Robust design and optimization of retroaldol enzymes

Abstract

Enzyme catalysts of a retroaldol reaction have been generated by computational design using a motif that combines a lysine in a nonpolar environment with water-mediated stabilization of the carbinolamine hydroxyl and β-hydroxyl groups. Here, we show that the design process is robust and repeatable, with 33 new active designs constructed on 13 different protein scaffold backbones. The initial activities are not high but are increased through site-directed mutagenesis and laboratory evolution. Mutational data highlight areas for improvement in design. Different designed catalysts give different borohydride-reduced reaction intermediates, suggesting a distribution of properties of the designed enzymes that may be further explored and exploited.

Introduction

De novo enzyme design is in its infancy. Active designs have been generated for several reactions, but the success rate is low.^1–4 Other approaches to designing catalysts, for example catalytic antibody selections, have also typically generated few active catalysts when successful.^{5, 6} While successes in enzyme design are encouraging, the ability to produce active designs robustly and reliably is clearly important for future applications of this approach.

We previously experimented with four different catalytic motifs for a simple retroaldol reaction [Fig. 1(A) and Supporting Information Fig. S1].¹ While little success was found for motifs involving extensive charged side chain networks analogous to those observed in naturally occurring (highly evolved) enzymes, active enzymes were successfully produced with a simpler design involving a lysine side chain within a hydrophobic binding pocket with a water molecule positioned to facilitate proton transfers involving the β-alcohol and the hydroxyl group of the carbinolamine reaction intermediate [Fig. 1(B) and Supporting Information S2]. The reaction utilizes a Schiff base from the lysine side chain, which acts as an electron sink during C➖C bond cleavage (Supporting Information Fig. S2).

Figure 1.

A. The retroaldol reaction of methodol. B. Schematic representation of the active site design incorporated into the protein scaffolds C. The 13 different scaffold backbones and 30 different catalytic lysine positions utilized by the active designs. Catalytic lysines are shown in pink.

In this article, we investigate the robustness with which new retroaldolase enzymes can be designed using this catalytic motif and assess the extent to which the designs can be improved by site-directed mutagenesis and laboratory evolution.

Results and Discussion

Computational design

To investigate the robustness of designs incorporating the catalytic motif described above we used RosettaMatch⁷ to find sites in scaffolds of known structure at which the active site (catalytic side chains and composite substrate/intermediate/transition state) could be placed to realize the specified catalytic geometry without steric clashes with the scaffold backbone. For each potential catalytic site found, Rosetta Design⁸ was used to optimize the surrounding side chains for favorable interactions with the substrate/transition state model. Compared to our previous study, a larger number of side-chain rotamers of the catalytic and binding-interaction residues was used during the computational design step. This increased diversity allowed finer sampling of the torsion angles and better designed binding and hydrogen bonding interactions to be identified, as one of the concerns about some of the designs from the previous set was a general “underpacking” of the active site. To favor designs with correctly preorganized active sites, the side chains in the designs were repacked and minimized in the presence and absence of the transition state (TS) model of the reaction to verify that the designed side chains do not move significantly from their intended location, and the catalytic lysine was required to be well packed by the surrounding residues. The designs with the best-packed catalytic residues and the most favorable calculated binding energy were selected for experimental characterization.

Genes were synthesized for 42 new designs in 13 different protein scaffolds at 30 different lysine positions [Fig. 1(C) and Supporting Information S3], and the proteins were expressed in E. coli, purified over nickel-NTA, and tested using a fluorometric retroaldolase assay.¹ All designs were expressed and soluble in E. coli. This solubility success rate is significantly higher than in our other enzyme design work,⁹ and may result from our use of thermostable starting scaffolds, the filtering strategy, and reversion of substitutions to the native residue where compatible with the designed active site (see Supporting Information).

Thirty-three of the 42 designs had rates more than 10-fold above the uncatalyzed rate in buffer, and 29 had v₀/v_uncat greater than 100-fold with 5–10 μM enzyme (Table I and Supporting Information Table SI). The active designs are based on a wide variety of different scaffolds. The range of scaffold folds and positions of the Schiff base-forming lysine are illustrated in Figure 1(C). Together with the active designed aldolases in our previous paper, there are a total of 65 active designs on 14 different protein backbone scaffolds (Supporting Information Fig. S5).

Table I.

Rate Enhancements of New Retroaldolase Designs

Design	Scaffold	Fold	v0/vuncat
RA113	1v04	Beta Barrel	490
RA77	1m4w	Jelly roll	86
RA83	1f5j	Jelly roll	140
RA84	1m4w	Jelly roll	180
RA81	1m4w	Jelly roll	250
RA80	1pvx	Jelly roll	280
RA85	1pvx	Jelly roll	300
RA79	1f5j	Jelly roll	310
RA76	1m4w	Jelly roll	410
RA78	1pvx	Jelly roll	550
RA75	1f5j	Jelly roll	770
RA72	1m4w	Jelly roll	1420
RA82	1m4w	Jelly roll	1970
RA89	3b5l	Jelly roll	3900
RA86	3b5l	Jelly roll	6900
RA111	1sjw	NTF2	130
RA110	1oho	KSI/NTF2-like	4300
RA112	1ilw	Rossman Fold	410
RA101	1thf	TIM barrel	57
RA93	1lbl	TIM barrel	82
RA100	3hoj	TIM barrel	100
RA99	1lbf	TIM barrel	130
RA97	2c3z	TIM barrel	170
RA94	1lbl	TIM barrel	250
RA96	1lbl	TIM barrel	290
RA71	1dl3	TIM barrel	290
RA92	1a53	TIM barrel	360
RA98	1a53	TIM barrel	440
RA67	1a53	TIM barrel	640
RA102	1thf	TIM barrel	660
RA90	1a53	TIM barrel	830
RA95	1lbl	TIM barrel	6400
RA91	1lbl	TIM barrel	7300

Assays were carried out at a substrate concentration of 435 μM and enzyme concentration of 5.0 μM in 25 mM HEPES, 2.7% CH₃CN, 100 mM NaCl (pH 7.5). k_uncat is 3.9 × 10⁻⁷ min⁻¹.¹⁰

The success rate (75% successful designs) is higher than the success rate in our original study (44%). Our initial design calculations in the original study [using motif I from Fig. 2(C) in Ref.1] were aimed at stabilizing the transition state for the C➖C bond breaking reaction. The low success with this motif may have been due to too great of a focus on the bond-breaking step resulting in too little space near the imine carbon to accommodate the carbinolamine alcohol.¹ In the current study, our designs ensured both sufficient space in the active site and an appropriate polar environment to favor carbinolamine formation and subsequent dehydration (Supporting Information Figs. S3 and S4). Success in enzyme design requires consideration of all steps in the reaction cycle, even those that are not rate limiting for the nonenzymatic reaction—this is one of the features that makes enzyme design so challenging.

Figure 2.

Comparison of active and inactive retroaldolase designs with identical catalytic lysine positions. The catalytic lysine and Glu (or Asp), used for the acid-base catalysis, are shown and the TS corresponding to each design is shown in the same color. A. RA76 (salmon) is an active design on scaffold 1m4w, a jelly roll; whereas RA77 (orange) is a catalytically inactive design with lysine in the same position in the scaffold. [The designs differ at 11 positions around the active site.] B. RA91 (cyan) and RA92 (magenta) are active enzymes in TIM barrel scaffold 1a53/1lbl; whereas RA93 (green) is an inactive design. [The designs differ at 19 positions around the active site.]

A remaining open question is what differentiates the designs with activity from those that are inactive. Of particular interest are subsets of designs in which the catalytic lysine is in the identical position on the same scaffold; several examples are shown in Figure 2. The binding pockets in the active designs may more correctly position the retroaldol substrate for reaction with the active site lysine, the lysine residues may be more activated for catalysis, or the active sites in the inactive designs may not be properly formed. We cannot currently distinguish between the active and inactive designs and this is clearly a challenge for future work. Supporting Information Table SI shows the contributions of the primary energetic terms in the Rosetta Design energy function for all of the working and nonworking designs. The results described below show that the differences responsible for the differences in activity may be fairly subtle as amino acid changes near the binding pocket can produce significant changes in activity.

Experimental active site optimization

Despite our high success rate in designing retroaldolase enzymes with measurable catalytic activity, the absolute activities of the individual enzymes are low relative to naturally occurring enzymes, with k_cat/K_M values of less than 1 M⁻¹ s⁻¹. It was found previously that the activity of a designed Kemp eliminase could be improved by directed evolution.² To learn more about the properties of the designed retroaldolase enzymes and the optimization of active site residues, we systematically mutated residues at or near the active site in a subset of the designs. Among the 65 active designed aldolases so far, we chose to focus on designs in three distinct folds, the jelly roll (RA60), the NTF2 fold (RA110), and the TIM barrel (RA45 and RA95). Active designs were found in the same TIM barrel scaffold with the catalytic lysine in 10 different positions, and we selected designs with two different catalytic lysine positions for sequence optimization.

For each of the four designs, Kunkel mutagenesis¹¹ with degenerate primers was used to generate individual amino acid substitutions at each position at or near the active site. We chose to mutate residues one at a time rather than simultaneously to facilitate correlation of activity changes with sequence changes. To increase solubility and expression, we also tested substitutions at surface positions to consensus residues frequently occurring in the protein family.¹² Variants were first screened in a plate assay, and mutants with increased or significantly decreased activity were subsequently expressed, purified, and assayed to confirm the lysate assay results (Supporting Information Tables SVII–SXIII). Mutations that increased activity 1.5- to 5-fold (Fig. 3) were combined by Kunkel mutagenesis and tested for additive effects.

Figure 3.

Models of improved active sites. For simplicity, only the carbinolamine intermediate is shown, rather than the transition state/intermediate ensemble used in the design calculations. Designed residues are in cyan sticks, and those changed during optimization are in purple. The carbinolamine and the catalytic lysine are in yellow. The improved enzyme designs are: A. RA45.2 from a TIM barrel fold, B. RA60.2 from a jelly roll fold, C. RA95.4 from a TIM barrel fold, D. RA110.4 from an NTF2 fold.

We selected several of the most active of the improved designs in each of the scaffolds for further characterization. All displayed Michaelis–Menten kinetics (Table II). With the exception of RA60.2 [Fig. 3(B)], which has a relatively exposed predicted substrate binding site, K_M values for the new variants range from 300 to 450 μM. The k_cat/k_uncat values range from 3 × 10⁴ to 2 × 10⁵, and the apparent second-order rate constants (k_cat/K_M values) for these enzymes are 10⁵-fold to 10⁶-fold above the second-order rate constant for the corresponding reaction with butylamine free in solution at pH 7.5.¹³ The cycles of site-directed mutation and combination improved catalytic efficiency (k_cat/K_M) for the four designs from a minimum of 7-fold up to 88-fold (Table II).

Table II.

Enzymatic Activity of Five Retroaldolase Designs Before and After Optimization^a,b

	KM (μM)	kcat (min−1)	kcat/kuncat	kcat/KM (M−1 s−1)	Fold increase in kcat/KM
RA60design	510	0.0093	2.4 × 104	0.27 ± 0.07	—
RA60.2improved	660	0.070	1.8 × 105	1.8 ± 0.2	7
RA110design	1600	0.005	1.2 × 104	0.048 ± 0.005	—
RA110.4improved	278	0.070	1.8 × 105	4.2 ± 0.2	88
RA110.4-6evolved	69	0.230	5.9 × 105	55 ± 6	1100
RA34design	630	0.0073	1.9 × 104	0.19 ± 0.01	—
RA34.6improved	30	0.022	5.5 × 104	12 ± 1	63
RA45design	800	0.0017	0.4 × 104	0.036 ± 0.003	—
RA45.2improved	439	0.0139	3.6 × 104	0.54 ± 0.02	15
RA45.2-10evolved	80	0.230	5.8 × 105	47 ± 3	1300
RA95design	540	0.0020	4.8 × 103	0.053 ± 0.009	—
RA95.4improved	310	0.015	3.4 × 104	0.84 ± 0.10	16
catAB 38C2	24 ± 1	0.71 ± 0.01	1.8 × 106	490	—

^aActivity was measured in 25 mM HEPES, 2.7% CH₃CN, 100 mM NaCl (pH 7.5). k_uncat is 3.9 × 10⁻⁷ min⁻¹. At concentrations above 500 μM substrate, solubility limits accurate determination of K_M. For comparison, the catalytic antibody 38C2¹⁰ was assayed under the same conditions. Enzyme sequences are in the Supporting Information.

^bBecause of slow reactions and complications of the fluorescence assay,^{1, 13} initial rates were determined before the first turnover of enzyme. For RA34.6, RA110.4, and RA95.5, single-turnover rates did not differ significantly from multiple-turnover rates under saturating conditions (see Supporting Information).

The results suggest several routes for improvement of retroaldolase activity in the designed enzymes. In the RA95 TIM barrel design, many of the favorable mutations involve increases in side chain volume and hydrophobicity, including T51Y, T83K, S110H, M180F, and R182M [Fig. 3(C) and Supporting Information Table SX]. Small-to-large mutations also increased activity in the jelly roll design, RA60, including S87W, A174M, V176I, and V178H (Supporting Information Table SIX). In contrast, many improvements to the RA45 design arose from large-to-small mutations including W8A/T/V, F133L, V159C, and R182V/I (Supporting Information Tables SVII and SVIII). These observations suggest that hydrophobic packing or positioning of the substrate was not optimal in the initial designs, and the substrate may be underpacked in two of the three cases investigated. The most significant increases in activity resulted from increasing the size of hydrophobic side chains predicted to surround the catalytic lysine; these mutations may better position the catalytic residue or increase its reactivity. The activity of RA95 was also increased by introducing lysine substitutions in nearby positions; these could shift the pK_a of the active site lysine,^{10, 13, 14} buttress the active site through hydrogen bonding or packing, or perhaps contribute to Schiff base formation directly.

Directed evolution

As the foregoing analysis makes apparent, computational design can successfully generate active enzymes in a wide range of scaffolds and these nascent enzymes are amenable to improvement. Further optimization through directed evolution can afford additional increases in activity. The improved RA110 and RA45 catalysts, RA110.4 and RA45.2, were chosen as starting points for in vitro evolution because of their different sizes, folds, and activities. The genes encoding these enzymes were diversified by error-prone PCR¹⁵ and DNA shuffling¹⁶ and more active variants identified in a plate assay. After six rounds of mutagenesis and screening, the activity of the RA110 variants in crude lysate further increased ∼7-fold, reflecting a >10-fold increase in k_cat/K_M but a twofold drop in protein yield. RA45 activity improved 700-fold over 10 evolutionary cycles, arising from a 100-fold increase in k_cat/K_M and a 10-fold increase in the yield of soluble protein.

Representative catalysts from the final round of directed evolution were chosen for more detailed characterization. Variants RA110.4-6 and RA45.2-10, for example, contain 8 and 14 new mutations, respectively (Fig. 4). As reported in Table II, these substitutions augment catalytic activity by simultaneously increasing k_cat and decreasing K_M. The cumulative >10³-fold increase in efficiency over the original computational designs from active site mutagenesis and directed evolution renders these enzymes comparable to catalytic antibody 38C2,¹⁰ one of the most active artificial aldolases reported to date, which promotes the same transformation with only a threefold higher k_cat and a 10-fold higher k_cat/K_M (Table II).

Figure 4.

Mutational history of RA110 and RA45. Panels A and B respectively show a model of RA110_design and a close-up of its active site; panels C and D show corresponding views of the RA45_design and its active site. The modeled carbinolamine intermediate is colored orange. Mutations introduced by Rosetta are depicted in cyan; residues that were altered during the active site optimization procedure are magenta; and residues that were substituted during directed evolution are shown as spheres (colored green if the site was not first mutated by design or cassette mutagenesis). In RA45, residue Trp8 (magenta spheres) was first mutated to valine during active site optimization and subsequently to arginine by directed evolution. Only residues that deviate from the starting designs are labeled.

As is often observed in directed evolution experiments, mutations in the evolved enzymes are scattered throughout the protein. In the case of RA110.^4–6, most of the substitutions are located at or near the surface of the protein [Fig. 4(A)], distant from the substrate binding pocket (e.g., Q59K, C69Y, S77N, A83T, P85L, H110Q, S126C); only one mutated amino acid (V38I) is in direct contact with putative active site residues [Fig. 4(B)]. In contrast, the mutations in RA45.2-10 [Fig. 4(C)] include three active site residues [V8R, L51M, V89M, Fig. 4(D)], four second-shell residues (L15Q, L131P, N161S, Y234F), three buried residues (L96Q, L108M, L146M), and four surface residues (R64H, E143D, N217S, K242E). It is noteworthy that four of these substitutions targeted sites that were originally modified by the design algorithm (Residues 51, 89, and 234) or in the subsequent active site optimization step (Residue 8). At least some of the changes remote from the active site presumably fine-tune the conformation of the binding pocket through subtle structural adjustments.

It is instructive to compare the designed retroaldolases with previously identified aldol catalysts created by design or selection methods.^{14, 17, 18–26} Building on earlier work on cationic polypeptides that promote oxaloacetate decarboxylation,^17–21 Barbas et al. selected several catalytically active peptides from a designed phage library that utilize a reactive lysine for the retroaldol cleavage of methodol.²² The best catalyst had a rate acceleration k_cat/k_uncat = 1400, 10-times lower than two of the in silico designs and considerably less active than the evolved variants. Hilvert et al. similarly developed β-peptide catalysts for a related retroaldol reaction that exhibited a k_cat/k_uncat of ∼3000,²³ comparable to the starting RA45 and RA110 designs but again lower than the improved variants. In addition to possessing a catalytic lysine with a depressed pK_a, these cationic peptides exhibit saturation kinetics consistent with substrate binding. However, their high K_m values (∼2 mM) likely reflect the lack of a defined substrate-binding pocket, which may make further optimization difficult. In contrast, the commercially available catalytic antibody 38C2,²⁴ which contains a catalytic lysine at the bottom of a deep hydrophobic pocket,^{25, 26} is a substantially more active aldolase. It was generated by reactive immunization with a haptenic diketone and has a catalytic efficiency 10³ to 10⁴-fold higher than the cationic peptides or the starting computational designs (Table II). This difference may reflect the strong selection for binding inherent in the immune system along with limitations in our design algorithms. Although the mutagenesis results show that computation does not yet predict an optimal binding pocket, active site mutations and directed evolution reduce K_M, consistent with stronger binding interactions, and also increase k_cat. The most active of our retroaldolases have k_cat and K_M values each within approximately threefold of the catalytic antibody. Thus, the design and subsequent empirical optimization is approaching the efficiency of an artificial enzyme selected via binding rather than catalysis. Nevertheless, artificial enzymes obtained by both strategies are orders of magnitude less efficient than natural aldolases. Elucidating the origins of the factors that distinguish these systems remains an important challenge for bridging the gap between natural and man-made catalysts.

Trapping intermediates during catalysis

A challenge faced by natural enzymes, and encountered in the design of nonnatural enzymes, is the presence of multiple reaction steps, each of which needs to be catalyzed to achieve an overall observed rate enhancement. In nature, the slowest or most rate-limiting step experiences the highest selection pressure, leading to a natural leveling of reaction barriers in enzymes. To address how similar or distinct different designed enzymes behave with respect to individual reaction steps we used the classical approach of trapping imine-type intermediates via borohydride reduction.^{27, 28}

Sodium borohydride was added to reactions with RA110.4, RA95.4, and RA34.6,²⁹ an improved variant of the RA34 design described in the original paper,¹ at intervals following initiation of the reaction, and the samples were analyzed using mass spectrometry. As shown in Figure 4 and Supporting Information Figure S11, during the reaction cycle the catalytic lysine forms covalent interactions with both the substrate and the acetone product, and we observe peaks with mass-to-charge ratios consistent with these adducts for RA34.6 and RA95.4 along with an off-pathway enzyme-naphthaldehyde species, which was also noted in reactions of previously-designed retroaldolases^{13, 27, 28} (Fig. 5). For RA34.6, the peak for the enzyme-acetone species accumulates early in the reaction, whereas for RA95.4 the Schiff base adduct with substrate is detected; for RA110.4 neither peak accumulates significantly. Independent labeling experiments with the diketone inhibitor 2,4-pentanedione also suggest that the RA110.4 lysine is less reactive than its counterpart in RA34.6 and RA95.4 due to a similar activity pattern (data not shown). The three enzymes tested showed continuous reaction progress over multiple turnovers, suggesting that even for RA34.6 reaction rates are not significantly limited by buildup of the enzyme–acetone intermediate (Figs. S11 and S12).

Figure 5.

Covalent intermediates accumulating on the catalytic lysine during the retroaldol reaction. Fourier transform mass spectrometry spectrum for RA34.6, RA95.4, and RA110.4 at different reaction time points. Protein was first mixed with substrate for the indicated time (in min), then NaBH₄ was added at each time point before injection into the mass spectrometer. Peak 1 is the protein peak. Peak 2 is protein MW+42 Da, which corresponds to an acetone adduct on a Lys side chain. Peak 3 is protein MW+170 Da, which corresponds to the product 6-methoxy-2-naphthaldehyde bound to a Lys side chain. Peak 4 is protein MW+226 Da, which corresponds to the substrate on a Lys side chain.

The simplest explanations for the variance in the trapped species with the designed enzymes are differences in the rate constants for the individual reaction steps of Schiff base formation, carbon–carbon bond cleavage, and hydrolysis of the lysine-acetone product. Nevertheless, differences in the susceptibility of adducts of the designed enzymes to borohydride reduction and in their detection by mass spectrometry can also contribute. These results suggest that variation in individual reaction steps and/or active site properties readily arise despite the application of uniform design rules, and highlight the challenges associated with catalysis of a multistep reaction. These variations might be harnessed to explore features needed to optimally catalyze the multiple reaction steps and point to additional challenges for design improvement.

Conclusions

We have been able to robustly produce new retroaldolase catalysts from a large variety of scaffolds. Improvements in the design protocol include increasing the rotamer sampling around the active site to promote better binding and hydrogen bonding interactions, increasing the space surrounding the carbinolamine oxygen atoms and the polarity of the surrounding residues, and screening for preorganization of the active site. Nevertheless, the catalysts still have far lower activities than naturally occurring enzymes. One indication of the imperfect state of current design methodology is provided by the saturation mutagenesis data—many point mutations led to increased activity and thus should have been identified in the original design calculations. Similarly, we are unable to differentiate the active and inactive designs computationally, even in matched sets that employ the same catalytic lysine position. These data provide a benchmark for guiding the development and testing of improved computational design algorithms.

Given the low activity of the designed enzymes it is tempting to speculate that they resemble ancestors of modern-day enzymes. However, there are significant differences between the active sites of native and computationally designed enzymes. For Schiff-base forming aldolases,^{27, 28} the active site lysine is positioned by multiple groups making hydrogen bonds—not solely hydrophobic or packing interactions as in the designed enzymes; the carbinolamine oxygen interacts directly with side chains that may stabilize charge development and facilitate proton transfer—not a solvent-accessible space as in the designed enzymes; and the substrate is typically positioned by a combination of hydrophobic and hydrophilic interactions—not solely hydrophobic interactions as in the designed enzymes. Design of such complex interaction networks is an important current challenge for computational enzyme design.

Despite these differences between the designed enzymes and natural enzymes, site-directed mutagenesis and directed evolution led to substantial rate improvements (Table II). Nevertheless, we do not know the extent to which these primitive designs can be improved, whether there are evolutionary pathways leading to catalysts with true enzyme-like efficiency, or what distinctive structural features might emerge in the course of laboratory evolution. These are key questions for future research.

Materials and Methods

Computational design

The geometric search and hashing technique for identifying the potential active sites, RosettaMatch, and the subsequent design of the site to maintain catalytic residue positions and to adequately bind the ligand in the designed protein structure scaffolds for the composite transition state with multiple catalytic side chain possibilities was described previously⁷ and is described for this reaction in the Supporting Information Part I Section 1.

Experimental active site optimization

Saturation mutagenesis was performed at positions around the active site using degenerate primers and Kunkel mutagenesis.⁹ Resulting transformants, 96 for each position, were sequenced and representatives of every amino acid present at each position were assayed for activity. In 96-well plates, 1 mL cultures of the mutants were grown and expression induced before being pelleted. The cells were lysed via freeze-thaw and the resulting supernatant containing the retroaldol enzymes was assayed for catalytic activity. The 4-hydroxy-4-(6-methoxy-2-naphthyl)-2-butanone substrate was added and the enzyme activity was monitored for formation of the product, 6-methoxy-2-naphthaldehyde, by observing the fluorescence using λ_ex of 330 nm and λ_em of 452 nm. The rate of product formation for each of the mutants compared to the original design for RA45, RA60, RA95, and RA110 is reported in Supporting Information Part II, Tables SVII—XIII. The most improved catalytic activity mutants were combined using Kunkel mutagenesis. The best combinations were iteratively subjected to the same degenerate codon mutagenesis to search positions around the active site for improved activity until large improvements were no longer achieved. Additional details are provided in Supporting Information Part I Section 2 and Part II Section 2.

Enzyme purification and catalytic activity assay

Enzymes were expressed in BL21-star cells using autoinduction media and by growing transformants for either 8 h at 37°C or 24 h at 18°C. The cells were sonicated and the proteins purified over Qiagen Ni-NTA resin. After elution, the proteins were dialyzed at least three times into 100-fold excess 25 mM HEPES, 100 mM NaCl at pH 7.5 or run over a desalting column to remove the imidazole. For the catalytic assays, fluorescence of the product was measured with an excitation wavelength of 330 nm and an emission wavelength of 452 nm in either 96-well black flat bottom plates or quartz cuvettes and compared to known product concentration curves for quantitation. All measurements were done in triplicate and averaged. Steady state kinetic parameters k_cat and K_M were derived by fitting the experimental data to the Michaelis–Menten equation: v₀/[E] = k_cat[S]/(K_M+[S]), where v₀ is the initial rate, [E] is the enzyme concentration, and [S] is the substrate concentration. The kinetic parameters achieved using the fluorescence detection method was confirmed using a mass spectrometry-based measurement which demonstrated that the two values were very similar. Additional details are provided in the Supporting Information Part I Sections III, IV, and VII.

Trapping of reaction intermediates

10 μM RA34.6, RA95.4, and RA110.4 were incubated with 200 μM substrate in 25 mM HEPES (pH 7.5), 100 μM NaCl, 5% acetonitrile and after the indicated time, NaBH₄ was added to a final concentration of 8 mM for 1 min to trap reaction intermediates bound to the active site lysine. Samples were then injected into a small C4 column (BioBasic4 30x1), washed with 0.1% formic acid in water for 1 min, and eluted with a 5 min acetonitrile gradient (up to 5:95 H₂O:acetonitrile in 0.1% formic acid) into a Fourier transform mass spectrometer where the peaks were observed at an approximate elution time of 2.5 min. Further information is available in the Supporting Information Part I Section 6.

In vitro evolution experiments

DNA libraries were constructed for designs by random mutagenesis using error-prone PCR (epPCR). A low mutation rate was chosen for all libraries to generate on average one to two amino acid substitutions per protein. After the first round of mutagenesis and screening, plasmids encoding the most active clones in the population (typically 1–6% of the total, Supporting Information Table SIII) were isolated, pooled, and further diversified by a combination of epPCR and DNA shuffling.^{15, 16} Further details can be found in the Supporting Information Part I Section 7.

Supplementary material

Additional Supporting Information may be found in the online version of this article.