Omega transaminases: discovery, characterization and engineering
Palacio, Cyntia Marcela
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2019
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Palacio, C. M. (2019). Omega transaminases: discovery, characterization and engineering. Rijksuniversiteit Groningen.
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
Enhancing thermal stability of
Pseudomonas jessenii transaminase
by computational library design
Cyntia M. Palacio, Hein J. Wijma,Marleen Otzen, Dick B. Janssen Manuscript in preparation
Transaminases catalyze the reversible conversion of amines to keto compounds or aldehydes. Their diversity in combination with high chemo-, regio- and stereoselectivity of individual enzymes allow applications in the synthesis of various terminal and chiral amines. We have recently discovered a transaminases from a caprolactam-degrading strain of Pseudomonas jessenii that is active with 6-aminohexanoic acid, suggesting that the enzyme could be attractive for in vivo or in vitro synthesis of primary amines including nylon precursors. However, the protein appeared to be rather unstable and showed disappointing heat lability. In order to pave the way for improving the stability, we applied the computational library design strategy FRESCO (Framework for Rapid Enzyme Stabilization by Computational libraries). Energy calculations on all single mutants at 397 out of 450 positions followed by molecular dynamics simulations and on-screen inspection predicted 96 mutations outside the dimer interface that could enhance stability, of which 64 mutants at 38 positions were experimentally characterized. To examine if the experimental verification can be further restricted to mutations that have a high probability of being effective, high-temperature molecular dynamics simulations were performed. Positions in the protein that showed a higher degree of unfolding in these MD simulations were assumed to represent labile regions subject to early unfolding and were assigned a high priority. Experimental testing identified five variants that showed enhanced thermal stability, and all these carried mutations in highly promising regions predicted by MD. The results suggest that high-temperature MD is a useful tool for identifying target regions for enzyme thermostability engineering.
INTRODUCTION
In theory, enzymes are ideal catalysts for application in synthetic chemistry. Their diversity is enormous and enzymes usually exhibit high chemo-, regio- and stereoselectivity and impressive catalytic power, allowing mild reaction conditions. Nevertheless, issues such as low thermal stability, modest lifetime, and fragility during preparation, storage and use may limit application of enzymes in industrial processes. Especially thermostability is highly important. Often, it is desirable to perform a conversion process at elevated temperature because this increases reaction rates and substrate solubility and at the same time decreases viscosity and risk of microbial contamination. Thus, enzymes with high thermal stability and long shelf life are in great demand (1–3). Native enzymes isolated from mesophilic organisms often are not suitable for industrial application and need to be improved to remain sufficiently stable and active at elevated reaction temperatures. Other robustness properties that are desirable include compatibility with the presence of organic co-solvents and tolerance to high mixing rates and associated shear and surface forces (2, 4, 5). Furthermore, high stability might benefit industrial application because less biocatalyst is used, in part because recycling is more effective (6–9).
Over the last decades, many protein engineering strategies have been developed for improving protein thermal stability, including directed evolution, rational design, consensus-based methods, and computation-based methodologies (6–9). Directed evolution is a powerful strategy that is often used to improve protein stability and can yield enzymes that are more active in organic solvents as well as enzyme variants with improved or modified substrate specificity (9–11). Several commercial enzymes, including α-amylases, lipases and subtilisins, have been evolved using this strategy (12). Directed evolution starts with generating sequence diversity, either in a random manner or inspired by insight in structure-function relationships of the protein under study. Random methods include, for example, DNA shuffling (13), which may be combined with error-prone PCR (14). Although the combination of these methods provides a powerful engineering method, the protocols are normally time-consuming because of the need to perform repeated rounds of mutagenesis and screening (5, 15–17). Furthermore, these methods become impractical in case a high-throughput expression and screening method is not feasible, as is the case with fungal or mammalian expression systems.
A second strategy to create thermostable enzymes is rational design. This strategy involves careful structural inspection of 3D structures to identify positions where mutations may improve interactions that on the basis of the biophysical principles of protein folding are expected to increase the stability of the folded state or to destabilize the unfolded state. Stabilizing mutations may act through disulfide bond formation, fill hydrophobic cavities or enlarge hydrophobic clusters, compensate unpaired hydrogen bond donors or acceptors or improve electrostatic interactions. When introduced at
the right spot, such mutations may increase thermodynamic stability or decrease the rate of unfolding by stabilizing labile regions (18–22). The abundance of information on protein folding and 3D structures have established rational design as a reliable approach for finding improved enzyme variants (23–25). However, single mutations often have a modest effect, and successful rational design of stable enzymes often requires the introduction of multiple mutations with independent or additive thermostabilization effects. It may be problematic to discover by rational design the required mutations, which have small effects individually (26–30). The search can become very time consuming.
Supported by the development of improved energy functions for calculating the effect of mutations and triggered by the successful design of highly stable small proteins, interest is rapidly increasing in the use of computational methods for enhancing enzyme stability (3, 31, 32). Computational design indeed can provide a rapid pathway to improved enzymes with better thermal stability for specific process conditions (33– 35). In our laboratory, we have explored a computational strategy for improving thermal stability named FRESCO (Framework for Rapid Enzyme Stabilization by Computational libraries) (36). This procedure aims to predict and identify a large number of stabilizing mutations through subsequent in silico screening steps, which reduce the number of variants to be tested experimentally. The initial step consists of molecular mechanics energy calculations on all possible point mutations. Mutants that are predicted to be more stable are subjected to molecular dynamics simulations (37) and visual inspection (21, 38) and those that pass these tests are examined experimentally. Thereafter, confirmed stabilizing mutations are checked for compatibility by model building and simulations are combined to obtain highly stable variants (3). This strategy was successful for increasing the thermal stability of the limonene epoxide hydrolase from Rhodococcus erythropolis DCL14 by +35˚C (3). The same computational strategy was also used to improve the stability of a halohydrin and a haloalkane dehalogenase for use in the presence of organic cosolvents. The improved variants reached a +27˚C increase in stability without decrease of activity (31, 39). Similarly, a stable variant of peptide amidase was created this way for use in neat organic solvent, allowing improved application of the enzyme in peptide C-terminal coupling reactions (40). The most attractive feature of the FRESCO protocol is the reduction of library size to a group of less than 190 mutants, which are easily tested in one or two 96-well microtiter plates using rapid unfolding assays. Since the success rate is 10-15%, this yields typically 20-30 stabilizing mutations. Some are at the same position or may be incompatible, but the remaining mutations can be combined to obtain a highly stable variant. Nevertheless, strategies to further reduce library size without decreasing the number of positive variants would be highly attractive. For example, mutagenesis could be focused on early unfolding regions, also called weak spots, in the protein structure (20).
Table 1. Success rates amongst predicted stabilizing point mutations in FRESCO libraries.
enzyme number of mutationsA successful (%)fraction reference
tested stabilizing
dehalogenase LinB 109 10 9 (31)
halohydrin dehalogenase HheC 218 29 13 (39)
HMF oxidase 140 17 12 (41)
limonene epoxide hydrolase 64 11 17 (3)
peptide amidase 120 12 10 (40)
xylanase 105 10 10 (42)
A This only includes point mutations introduced in the starting variant of the enzyme.
Omega transaminases are PLP-dependent fold type I subgroup AT-II enzymes, which have been studied for the conversion of aldehydes and ketones to ω-amino acids, β-amino acids, chiral amines and terminal amines (43). The amine donor is a primary amine or ω-amino acid that transfers its amino group via an enzyme-PMP intermediate to the amino acceptor. These enzymes are considered to have high potential for industrial purposes (44). They show broad substrate specificity and apart from a cheap amine donor (alanine, isopropylamine) transaminases do not require expensive cosubstrates or redox-active cofactors (45). Due to these advantages and the environmentally friendly properties of the catalyst, transaminases are widely investigated for use in kinetic resolution of racemic amines and for the asymmetric synthesis of enantiopure amines from ketones (46, 47). One of the most relevant examples is their application in the manufacturing process of Sitagliptin, a top-selling anti-diabetic drug (48). In the original process, the non-enzymatic conversion of a β-ketoamide to Sitagliptin required the use of transition metal catalysts and high pressure. The metal catalyst was later replaced by a highly evolved transaminases, resulting in a reduction of waste production and energy consumption, as well as an increased overall yield and volumetric productivity (49). In general, transaminases are attractive for conversions which demand high stereo-, regio-, and enantioselectivity in combination with high reaction rates.
In this chapter we explore the use of the FRESCO workflow to improve the thermal stability of a recently discovered and characterized ω-transaminase involved in the caprolactam degradation pathway (50). This enzyme was isolated from the caprolactam degrading bacterium Pseudomonas jessenii strain ODJ3. Caprolactam is widely used for the production of Nylon 6 (51), a polymer used for manufacturing of numerous products, such as toys, fabrics, clothing, utensils, mechanical parts, etc. Protein mass spectrometry demonstrated that this transaminase is induced in caprolactam-grown cells. The purified transaminase catalyzes the reversible transamination of the aldehyde 6-oxohexanoic acid into 6-aminohexanoic acid. Both these compounds are intermediates in an artificial biosynthetic pathway towards caprolactam and in the natural pathway of caprolactam biodegradation (50, 52). The Pseudomonas jessenii transaminase showed high activity
towards a wide variety of aromatic and aliphatic linear amines. However, the low stability of the transaminase at high temperatures limits its potential applications.
For the above reasons, we set out to target the PjTA enzyme as a model system to apply the FRESCO protocols for enhancing the stability of a PLP dependent aminotransferase. Since the enzyme is rather large (455 amino acids), we examined the use of high-temperature MD simulations to examine if this approach can be used to reduce library size and increase library quality. Whereas reports on improving stability of transaminase emerged while this work was ongoing (53, 54), these studies did not focus on surface mutations or computational selection of target regions. After selection of mutants, we performed experimental screening and identified five variants with increased thermal stability, including two mutants with increased catalytic activity. These were in regions that by MD simulations were predicted to be priority targets for enhancing stability.
MATERIALS AND METHODS
Substrates and chemicals
Nicotinamide adenine dinucleotide (NAD+), 6-aminohexanoic acid and alanine
dehydrogenase from Bacillus cereus were purchased from Sigma-Aldrich. Pyridoxal phosphate (PLP) was purchased from Fisher Scientific. Pyruvic acid was acquired from Acros Organics. Potassium phosphate dibasic trihydrate and potassium phosphate monobasic were obtained from Merck Millipore. The substrate (S)-α-methylbenzylamine was obtained from Acros Organics. The fluorescent stain SYPRO Orange was purchased from Life Technologies.
Mutagenesis
Mutants of PjTA were created by QuikChange mutagenesis in 96-well plate format. The vector was pET20b(+)-His-PjTA, containing an in-frame fusion of the PjTA gene as described ((55), see also Chapter 2 of this thesis). The reactions were performed in parallel. Primers were designed with the QuikChange Primer Design Program of Agilent Technologies. After transformation to strain E. coli TOP10, also in a 96-well microtiter plates, transformants from individual wells were plated onto agar in 24-well microtiter plates. Plasmids from colonies were analyzed by agarose gel electrophoresis (0.8%) and clones showing plasmid with the expected size were transferred to 96-well plates, grown at 37ºC and stored in the presence of 15% glycerol at -80ºC. The mutations were verified by DNA sequencing (Eurofins Genomics).
Enzyme expression and purification
The enzymes were produced as described by Palacio et al. (56). Briefly, cells were cultivated in 250 mL flasks overnight at 30°C on TB medium with induction by IPTG, collected by centrifugation, and lysed by sonication. Purification was carried out in Pall AcroWell 96-well filter plates (Sigma Aldrich) loaded with TALON metal affinity chromatography resin (Clontech). The desalting step was done with a multi-well gravity column PD MultiTrap G-25 (GE Healthcare). The purification of PjTA from larger cultures was performed as described by Palacio et al. (56). The enzymes were stored at 4°C in 100 mM potassium phosphate, pH 8.0.
Thermal shift assay
Melting temperatures (Tm) were determined by the thermofluor method, which is based on an enhancement of fluorescence upon binding of a hydrophobic fluorescent compound to protein that becomes unfolded when the temperature is increased (57). Five µl of 100-fold diluted SYPRO orange and 20 µl of a 0.1 to 1.5 mg/ml enzyme solution were added to an IQ 96-wells PCR plate (Bio-Rad). Phosphate buffer without protein was used as a control. The plates were sealed with Microseal B adhesive sealer (Bio-Rad) and heated in a MyiQ real-time PCR machine (Bio-Rad) from 25°C to 90°C at 1°C/min. The wavelengths for excitation and emission were 490 nm and 575 nm, respectively (58). The temperature at the moment of maximum rate of fluorescence change (dRFU/dT) was taken as the apparent melting temperature (TM,app) (31).
Activity assay
Enzyme activities were measured by coupling the aminotransferase to B. cereus alanine dehydrogenase as described by Palacio et al. (55). The appearance of NADH at 340 nm (εNADH = 6.22×103 M-1∙cm-1) was monitored using a microtiter plate reader (Synergy Mx,
BioTek Instruments). Assays were performed in 96-well plates containing reaction mixtures lacking the amino acceptor pyruvate. A solution of pyruvate was added to each well to initiate the reaction. The absorbance at 340 nm was monitored for 30 min at 30˚C. The initial rates of appearance of NADH by the alanine dehydrogenase were used to determine the kinetic constants of the aminotransferase. Specific activities are expressed in units/mg (µmol∙min-1∙mg-1).
Identification of priority regions by MD
To predict regions in the PjTA structure that might be sensitive to unfolding and thereby serve as targets for stability engineering, the PjTA-PLP structure was subjected to high-temperature molecular dynamics simulations. The simulations were done at 500 K to allow detection of structural changes that happen too slowly at ambient temperature. Three X-ray structures of P. jessenii TA which were solved in-house were used: PjTA-ap, the apoenzyme with phosphate bound (1.87 Å resolution with an Rfree of 0.184); PjTA
-as, the apoenzyme with succinate bound (resolution 1.80 Å, Rfree 0.203); and PjTA -plp, the PLP-bound enzyme (2.15 Å resolution, Rfree of 0.261). First, molecules originating from the crystallization buffer were removed and protonation states of ionizable groups were assigned (59) assuming pH = 8.0.
Simulations started with an energy-minimized structure, and the temperature was increased from 5 K to the final simulation temperature of 500 K in the first 30 ps, followed by 32 ns of MD simulation at 500 K. A periodic boundary simulation cell was used with distances of at least 7.5 Å between enzyme and the cell walls. The integration time step was 1.25 fs with non-bonded interactions updated every 2 simulation steps. Simulations were performed with the Amber11 force field under Yasara (60). The simulations were performed in triplicate, using different initial atom velocities according to a Boltzmann distribution (61, 62). Snapshots were saved every 25 ps. For every snapshot the RMSD (root mean square deviation) was determined and recorded for each amino acid. A local increase in RMSD observed in the 500 K simulations was interpreted as indicative of unfolding. Some of these structural changes might be due to the omission of crystal contacts. To avoid these being interpreted as a sign of unfolding as well, an averaged structure from a similar set of simulations at 298 K was used as a reference instead of the crystal structure. This reference was obtained from 10 simulations of each 2.5 ns of which the equilibrated last 0.5 ns was averaged. The use of repetitive simulation runs with independent initialization improves coverage of conformational space (62–64) whereas in case of high temperature MD simulations we used timescales long enough to detect early unfolding events.
When a partially unfolded protein is superpositioned on its folded structure using standard methods, the apparent RMSD is increased throughout the protein structure since the alignment algorithm minimizes the overall RMSD. Since this can mask local unfolding, we used the motif superpositioning method of Vriend and Sander (65). This algorithm only superimposes the folded parts of the protein and gave higher RMSDs for truly unfolded residues and a lower RMSD for stable regions.
A method to convert high temperature RMSD values to priorities for mutagenesis was developed in house (to be published elsewhere). Briefly, the RMSD per residue calculated as explained above was averaged and priorities were assigned to positions as a function of this RMSD and the RMSD of residues with which it interacts in the 3D structure. The latter was included because residues critical for stabilizing interactions may not appear as early unfolding themselves, but may allow unfolding of regions with which they interact in a suboptimal manner.
Computational design of stabilizing mutations
To design potentially stabilizing mutations in the PjTA protein, we followed the FRESCO protocol (framework for enzyme stabilisation by computational library design) (3, 66, 67). First, we calculated effects on free energy of folding for point mutations at all
positions in the protein, substituting with all proteinogenic amino acids except cysteine. The difference in free energy of unfolding of each point mutant (ΔΔGfold) was calculated
using Rosetta (68) with the Row 3 protocol [options: –ddg::local_opt_only true – ddg::opt_radius 8.0 –ddg::weight_file soft_rep_design -ddg::iterations 50 -ddg::min_ cst false -ddg::mean true -ddg::min false -ddg::sc_min_only false -ddg::ramp_repulsive false]. The same calculations were performed with FoldX (69), using its standard settings. All FoldX calculations were repeated five times to obtain better averaging. Both for the FoldX and the Rosetta protocol, the results over the three dimeric aminotransferase structures were averaged.
To eliminate mutations that are unlikely to be stabilizing we used visual inspection and MD simulations, which is a standard part of the FRESCO procedure (3, 31, 40, 67) For each mutant, a 3D structure was predicted by FoldX using the crystal structure of the wild-type PjTA -plp as the template. The resulting models were used for 100 ps MD simulations. For each mutant, five simulations at 298 K were performed with a different set of initial atom velocities. Yasara was used with the same settings as above, except that the Yamber3 force-field was used (70). The visual inspections were recently described in detail (36) and result in the dismissal of mutations causing solvent exposure of hydrophobic side chains, introduction of unsatisfied H-bond donors and acceptors, and other obvious structural problems.
We also examined the effect of the use of more strict energetic criteria for mutant selection, similar to protocol recently published under the acronym FireProt (71). Again, FoldX and Rosetta are used to calculate the effect of mutations on ΔΔGfold, but only
mutations were accepted for which the calculated ΔΔGfold is better both by at least
-4.184 kJ mol-1 for FoldX and by -8.368 kJ mol-1 for Rosetta. The stricter criteria could
prevent false-positive predictions causing the accidental introduction of destabilizing mutations. The settings for the Rosetta Row 16 protocol were: -ddg::weight_file soft_ rep_design -ddg::iterations 20 -ddg::local_opt_only false -ddg::min_cst true -ddg::mean false -ddg::min true -ddg::sc_min_only false -ddg::ramp_repulsive true. The Row 16 protocol uses energy minimization of the entire protein while the Row 3 protocol performs energy minimization only on amino acid sidechains that are within 8 Å of the mutated residue. The prediction accuracy of the computationally much more expensive Row16 protocol is slightly better: for a large benchmark set of mutations with known ΔΔGfold the correlation coefficient R2 was 0.68 for the Row 3 protocol and 0.69 for the
Row 16 protocol (68). The less strict criteria commonly used in the FRESCO workflow (in this study ΔΔGfold = -2.5 kJ/mol per subunit both for FoldX and Rosetta, unlike with
the strict criteria the mutation needs to pass only FoldX or Rosetta to be accepted) will initially select a large number of mutations, from which pathological mutations are removed by orthogonal tests, i.e. molecular dynamics simulations and visual inspection. The few mutations that passed the stricter energy criterion were also experimentally characterized if they did not pass these orthogonal tests.
RESULTS
Stability of P. jessenii aminotransferase
Preliminary experiments reported by Otzen et al. (50) indicated that a P. jessenii transaminase (PjTA) that is active on 6-aminocaproic acid is unstable upon freezing and storage. Formation of a precipitate was observed when the enzyme was thawed after storage at -20˚C in 50 mM potassium phosphate, pH 8, containing 10% glycerol. After removing this material by centrifugation, an activity test was performed with the supernatant and no activity was observed. When the enzyme was stored in the same buffer at 4˚C, activity decreased 36% over a period of 15 days. In view of these stability problems, we decided to study the possibility to stabilize the enzyme by protein engineering, at the same time testing if the use of high-temperature molecular dynamics would be helpful to identify priority mutations that contribute to enhanced stability.
Selection of target regions
The FRESCO workflow for stability engineering makes use of computational protocols to identify substitutions that have a high probability of contributing to enhanced stability. Although the in silico design and screening strategy is effective in terms of final outcome, the number of experimentally confirmed positives (more stable variants) in the designed libraries is still modest (Table 1). In case of large proteins FRESCO suggests a larger number of stabilizing mutations than what one would like to examine in the laboratory. Earlier applications of the FRESCO protocol mainly concerned enzymes with lower molecular weight than PjTA, which is a 455 amino acid dimeric protein with 50 kDa subunits. To reduce the numbers of variants that must be tested in the lab and increase the success frequency, the FRESCO strategy would benefit from an additional computational step that can discern topological regions in a protein where mutations are effective from regions where mutations cannot make a contribution to enhanced stability. Such a method should preferably be orthogonal to the already applied methods of energy calculations, MD simulations, and visual inspection.
The irreversible denaturation of proteins at high temperature may start at early unfolding regions (20). Exposing hydrophobic patches may trigger aggregation and thereby cause cooperative irreversible inactivation. Mutations in such labile regions may enhance overall stability. This is the basis of the so-called B-fit approach, which aims to stabilize flexible regions identified by high crystallographic B-factors (72, 73). An alternative approach would be to pinpoint early unfolding regions where mutations may enhance stability by using molecular dynamics simulations (74). However, the timescales of MD simulations are too short in most cases to detect such functionally relevant structural events. A way to enhance the unfolding rate in simulations is to increase the temperature (75). We therefore performed MD simulations both at 298 and at 500 K
and compared RMSD changes. Starting with the recently solved crystal structure with PLP bound (PjTA-plp) 32 ns trajectories were generated in triplicate by MD simulations at 500 K. For the room temperature simulations, 10 trajectories of each 2.5 ns were averaged to obtain an MD equilibrated structure, which served as a reference for the 500K simulations. This averaged structure was used as a reference instead of the X-ray structure because the latter would make the comparisons sensitive to effects of loss of crystal contacts or inaccuracies in the force-field. The results of the high temperature MD simulations and RMSD calculations were used to assign each position with a priority rank, ranging from 1 to 449, where the highest rank corresponds to the highest regional RMSD, calculated as described in the Materials and Methods.
The usefulness of the unfolding predictions was subsequently examined in the laboratory with PjTA residues that are not located at the interface. In total 100 positions with high priority (rank 1-100) and 40 positions with low priority (rank 410-449) were selected for experimental analysis (Table 2; Fig. 1).
Computational design of point mutations
Initial energy calculations were carried out for all possible substitutions at all 397 positions that were further than 10 Å from the PLP cofactor. Changes in free energy of unfolding (ΔΔGFold) were calculated using FoldX and Rosetta with the three different PjTA structures
and the results were averaged, yielding 323 potentially stabilizing substitutions (ΔΔGFold
per dimer better than -2.5 kJ/mol for FoldX or Rosetta) in the high-priority regions and 128 mutations in the low priority region (Table 2). The predicted structures of these 451 mutants were used for MD simulations to identify some mutations that should be dismissed because of expected destabilizing effects as indicated by structural problems or enhanced flexibility (Table 2). Due to inaccuracies in the energy functions and the conformational sampling methods, both FoldX and Rosetta tend to select a mixture of stabilizing mutations and mutations decrease stability due to the biophysical problems that they introduce. Therefore, mutations were inspected on screen, which dismissed a further set of mutations (Table 2). Interface mutations were excluded because the priority assignment algorithm was known not to work for such mutations (see above).
Fig. 1. Identification of surface-located target regions for thermostabilisation of P. jessenii transaminase by high temperature MD simulations. The priority (thicker lines indicate higher priority) and the location of the selected mutations (numbered spheres) are indicated. Orange spheres indicate positions which were given high priority by the 500K RMSD + neighbour correction algorithm. Positions with black spheres were given a low priority. The PLP cofactor is indicated with yellow carbon atoms. The two subunits are in cyan green and marine blue.
The combined selection steps in the FRESCO protocol reduced the set of promising mutations in high- and low priority regions to a total of 88 (Table 2). At the same positions, the use of a set of stricter energy criteria (but without the inspection steps) resulted in selection of only 18 mutations (Table 2). This included 8 mutations that were not selected by the FRESCO protocol (Table 3). As a result, a total of 96 mutations were selected for experimental characterization.
Mutagenesis and expression
Of the 96 selected mutations remaining by in silico design and MD screening, 68 were obtained by QuikChange PCR as confirmed by DNA sequencing. For examining expression and thermostability, the mutants were grown in 50 mL cultures in 250 mL flasks after which cells were lysed and mutant enzymes were isolated by His-tag metal affinity chromatography in microtiter plates. Using this procedure, we isolated from each 50 ml culture 10-45 mg of purified PjAT variant. Concentrations were determined by Bradford assay. In total 64 PjAT variants were obtained in soluble form, indicating that detrimental mutations which completely abolished expression were rare. The purified enzyme samples were used for thermofluor stability assays.
substitutions (''G
per dimer better than -2.5 kJ/mol for FoldX or Rosetta) in the
high-priority regions and 128 mutations in the low priority region (Table 2). The
predicted structures of these 451 mutants were used for MD simulations to identify
some mutations that should be dismissed because of expected destabilizing effects as
indicated by structural problems or enhanced flexibility (Table 2). Due to inaccuracies
in the energy functions and the conformational sampling methods, both FoldX and
Rosetta tend to select a mixture of stabilizing mutations and mutations decrease stability
due to the biophysical problems that they introduce. Therefore, mutations were
inspected on screen, which dismissed a further set of mutations (Table 2). Interface
mutations were excluded because the priority assignment algorithm was known not to
work for such mutations (see above).
The combined selection steps in the FRESCO protocol reduced the set of
promising mutations in high- and low priority regions to a total of 88 (Table
2)
. At the
same positions, the use of a set of stricter energy criteria (but without the inspection
Fig. 1. Identification of surface-located target regions for thermostabilisation of P.
jessenii transaminase by high temperature MD simulations. The priority (thicker lines
indicate higher priority) and the location of the selected mutations (numbered spheres) are indicated. Orange spheres indicate positions which were given high priority by the 500K RMSD + neighbour correction algorithm. Positions with black spheres were given a low priority. The PLP cofactor is indicated with yellow carbon atoms. The two subunits are in cyan green and marine blue.
CHAPTER 4
Thermostability assays
The stability of the isolated PjTA variants was examined by thermofluor assays, which measure real-time thermal unfolding in the presence of a fluorescent dye. The assays were done in microtiter plates to allow comparison of large numbers of mutants, using 1-20 μg/well samples of protein per assay. The changes in apparent melting temperature (ΔTM,app) in comparison to the wild-type TM,app revealed the stabilizing effect of the designed mutations. The wild-type enzyme displayed modest stability with a TM,app of 60.25 ºC. All 64 mutants that gave expression of soluble protein gave a good signal in these thermofluor assays (Table 3).
Of the 13 mutants (12 FRESCO mutants plus 1 mutant predicted by the stricter FireProt energy criteria) substituted in low priority regions, none displayed an enhanced stability (ΔTM,app ≥1.0˚C) in thermofluor assays compared to the wild-type PjTA. On the other hand, when examining mutations in the high priority regions, it appeared that out of 51 mutations (39 from FRESCO only, 6 from strict energy criteria, and 6 from both) attempted mutations, 5 mutations gave a modest but significant increase in thermostability. These 5 all came from the more relaxed FRESCO energy criteria and orthogonal inspection by MD and on-screen visualization (Table 3). Of the other 46 mutants in high priority regions, 34 showed a change in ΔTm,app of less than 1.0°C, which is too low for reliable conclusions, whereas 12 of those 46 mutants gave a slightly lower stability than the wild type. The discovery of 5 stabilizing mutations only in the high-priority group does not unequivocally proof that high-temperature MD identified regions where computationally designed mutations can contribute to an increase in stability because of the smaller number of tested mutations localized in low priority regions. Also the fraction of stabilizing mutations (5/45=11%) amongst the high priority mutations does not exceed the success rate in earlier FRESCO studies where no strategy was applied to target early unfolding regions (Table 1). Thus, the current system and data do not provide support for the hypothesis that high temperature MD simulations helped to decrease the screening effort required to find good mutations.
Table 2. Computational design of stabilizing mutations using different energy criteria and ranking methods.
FRESCO criteria Stricter criteria
Step total priority1 - 100 410-449priority total priority 1 - 100 410-449priority
theoretical possibilities 2520 1800 720 2660 1900 760
FoldX criteria (kJ/mol) A < -2.5 < -2.5 < -2.5 <-4.2 <-4.2 <-4.2
No. of FoldX mutants 167 124 43 81 58 23
Rosetta row protocol
criteria (kJ/mol) A, B < -2.53 < -2.53 < -2.53 < -8.416 < -8.416 < -8.416
No. of Rosetta mutants 375 267 108 24 17 7
total number of energy-based mutationsC 451 323 128 24 17 7
position at the dimer interface (eliminated) 38 0 38 6 0 6
solvent exposed hydrophobic groups 130 96 34 - -
-unsatisfied H-bond donors/acceptors 46 42 4 - -
-other problems D 86 62 24 - -
-increased side chain flexibility 6 5 1 - -
-increased local backbone flexibility 21 19 2 - -
-increased flexibility of entire protein 2 0 2 - -
-no solid conclusion by inspectionE 6 30 6 - -
-total number of mutations remaining 88 69 19 18 17 1
protein expression obtained 57 45 12 12 11 1
increased Tm,app 5 5 0 0 0 0
decreased Tm,app 11 9 2 4 4 0
no effect on Tm,app 42 32 10 8 7 1
success (%)F 8 11 0 0 0
-A Energy criteria are per monomer. B Calculations with the Rosetta Row 16 protocol were only carried out for mutants that passed the FoldX criterion because the calculations were very CPU expensive. C passed either Rosetta or FoldX criteria or both. D Occasionally it occurred that all mutations at a certain position were predicted to be stabilizing, suggesting a systematic error in the energy calculation for the wild-type structure. E In the standard FRESCO procedure such mutants are included but they were omitted here to restrict library size. F equals the number of stabilizing mutations divided by the number of mutations examined experimentally.
No stabilizing mutations were discovered within the smaller set of 12 tested mutant proteins that were suggested when using the stricter energy selection criteria (Table 2). These 12 mutations were almost exclusively located at high priority positions, and 6 of them were selected while that the visual inspection or MD simulations of the FRESCO protocol suggested their elimination. Of these disputed mutations, 4 were even significantly destabilizing. Of these 4, 3 had been eliminated from the FRESCO library because they introduced a solvent exposed hydrophobic group on the protein surface (D45F, D45Y, and E345M).
Table 3. Mutations influencing stability of PjTA.
Priority Mutation Predicted ΔΔG
Fold (kJ/mol)A
FRESCO
selected selectedFireProt ObservedΔTM,app FoldX Rosetta Row 3 Rosetta Row 16
23 S6K 0.7 -5.3 ND + - 0.5 58 P9A 17.9 -8.7 ND + - 1.5C 58 P9K 6.7 -14.8 ND + - 1.5C 58 P9R 6.1 -9.9 ND + - 0.25 39 A23L -18.4 1.8 -33.7 + + 0.25 11 E28K -4.5 -16.4 ND + - 0 43 D45I -12.2 -11.3 -18 + + 0.5 43 D45V -6.5 -5 ND + - 0 43 D45F -12.8 -17.5 -22.2 -B + -1 43 D45K -8.8 -2.1 12.3 + - -1.25 43 D45Y -14.3 -23 -24 -B + -1.25 43 D45W -9 -31.8 -25.5 -B + -0.25 40 Q46K -0.5 -6 ND + - 0.25 40 Q46N 0.3 -8.5 ND + - -0.5 69 G49A -1.1 -7.9 ND + - -2.25 69 G49P -13.3 12.9 -22.1 + + -0.25 61 E103A 6.2 -8.2 ND + - -0.25 61 E103K 0.3 -13 ND + - 0 86 P139K 11.2 -6.4 ND + - 1 72 G166S -1.6 -13.2 ND + - -0.25 425 D252K -1.9 -7.2 ND + - 0 45 Y187M -7.1 30.5 ND + - -1.5 45 Y187F -7.3 2.8 ND + - -1.25 13 L191A 1.5 2.1 ND + - -0.25 49 E196P -7.1 14.1 ND + - -0.25 64 H198A 3.2 -9.1 ND + - 0.25 64 H198R -1.7 -7.6 ND + - 0 64 H198Q -1.7 -10 ND + - 0.25 15 A192P -7.5 -11.1 ND + - 0.25 410 K210L -10.1 -10.1 -10.1 + - -0.5 89 R238H -6.3 14.9 ND + - 0.25 419 E242A 0.1 -7.6 ND + - 0.5 419 E242K -1.3 -9.3 ND + - -0.75 419 E242P -8.2 14.4 ND + - -0.5 425 D252K -1.9 -7.2 ND + - 0 85 Q269H 0.6 -11.2 ND + - 0.75 85 Q269N 0.7 -8.3 ND + - 0.25 411 A303P -8.6 8.2 -5.8 + - 0.25 412 P304A 15 -9.6 ND + - -1.25 88 S360K -5.3 -12.1 ND + - -0.75 88 S360A -2.1 -9.4 ND + - -0.75 67 E345M -11.5 -14.4 -17.5 -B + -1.75 67 E345A 0.7 -9.6 ND + - -2.5 67 E345K -9 -19.2 -12.9 + - -3 67 E345N -3.2 -11 ND + - -3 423 Q353K -0.2 -7.3 ND + - 0.75 99 R359L -15.8 -21.6 -28.8 + + 0.25 99 R359I -11.6 -9.6 -25.9 + + -0.25 47 D367K -0.9 -9.4 ND + - 0.5 47 D367N 1.3 -11.3 ND + - 0.75 1 S391K 3.6 -7 ND + - -0.25 4 A393R -7.9 9.2 ND + - 1.0 18 L398Y -6.4 3.7 ND + - -1.5 18 L398P -7 -7.6 ND + - -0.25 38 R403A -0.1 -5.5 ND + - 0.5 62 G407N -7.3 -15.3 ND + - -0.25 62 G407Y -9.1 -16.6 -26.3 -B + 0 78 M419L 4.6 -8.3 ND + - 1.25C 437 E433P -3 -8.9 ND + - 0.25 437 E433Q 4.5 -6.4 ND + - 0 420 G437Y -18.7 -22.2 -32.1 -B + 0.5 420 G437N -1.7 -11.8 ND + - -1 83 T450L -22.6 -9.4 -29.2 + + -1.25 73 T451F -6.3 -7.9 ND + - -0.25
4
20 40 60 80 100 -500 0 500 1000 1500 Temperature (°C) dR el at ive flu or escen ce/ dT WT M419L P139K P9A P9K A393R
Characterization of stabilized PjTA variants
The five variants that showed a significantly improved apparent melting temperature (ΔTM,app > 1°C, Fig. 2) carried mutations that were mainly located at low B-factor regions (Fig. 3). These stabilizing mutations were distributed over four different positions: P9, P139, A393 and M419 (Fig. 3). Of these, only A393 is located at a high B-factor region. The presence of most of the mutations in relatively rigid areas indicates that high B-factor regions are not the only areas where mutations can stabilize a protein.
Fig. 2. Thermostabilities of wild-type PjTA and the five best variants determined by the thermofluor method.
To analyze whether the increased thermostability affected the catalytic performance of the mutants, specific activities of the five most stable variants and the wild-type PjTA were tested with (S)-α-methylbenzylamine (MBA) and 6-aminohexanoic acid (6-ACA) (Table 4). MBA is often used for the detection of transaminase activity (76) and 6-ACA is of interest because of its role as an intermediate in the caprolactam/nylon 6 degradation pathway (31, 50). Variants A393R and P9A showed higher activity while P9K showed wild-type activity. Different effects on activity are observed more often as result of thermostabilization protocols (3). On the other hand, variants M419L and P139K displayed slightly decreased activities. Thus, the single-site mutations with improved thermal stability seem to have only a moderate impact on the specific activity.
103 Fig. 3. Positions of stabilizing mutations in the 3D structure of PjTA. The colored spheres show the stabilizing mutations and the effect on TM,app. Protein flexibility determined by crystallographic B-factors is represented by the thickness of the backbone.
Table 4. Specific activities of the PjTA mutants with enhanced thermal stability.
Variant Specific activity (μmol min-1 mg-1)
(S)-α-Methylbenzylamine 6-Aminohexanoic acid
WT 0.54±0.01 0.17±0.01 M419L 0.48±0.01 0.14±0.01 P139K 0.40±0.01 0.13±0.01 P9A 0.64±0.01 0.16±0.01 P9K 0.55±0.01 0.16±0.01 A393R 0.71±0.01 0.20±0.01
Activities were measured with a coupled assay, using pyruvate as amine acceptor and alanine dehydrogenase to couple its conversion to alanine to NADH formation.
DISCUSSION
With six published examples, the FRESCO protocol has emerged as an efficient strategy for enhancing thermostability (3, 31, 40–42, 67). Computational design and in silico screening of mutant libraries preceeds laboratory expression and stability screening, thereby strongly reducing the amount of work that needs to be done to find stable variants. However, even with careful energy calculations, inspection and MD simulation, the typical success rate of point mutations still does not exceed 10-20%, suggesting that further improvements are desirable. In the work reported here, we examine the use of high-temperature MD simulations as a means to identify relatively flexible surface regions that may be involved in early unfolding and triggering aggregation leading
!
Fig. 3. Positions of stabilizing mutations in the 3D structure of PjTA. The colored
spheres show the stabilizing mutations and the effect on TM,app. Protein flexibility
determined by crystallographic B-factors is represented by the thickness of the backbone.
Table 4. Specific activities of the PjTA mutants with enhanced thermal stability.
d('0(,)! ".%$0;0$!($)0Q0):!=h*+2!*0, WC!*-WC?! ="?WiWJ%)@:21%,3:2(*0,%! NW<*0,+@%K(,+0$!($0/! fP! D9OLjD9DC! D9CMjD9DC! JLCEA! D9LIjD9DC! D9CLjD9DC! \CFEb! D9LDjD9DC! D9CFjD9DC! \E<! D9NLjD9DC! D9CNjD9DC! \Eb! D9OOjD9DC! D9CNjD9DC! <FEF5! D9MCjD9DC! D9HDjD9DC! <$)0Q0)0%&!Y%'%!*%(&#'%/!Y0)@!(!$+#.2%/!(&&(:Z!#&0,-!.:'#Q()%!(&!(*0,%!($$%.)+'!(,/!(2(,0,%! /%@:/'+-%,(&%!)+!$+#.2%!0)&!$+,Q%'&0+,!)+!(2(,0,%!)+!X<[G!;+'*()0+,9
Discussion
With six published examples, the FRESCO protocol has emerged as an efficient
strategy for enhancing thermostability (3, 31, 40–42, 67). Computational design and in
silico screening of mutant libraries preceeds laboratory expression and stability
screening, thereby strongly reducing the amount of work that needs to be done to find
stable variants. However, even with careful energy calculations, inspection and MD
simulation, the typical success rate of point mutations still does not exceed 10-20%,
1.5 ˚C 1.25 ˚C 1 ˚C
!
to irreversibe inactivation. Mutations located in or contacting early unfolding regions are expected to have a higher probability of increasing stability, especially in case of large proteins where unfolding is irreversible and controlled by kinetics rather than thermodynamic factors.
To examine this approach we designed and constructed a library of 64 variants with single mutations (none of which were located at the subunit interface) at positions that encompass different classes of flexibility. The flexibility classes were identified in high-temperature molecular dynamics simulations and used to assign priority scores to the mutations, high priority areas being regions with highest increase in RMSD in MD simulations. Regions were defined not only as amino acids showing large Cα RMSD but also nearby residues. The mutants were produced and expressed in soluble form. Of the set of 64 variants, 51 were located at 29 positions in high priority areas and 13 substitutions at 9 positions were in non-favoured areas. The effects on stability, as determined by the thermofluor assays, were modest, even for the high priority regions, with five mutants showing an increase of more than 1˚C of TM,app (success rate 11%, 5/46, 46 corresponding to the 51 mutations in high priority areas with 5 negative controls included to test the stricter energy criteria excluded, see Table 3). The vast majority of the mutations had neutral effects and some were detrimental for stability.
The percentage of success found here with FRESCO-designed surface mutations in PjTA is lower than what was reported by Wijma et al. 2014 (3) for another dimeric protein (39). There, a percentage of success of 17% was found amongst point mutations in the first stage of experimental screening with the limonene epoxide hydrolase. Arabnejad et. al. 2016 (67) reported 29 stabilized mutants of halohydrin dehalogenase after the screening of 218 mutants (13%), which shows a similar percentage of success to what is obtained here with PjAT. All five mutants with significant positive effect were located in the predicted early unfolding regions, whereas no such stabilizing mutations were found in the non-flexible regions.
We also observed that the use of the stricter energy criteria as implemented elsewhere (71) (requiring -8.2 kJ/mol for Rosetta and simultanously -4.2 kJ/mol for FoldX predictions instead of requiring -2.5 kJ/mol for at least one of both software predictions) did not improve the selection of mutants. No stabilizing variants were found in the resulting smaller set of 12 mutations and 4 out of these 12 mutations were even destabilizing. The destabilization can be due to the introduction in 3 (out of the 4) variants of a hydrophobic group on the protein surface. We have frequently observed that such mutations are proposed by energy calculations (negative ΔGFold)
and elsewhere it has been found for a small and well soluble model protein (ThreeFoil) that such mutations indeed had a beneficial effect on ΔGFold but decreased solubility
(77). For larger proteins such as enzymes, the lower solubility may decrease thermal stability by accelerating irreversible aggregation, which outweighs the benefits of the improved ΔGFold. Also with limonene epoxide hydrolase it was observed that mutations
4
that were predicted to be strongly stabilizing were destabilizing due to introduction of hydrophobic aromatic groups on the protein surface (3). Based on these observations, mutations that introduce hydrophic groups are usually dismissed in the FRESCO protocol before laboratory testing. It appears that introduction of hydrophobic groups on the protein surface is a systematic problem of the prediction algorithms that cannot be solved just by setting the energy criteria for accepting mutations to more restrictive values.
Four of the five mutants with enhanced stability showed a modest reduction of catalytic activity (Table 4). An inverse relationship between stability and function has been observed in other enzymes, including citrate synthase (78), a Staphylococcal nuclease (79), thioredoxin (80) and lysozyme (3). Structural inspection provided clues about the cause of loss of (some) activity. For example, M419L is located close to the tunnel-like active site of the enzyme and may influence substrate entry, causing some reduction of enzyme activity. The P9K and P9A surface mutations displayed the highest positive ΔTM,app values (1.5˚C). Position 9 is in the vicinity of a predicted flexible region (Fig. 3). Previously, areas near flexible regions were found to be critical for enzyme stability (81). Additionally, proline is a potent helix disruptor which can adopt just a few conformations. Thus, the substitution by alanine or lysine seems to help the formation of stabilizing secondary structure. The PjAT variant P139K also showed a decrease in the specific activity and an improvement of the thermostability. The replacement of a proline by a lysine may cause an increase in loop flexibility and added a positive charge. Results from studies on ubiquitin, human acylphosphatase, and Cdc42GTPase have shown that the stability of enzymes can be increased via the optimization of the surface charge–charge interactions (82).
Fig. 4. Location of stabilizing mutations. A, position of M419L, which caused a minor decrease in activity. The PLP cofactor bound in the active site is shown in pink while the location of M419 is shown by the magenta patch on the surface. B, position of P9 in a surface loop. The carbonyl O is hydrogen bonded to Lys142.
The fifth mutation is A393R, located in one of the regions with the highest crystallographic B-factor. It provides a moderate increase in thermostability and a 30% higer activity than wild-type PjTA with both of the tested substrates. Although position 393 is distant from the active site, the replacement of alanine by arginine seems to indirectly influence the catalytic activity. Leferink et al. (83) showed that mutations distant from the active site of a copper-containing nitrite reductase resulted in altered catalytic activity due to minor structural perturbations. Variant P9A, also carrying a substitution distant from the active site, showed a modest increase in activity towards (S)-α-methylbenzylamine. As described above, mutations located far from the active site can influence the activity of the enzymes through minor structural changes.
The work reported here demonstrates that this computational-based point mutation strategy has the potential to improve PjTA thermostability and that the use of high-temperature MD simulations is a promising orthogonal method for improving library quality. Nevertheless, in terms of absolute values, both the number of positive mutations and their contribution to thermostability was very modest as compared to results with other proteins. For example, Arabnejad et al. (39) working with a tetrameric enzyme reported ∆Tmapp between 1 and 13.5˚C. Recent results in our laboratory (84) and reported in the literature by Börner et al. (54, 85) indicate that mutations at the subunit interfaces and mutations influencing cofactor binding may have a large and positive effect on stability of related PLP fold type I transaminases. Dissociation of dimeric or
4
tetrameric transaminases leads to inactivation, denaturation, and release of the cofactor pyridoxal phosphate (86, 87), which results in aminotransferases being active just in the multimeric state (88). Further attempts to engineer the stability of PjAT should thus be focused on improving interactions at the dimer interface or with the cofactor. Mutations in these regions may deliver stabilized proteins in which new weak spots appear, maybe on the protein surface, in which case introduction of the surface mutations examined here may have a larger positive effect.
In conclusion, the current work suggests that introduction of stabilizing mutations in priority regions identified by high temperature MD may become a useful addition to the FRESCO strategy. Its impact on the engineering of robust enzymes is probably much larger in case of monomeric proteins of low stability as compared to stabilization of multimeric enzymes where irreversible denaturation may be governed less by local unfolding than by global dissociation events.
AUTHOR CONTRIBUTIONS
All authors designed experiments and contributed to the interpretation of the data. H.J.W. performed computational design and simulations. C.M.P. constructed mutans and expressed and characterized the mutant enzymes. C.M.P., H.W., and D.B.J. wrote the manuscript.
REFERENCES
1. Mozhaev V V. 1993. Mechanism-based strategies for protein thermostabilization. Trends Biotechnol 11:88–95. 2. Vieille C, Zeikus GJ. 2001. Hyperthermophilic enzymes: Sources, uses, and molecular mechanisms for thermostability.
Microbiol Mol Biol Rev 65:1–43.
3. Wijma HJ, Floor RJ, Jekel PA, Baker D, Marrink SJ, Janssen DB. 2014. Computationally designed libraries for rapid enzyme stabilization. Protein Eng Des Sel 27:49–58.
4. Turner P, Mamo G, Karlsson E. 2007. Potential and utilization of thermophiles and thermostable enzymes in biorefining. Microb Cell Fact 6:1–23.
5. Bommarius AS, Paye MF. 2013. Stabilizing biocatalysts. Chem Soc Rev 42:6534–6565.
6. Emond S, André I, Jaziri K, Potocki-Véronèse G, Mondon P, Bouayadi K, Kharrat H, Monsan P, Remaud-Simeon M. 2008. Combinatorial engineering to enhance thermostability of amylosucrase. Protein Sci 17:967–976.
7. Lehmann M, Wyss M. 2001. Engineering proteins for thermostability: The use of sequence alignments versus rational design and directed evolution. Curr Opin Biotechnol 12:371–375.
8. Burg B Van Den, Eijsink VGH. 2002. Selection of mutations for increased protein stability. Protein Technol Commer Enzym 333–337.
9. Eijsink VGH, GÅseidnes S, Borchert T V., Van Den Burg B. 2005. Directed evolution of enzyme stability. Biomol Eng 22:21–30.
10. Giver L, Gershenson A, Freskgard PO, Arnold FH. 1998. Directed evolution of a thermostable esterase. Proc Natl Acad Sci U S A 95:12809–12813.
11. Chen K, Robinson AC, Dam ME Van, Martinez P, Economou C, Arnold FH. 1991. Enzyme engineering for nonaqueous solvents . II . Additive effects of mutations on the stability and activity of subtilisin E in polar organic media. Biotechnol Prog 7:125–129.
12. Schmidt-Dannert C, Arnold FH. 1999. Directed evolution of industrial enzymes. Trends Biotechnol 17:135–136. 13. Stemmer WP. 1994. Rapid evolution of a protein in vitro by DNA shuffling. Nature 370:389–391.
14. van Loo B, Lutje Spelberg JH, Kingma J, Sonke T, Wubbolts MG, Janssen DB. 2004. Directed evolution of epoxide hydrolase from A. radiobacter toward higher enantioselectivity by error-prone PCR and DNA shuffling. Chem Biol 11:981–990.
15. Bommarius AS, Broering JM, Chaparro-Riggers JF, Polizzi KM. 2006. High-throughput screening for enhanced protein stability. Curr Opin Biotechnol 17:606–610.
16. Magliery TJ, Lavinder JJ, Sullivan BJ. 2011. Protein stability by number: High-throughput and statistical approaches to one of protein science’s most difficult problems. Curr Opin Chem Biol 15:443–451.
17. Porter JL, Rusli RA, Ollis DL. 2016. Directed evolution of enzymes for industrial biocatalysis. ChemBioChem 17:197– 203.
18. Van den Burg B, Vriend G, Veltman OR, Venema G, Eijsink VG. 1998. Engineering an enzyme to resist boiling. Proc Natl Acad Sci U S A 95:2056–2060.
19. Bryan PN. 2000. Protein engineering of subtilisin. Biochim Biophys Acta - Protein Struct Mol Enzymol 1543:203–222. 20. Eijsink VGH, Bjørk A, Gåseidnes S, Sirevåg R, Synstad B, Burg B Van Den, Vriend G. 2004. Rational engineering of
enzyme stability. J Biotechnol 113:105–120.
21. Wijma HJ, Janssen DB. 2013. Computational design gains momentum in enzyme catalysis engineering. FEBS J 280:2948–2960.
22. Lehmann M, Pasamontes L, Lassen SF, Wyss M. 2000. The consensus concept for thermostability engineering of proteins. Biochim Biophys Acta - Protein Struct Mol Enzymol 1543:408–415.
23. Bornscheuer UT, Pohl M. 2001. Improved biocatalysts by directed evolution and rational protein design. Biocatal Biotransformation 5:137–143.
24. Soskine M, Tawfik DS. 2010. Mutational effects and the evolution of new protein functions. Nat Publ Gr 11:572–582. 25. Schoemaker HE, Mink D, Wubbolts MG. 2003. Dispelling the myths - Biocatalysis in industrial synthesis. Science
299:1694–1698.
26. Pantoliano MW, Whitlow M, Wood JF, Dodd SW, Hardman KD, Rollence ML, Bryan PN. 1989. Large increases in general stability for subtilisin BPN’ through incremental changes in the free energy of unfolding. Biochemistry 28:7205–13.
27. Serrano L, Day AG, Fersht AR. 1993. Step-wise mutation of barnase to binase. A procedure for engineering increased stability of proteins and a experimental analysis of the evolution of protein stability. J Mol Biol 233:305–312.
28. von der Osten C, Branner S, Hastrup S, Hedegaard L, Rasmussen MD, Bisgard-Frantzen H, Carlsen S, Mikkelsen JM. 1993. Protein engineering of subtilisins to improve stability in detergent formulations. J Biotechnol 28:55–68. 29. Akasako A, Haruki M, Oobatake M, Kanaya S. 1995. High resistance of Escherichia coli ribonuclease HI variant with
quintuple thermostabilizing mutations to thermal denaturation, acid denaturation, and proteolytic degradation. Biochemistry 34:8115–8122.
30. Shih P, Kirsch JF. 1995. Design and structural analysis of an engineered thermostable chicken lysozyme. Protein Sci 4:2063–2072.
31. Floor RJ, Wijma HJ, Colpa DI, Ramos-Silva A, Jekel PA, Szymánski W, Feringa BL, Marrink SJ, Janssen DB. 2014. Computational library design for increasing haloalkane dehalogenase stability. ChemBioChem 15:1660–1672. 32. Arnold FH. 2001. Combinatorial and computational challenges for biocatalyst design. Nature 409:253–257. 33. Lippow SM, Tidor B. 2007. Progress in computational protein design. Curr Opin Biotechnol 18:305–311. 34. Korkegian A. 2005. Computational thermostabilization of an enzyme. Science 308:857–860.
35. Kang SG, Saven JG. 2007. Computational protein design: structure, function and combinatorial diversity. Curr Opin Chem Biol 11:329–334.
36. Wijma HJ, Fürst MJLJ, Janssen DB. 2018. A computation library design protocol for rapid improvement of protein stability: FRESCO, p. 69–85. In Protein Engineering.
37. Vihinen M. 1987. Relationship of protein flexibility of thermostability. Protein Eng 1:477–480.
38. Kiss G, Çelebi-Ölçüm N, Moretti R, Baker D, Houk KN. 2013. Computational enzyme design. Angew Chemie - Int Ed 52:5700–5725.
39. Arabnejad H, Dal Lago M, Jekel PA, Floor RJ, Thunnissen, Andy-Mark W. H. Terwisscha van Scheltinga A, Wijma HJ, Janssen DB. 2016. A robust cosolvent-compatible halohydrin dehalogenase by computational library design. Protein Eng Des Sel 1–15.
40. Wu B, Wijma HJ, Song L, Rozeboom HJ, Poloni C, Tian Y, Arif MI, Nuijens T, Quaedflieg PJLM, Szymanski W, Feringa BL, Janssen DB. 2016. Versatile peptide C-terminal functionalization via a computationally engineered peptide amidase. ACS Catal 6:5405–5414.
41. Martin C, Maqueo AO, Wijma HJ, Fraaije MW. 2018. Biotechnology for biofuels creating a more robust 5 - hydroxymethylfurfural oxidase by combining computational predictions with a novel effective library design. Biotechnol Biofuels 11:1–9.
42. Bu Y, Cui Y, Peng Y, Hu M, Tian Y, Tao Y, Wu B. 2018. Engineering improved thermostability of the GH11 xylanase from
Neocallimastix patriciarum via computational library design. Appl Microb Biotechnol 102:3675–3685.
43. Taylor PP, Pantaleone DP, Senkpeil RF, Fotheringham IG. 1998. Novel biosynthetic approaches to the production of unnatural amino acids using transaminases. Trends Biotechnol 16:412–418.
44. Nestl BM, Nebel BA, Hauer B. 2011. Recent progress in industrial biocatalysis. Curr Opin Chem Biol 15:187–193. 45. Jiang J, Chen X, Feng J, Wu Q, Zhu D. 2014. Substrate profile of an ω-transaminase from Burkholderia vietnamiensis and
its potential for the production of optically pure amines and unnatural amino acids. J Mol Catal B Enzym 100:32–39. 46. Bea H-S, Park H-J, Lee S-H, Yun H. 2011. Kinetic resolution of aromatic β-amino acids by ω-transaminase. Chem
Commun (Camb) 47:5894–5896.
47. Koszelewski D, Grischek B, Glueck SM, Kroutil W, Faber K. 2011. Enzymatic racemization of amines catalyzed by enantiocomplementary ω-transaminases. Chemistry 17:378–383.
48. Hou A, Deng Z, Ma H, Liu T. 2016. Substrate screening of amino transaminase for the synthesis of a sitagliptin intermediate. Tetrahedron 72:4660–4664.
49. Desai AA. 2011. Sitagliptin manufacture: A compelling tale of green chemistry, process intensification, and industrial asymmetric catalysis. Angew Chem Int Ed 50:1974–1976.
50. Otzen M, Palacio CM, Janssen DB. 2018. Characterization of the caprolactam degradation pathway in Pseudomonas
jessenii using mass spectrometry-based proteomics. Appl Microb Biotechnol 102:6699-6711
51. Dahlhoff G, Niederer JPM, Hoelderich WF. 2001. ε-Caprolactam: New by-product free synthesis routes. Catal Rev Sci Eng 43:381–441.
52. Turk SCHJ, Kloosterman WP, Ninaber DK, Kolen KPAM, Knutova J, Suir E, Schürmann M, Raemakers-Franken PC, Müller M, De Wildeman SMA, Raamsdonk LM, Van Der Pol R, Wu L, Temudo MF, Van Der Hoeven RAM, Akeroyd M, Van Der Stoel RE, Noorman HJ, Bovenberg RAL, Trefzer AC. 2016. Metabolic engineering toward sustainable production of nylon-6. ACS Synth Biol 5:65–73.
53. Buß O, Muller D, Jager S, Rudat J, Rabe KS. 2018. Improvement in the thermostability of a β-amino acid converting ω-transaminase by using FoldX. ChemBioChem 19:379–387.
54. Börner T, Bartsch S, Vogel A, Adlercreutz P, Grey C. 2017. Three in one: temperature, solvent and catalytic stability by engineering the cofactor-binding element of amine transaminase. ChemBioChem 18:1482–1486.
55. Palacio CM, Rozeboom HJ, Lanfranchi E, Otzen M, Janssen DB. 2018. Biochemical properties of a Pseudomonas aminotransferase involved in caprolactam metabolism. Submitted for publication.
56. Palacio CM, Crismaru CG, Bartsch S, Navickas V, Ditrich K, Breuer M, Abu R, Woodley J, Baldenius K, Wu B, Janssen DB. 2016. Enzymatic network for production of ether amines from alcohols. Biotechnol Bioeng 113:1853–1861. 57. Ericsson UB, Hallberg BM, DeTitta GT, Dekker N, Nordlund P. 2006. Thermofluor-based high-throughput stability
optimization of proteins for structural studies. Anal Biochem 357:289–298.
58. Boivin S, Kozak S, Meijers R. 2013. Optimization of protein purification and characterization using Thermofluor screens. Protein Expr Purif 91:192–206.
59. Krieger E, Nielsen JE, Spronk CAEM, Vriend G. 2006. Fast empirical pKa prediction by Ewald summation. J Mol Graph Model 25:481–486.
60. Krieger E, Vriend G. 2015. New ways to boost molecular dynamics simulations. J Comput Chem 36:996–1007. 61. Genheden S, Ryde U. 2011. A comparison of different initialization protocols to obtain statistically independent
molecular dynamics simulations. J Comput Chem 32:187–195.
62. Caves LSD, Evanseck JD, Karplus M. 1998. Locally accessible conformations of proteins: Multiple molecular dynamics simulations of crambin. Protein Sci 7:649–666.
63. Genheden S, Diehl C, Akke M, Ryde U. 2010. Starting-condition dependence of order parameters derived from molecular dynamics simulations. J Chem Theory Comput 6:2176–2190.
64. Zagrovic B, Gunsteren WF Van. 2007. Computational analysis of the mechanism and thermodynamics of inhibition of phosphodiesterase 5A by synthetic ligands. J Chem Theory Comput 3:301–311.
65. Vriend G, Sander C. 1991. Detection of common three-dimensional substructures in proteins. Proteins Struct Funct Bioinforma 58:52–58.
66. Floor RJ, Wijma HJ, Jekel PA, Terwisscha van Scheltinga AC, Dijkstra BW, Janssen DB. 2015. X-ray crystallographic validation of structure predictions used in computational design for protein stabilization. Proteins Struct Funct Bioinforma 83:940–951.
67. Arabnejad H, Lago MD, Jekel PA, Floor RJ, Thunnissen AWH, Scheltinga ACT Van, Wijma HJ, Janssen DB. 2017. A robust cosolvent-compatible halohydrin dehalogenase by computational library design. Protein Eng Des Sel 30:175– 189.
68. Kellogg EH, Leaver-Fay A, Baker D. 2011. Role of conformational sampling in computing mutation-induced changes in protein structure and stability. Proteins 79:830–838.
69. Guerois R, Nielsen JE, Serrano L. 2002. Predicting changes in the stability of proteins and protein complexes: A study of more than 1000 mutations. J Mol Biol 320:369–387.
70. Krieger E, Darden T, Nabuurs SB, Finkelstein A, Vriend G. 2004. Making optimal use of empirical energy functions: Force-field parameterization in crystal space. Proteins Struct Funct Bioinforma 57:678–683.
71. Bednar D, Beerens K, Sebestova E, Bendl J, Khare S, Chaloupkova R, Prokop Z, Brezovsky J, Baker D, Damborsky J. 2015. FireProt: Energy- and evolution-based computational design of thermostablemultiple-point mutants. PLOS Comput Biol 11:1–20.
72. Reetz MT, Soni P, Fernández L. 2008. Knowledge-guided laboratory evolution of protein thermolability. Biotechnol Bioeng 102:1712–1717.
73. Reetz MT, Soni P, Fernáandez L, Gumulya Y, Carballeira J. 2010. Increasing the stability of an enzyme toward hostile organic solvents by directed evolution based on iterative saturation mutagenesis using the B-FIT method. Chem Comm 46:8657–8658.
74. Pikkemaat MG, Linssen ABM, Berendsen HJC, Janssen DB. 2002. Molecular dynamics simulations as a tool for improving protein stability. Protein Eng 15:185–192.
75. Meharenna YT, Poulos TL. 2010. Using molecular dynamics to probe the structural basis for enhanced stability in thermal stable cytochromes P450. Biochemistry 49:6680–6686.
76. Caspi R, Altman T, Billington R, Dreher K, Foerster H, Fulcher CA, Holland TA, Keseler IM, Kothari A, Kubo A, Krummenacker M, Latendresse M, Mueller LA, Ong Q, Paley S, Subhraveti P, Weaver DS, Weerasinghe D, Zhang P, Karp PD. 2014. The MetaCyc database of metabolic pathways and enzymes and the BioCyc collection of Pathway/ Genome Databases. Nucleic Acids Res 42:459–471.
77. Broom A, Jacobi Z, Trainor K, Meiering EM. 2017. Computational tools help improve protein stability but with a solubility tradeoff. J Biol Chem 292:14349–14361.
78. Poole LB, Loveys D a, Hale SP, Gerlt J a, Stanczyk SM, Bolton PH. 1991. Deletion of the omega-loop in the active site of staphylococcal nuclease. 1. Effect on catalysis and stability. Biochemistry 30:3621–7.
79. Gleason FK. 1992. Mutation of conserved residues in Escherichia coli thioredoxin: Effects on stability and function. Protein Sci 1:609–616.
80. Shoichet BK, Baase WA, Kuroki R, Matthews BW. 1995. A relationship between protein stability and protein function. Proc Natl Acad Sci 92:452–456.
81. Jacob J, Duclohier H, Cafiso DS. 1999. The role of proline and glycine in determining the backbone flexibility of a channel-forming peptide. Biophys J 76:1367–1376.
82. Gribenko A V, Patel MM, Liu J, McCallum SA, Wang C, Makhatadze GI. 2009. Rational stabilization of enzymes by computational redesign of surface charge-charge interactions. Proc Natl Acad Sci 106:2601–2606.
83. Leferink NGH, Antonyuk S V., Houwman JA, Scrutton NS, Eady RR, Hasnain SS. 2014. Impact of residues remote from the catalytic centre on enzyme catalysis of copper nitrite reductase. Nat Commun 5:1–8.
84. Q. Meng, C. Palacio, H.J. Wijma EL. Robust ω-transaminases by computational stabilization of subunit interfaces. In preparation.
85. Börner T, Rämisch S, Reddem ER, Bartsch S, Vogel A, Thunnissen A-MWH, Adlercreutz P, Grey C. 2017. Explaining operational instability of amine transaminases: Substrate-induced inactivation mechanism and influence of quaternary structure on enzyme–cofactor intermediate stability. ACS Catal 7:1259–1269.
86. Kishimoto K, Yasuda C, Manning JM. 2000. Reversible dissociation/association of D-amino acid transaminase subunits: Properties of isolated active dimers and inactive monomers. Biochemistry 39:381–387.
87. Polyanovsky OL. 1965. Reversible dissociation of succinylated aspartate transaminase into subunits. Biochem Biophys Res Commun 19:364–370.
88. Chen A, Li Y, Nie J, McNeil B, Jeffrey L, Yang Y, Bai Z. 2015. Protein engineering of Bacillus acidopullulyticus pullulanase for enhanced thermostability using in silico data driven rational design methods. Enzyme Microb Technol 78:74–83.
