme activities were measured spectrophotometrically at 30°C in 96-well half-area plates in a Thermo Scientific Multiskan SkyHigh Microplate Spectrophotometer (Thermo Fisher Scientific Inc., Waltham, MA, USA) in triplicates. The decrease in absorbance was monitored at 340 nm. Initial velocities were calculated from the linear section of the plots using an NADPH calibration and specific activities calculated based on enzyme amounts used in each reaction.

Additional resources

Quick Protocol for Q5 Site-Directed Mutagenesis Kit (E0554) | NEB.

Mag-Bind TotalPure NGS – Omega Bio-tek.

NEBExpress Cell-free Ecoli Protein Synthesis System | NEB.

NEBExpress Ni-NTA Magnetic Beads | NEB.

Published: November 27, 2025