Fine‐tuning the activity and stability of an evolved enzyme active‐site through noncanonical amino‐acids

Site‐specific saturation mutagenesis within enzyme active sites can radically alter reaction specificity, though often with a trade‐off in stability. Extending saturation mutagenesis with a range of noncanonical amino acids (ncAA) potentially increases the ability to improve activity and stability simultaneously. Previously, an Escherichia coli transketolase variant (S385Y/D469T/R520Q) was evolved to accept aromatic aldehydes not converted by wild‐type. The aromatic residue Y385 was critical to the new acceptor substrate binding, and so was explored here beyond the natural aromatic residues, to probe side chain structure and electronics effects on enzyme function and stability. A series of five variants introduced decreasing aromatic ring electron density at position 385 in the order para‐aminophenylalanine (pAMF), tyrosine (Y), phenylalanine (F), para‐cyanophenylalanine (pCNF) and para‐nitrophenylalanine (pNTF), and simultaneously modified the hydrogen‐bonding potential of the aromatic substituent from accepting to donating. The fine‐tuning of residue 385 yielded variants with a 43‐fold increase in specific activity for 50 mm 3‐HBA and 100% increased kcat (pCNF), 290% improvement in Km (pNTF), 240% improvement in kcat/Km (pAMF) and decreased substrate inhibition relative to Y. Structural modelling suggested switching of the ring‐substituted functional group, from donating to accepting, stabilised a helix‐turn (D259‐H261) through an intersubunit H‐bond with G262, to give a 7.8 °C increase in the thermal transition mid‐point, Tm, and improved packing of pAMF. This is one of the first examples in which both catalytic activity and stability are simultaneously improved via site‐specific ncAA incorporation into an enzyme active site, and further demonstrates the benefits of expanding designer libraries to include ncAAs.


Introduction
As our understanding and control of biocatalytic reactions improves, organic chemists are increasingly able to employ naturally occurring or engineered enzymes to catalyse otherwise difficult synthetic chemical reactions. Transketolase has considerable potential for asymmetric C-C bond formation by catalysing the transfer of a two-carbon ketol group from a donor substrate to an aldehyde acceptor substrate [1]. The synthesis of α,α'-dihydroxyketone products provides a versatile backbone as a precursor to ketosugars, chiral aminodiols and other high-value molecules such as fragrances and flavours [2][3][4], while the stereoselectivity of transketolase is highly appealing when attempting C-C bond formation in order to form more complex biologics with multiple stereocentres such as antibiotics. While inherently reversible, the reaction can be rendered irreversible through use of β-hydroxypyruvate (HPA) as the donor substrate, thus increasing biocatalytic product yields.
Yeast and E. coli transketolases function as a homodimer of apparently structurally identical subunits, which is activated upon TPP binding [5]. Each homodimeric unit has two active sites located at the subunit interface [6,7], and each active site can be occupied by a maximum of one TPP molecule and one divalent cation (M 2+ ) such as Ca 2+ , Mg 2+ or Mn 2+ . Recently, the two-species model of transketolase activation was proposed that described the existence of two transketolase subpopulations within a cellular redox regulatory mechanism. Unmodified, inactive TK low is oxidised to form active TK high in response to oxidative stress [8], most likely through sulfenylation/ sulfonylation at Cys157 within the TPP-binding site. The proportion of TK high relative to TK low was correlated with the proportion of singly-(+16 Da) and doubly (+32 Da) oxidised TK, relative to unmodified TK low in mass spectra. These monomeric species could combine to form three dimeric species, TK high -TK high , TK low -TK low and the mixed dimer TK high -TK low .
The natural substrates of transketolase are generally phosphorylated sugars such as ribose-5-phosphate and xylulose-5-phosphate, but transketolase has been engineered to accept a plethora of novel substrates by employing various directed evolution strategies. For example, the substrate specificity of transketolase was successfully shifted from phosphorylated sugars, first towards nonphosphorylated, polar substrates [9], then aliphatic nonphosphorylated and heteroaromatic substrates [10][11][12], and finally to three aromatic benzaldehyde derivatives; 3-formylbenzoic acid (3-FBA), 4-formylbenzoic acid (4-FBA) and 3-hydroxybenzaldehyde (3-HBA) [13,14]. Crystallographic structure analysis of the latter variant, S385Y/D469T/R520Q (3M), coupled with in silico molecular docking of the three benzaldehyde derivatives, revealed the creation of two distinct binding pockets that are sterically separated by the D469T mutation [15]. The S385Y mutation was predicted to play a crucial role in aromatic substrate binding to both pockets through π-π stacking interactions with F434 and the aromatic ring of the substrate. While 3-FBA bound into one pocket and 4-FBA the other, both with relatively high affinity and catalytic turnover, 3-HBA was found to bind to both pockets 1 and 2 with low-affinity and poor catalytic productivity.
In previous rounds of directed evolution, activity gains were realised by semirational engineering of important active-site residues through relatively large saturation mutagenesis libraries, and yet also required re-engineering of protein stability due to critical stability losses incurred by active-site mutations [16]. Indeed, the 3M variant was similarly found to be less stable than the wild-type and has recently been restabilised through additional nonactive-site mutations [17]. The aim of this study was to further explore the critical aromatic ring of Y385, through altered aromatic ring electron density, to probe and potentially improve the catalytic activity, substrate inhibition, enzyme stability and binding pocket preference of the 3M variant simultaneously, without the need for additional restabilising mutations. We then also rationalised the observed trends in the context of active-site electronics, hydrophobic packing, and the size, polarity and hydrogen binding potential of the ring-substituted functional group.
The genetic code limits us to only 20 amino acids. Only phenylalanine has a less electron-dense single aromatic ring than tyrosine, and none are more electrondense. We therefore expanded the genetic code beyond its natural limits via incorporation of noncanonical aromatic amino acids (ncAAs) to create a series of five variants with highly electron-donating to highly electronwithdrawing aromatic ring substitutions at the para-position, (p-aminophenylalanine (pAMF)> tyrosine (Y)> phenylalanine (F)> p-cyanophenylalanine (pCNF) and p-nitrophenylalanine (pNTF), in order of decreasing aromatic ring electron density).
The incorporation efficiency (protein yield relative to wild-type) for various incorporation systems and their respective ncAAs has been well documented and quantified in the majority of studies to date. However, their incorporation fidelity (proportion of incorporated ncAA relative to misincorporated amino acids) is comparatively under-reported. Often, the misincorporation mass spectrometry peaks are hidden by the dominant ncAA-incorporated peak [18,19], or mass spectra are provided but not deconvoluted and/or quantified [18,20]; otherwise, the deconvoluted spectra are poorly resolved due to the low sensitivity of the LCMS [21,22]. In many others, no supporting mass spectra data are provided at all, giving no information on the level of misincorporation in the presence of ncAA [23][24][25][26][27][28]. Incorporation fidelity is a factor that should not be ignored for enzyme activity, ligand binding or protein stability studies, since heterogeneity is almost always unavoidable in ncAA-incorporation studies, and may hide the true performance of the ncAA-incorporated species. We therefore describe a methodology to (a) accurately quantify incorporation fidelity from deconvoluted mass spectra and (b) account for the contribution of misincorporated species to the experimentally determined or 'apparent' activity and stability parameters, to reveal the 'true' parameters of the ncAA-incorporated species of interest. The suitability of the 'apparent' and 'true' parameters as indicators of biocatalytic output, structure-function relationships and enzymatic potential is also discussed.
Global incorporation of ncAAs has enhanced the activity and thermal stability of a number of enzymes, mainly through incorporation of fluorinated natural amino acid analogues [29][30][31][32][33][34]. Site-specific, active-site ncAA incorporation is a nascent and promising field of research that was initially held back by low incorporation efficiencies and fidelities of incorporation. Over the last decade, the genetic code has been expanded to incorporate a myriad of ncAAs with much improved efficiencies and fidelities. To date, only a handful of studies have successfully improved [26,27,35] or introduced novel catalytic function [18,28], or enhanced thermostability [19,25] via sitespecific ncAA incorporation, and even less introduced them into an active-site. While the latter study [25] observed both a 1% increase in T m and a modest 15% improvement in catalytic activity at 40°C relative to wild-type, the thermostable mutant was less-active than the wild-type at 23°C, which suggested the improved activity reflected only the improved thermostability at 40°C rather than a genuine improvement in catalysis. To our knowledge, we therefore report the first example of a genetically encoded, sitespecific active-site ncAA-incorporated variant with both enhanced activity at 22°C and thermostability (T m ), and demonstrate the benefits of including ncAAs in site-specific, smart designer libraries.

Results
The electronic properties of para-substituted phenylalanine derivatives The strength of π-π stacking interactions is influenced by aromatic ring electron density. Electron-withdrawing groups (e.g. -NO 2 -) strengthen hydrophobic interactions between aromatic rings [36], while the opposite is true for electron-donating groups (e.g. -NH 2 ). The amino acids used in this study can be ranked in order of aromatic ring electron density from most dense to least dense: p-aminophenylalanine (pAMF)> tyrosine (Y)> phenylalanine (F)> p-cyanophenylalanine (pCNF)> pnitrophenylalanine (pNTF) (Fig. 1). The aromatic ring electron density at position 385 may therefore influence: (a) active-site hydrophobicity; (b) the overall hydrophobic packing of active-site residues; and (c) the strength of substrate binding due to π-π stacking interactions, in either catalytically productive or inhibitory orientations.
Nevertheless, substrate binding and catalytic turnover are dictated by a multiplicity of additional factors, including hydrogen bonding (H-bonds) with other active-site residues, steric hindrance and orientation of the acceptor substrate relative to the dihydroxyethyl-TPP (DHE-TPP) intermediate in threedimensional space. Furthermore, the characteristics of the ring-substituted functional groups may also have an impact on both catalysis and stability, as they differ in size, polarity and H-bonding potential from donating (pAMF and Y), to nonbonding (F), and finally accepting (pCNF and pNTF). While it is relatively hard to predict the outcome of such changes without far more structural information and computational power, one can predict with some confidence that one or more properties of an enzyme will at least change when altering the electronics of an active-site residue that is already known to be critical to the acceptance of aromatic substrates.

Efficiency and fidelity of ncAA incorporation
An evolved, orthogonal M. jannaschii aminoacyl-tRNA synthetase/tRNA pair [37], which incorporates ring-substituted phenylalanine derivatives at amber stop codons with high incorporation fidelities (i.e. low levels of natural amino acid misincorporation) and efficiencies (i.e. high protein yield) [20], but considerable ncAA promiscuity [22], was coexpressed using an optimised plasmid, pUltra, with the TK expression plasmid, pQR791-TK-S385X/D469T/R520Q, in C321.ΔA.exp. This 'amberless' E. coli strain, which has all amber codons replaced, and RF1 deleted, and thus cannot terminate translation at amber stop codons, giving an improved absolute efficiency of ncAA incorporation by preventing the formation of truncated protein [38]. The TK 3M variants generated are henceforth designated as 385pAMF, 385Y, 385F, 385pCNF and 385pNTF.
A methodology to quantify incorporation fidelity from deconvoluted mass spectra The major drawback of using expression strains such as BL21 (DE3) for ncAA incorporation is the potential to produce truncated protein rather than fulllength protein. Protein expression in C321.ΔA.exp prevented translation of truncated protein and significantly improved the efficiency of multisite ncAA incorporation [39]. However, only marginal improvements were observed in single-site incorporation [38] and it is still unclear what impact strain usage may have on the fidelity of incorporation. We therefore compared the incorporation into N-terminal Histagged TK using the pUltra plasmid system in both C321.ΔA.exp and BL21(DE3), to choose the best expression strain. We developed a methodology to quantify the incorporation fidelity based on the hierarchy of TK species and subspecies outlined below (Fig. 2 oxidised, active TK subspecies (i.e. TK high ). 7 TK inactiveoveroxidised (+48 Da, +64 Da, + 80 Da etc.), inactive TK subspecies. 8 %(nc)AA incorporatedproportion of (nc)AA-TK incorporated relative to all TK species. 9 %(nc)AA activeproportion of all TK high subspecies of (nc)AA incorporated relative to all TK species.
The incorporation fidelity of each ncAA was quantified by fitting triplicate mass spectra to the sum of multiple Gaussian functions in OriginLabs (Fig. 3). After assigning initial values by eye, the peak width parameter was shared between the summated Gaussian functions, while the peak area, peak width and peak centre parameters were unconstrained in order to converge towards their best fit values.
TK was recently shown to exist in multiple oxidation states [8], and therefore, each TK misincorporated and TK incorporated species consisted of a series of subspecies of unmodified, low-activity TK low (peak 1 = +0 Da); two oxidised, active TK high species (collectively called TK active from henceforth) (peak 2 = +16 Da; peak 3 = +32 Da) and overoxidised, inactive TK inactive (peak 3 = +48 Da, peak 4 = +64 Da, peak 5 = +80 Da, etc.) (Fig. 3). The two TK high peaks of each misincorporated species were small and undetectable by eye in Fig. 2. A schematic diagram outlining the hierarchy of TK species (grey) and subspecies (blue). To provide clarity, the following example is given for variant pAMF. pAMF-TK biocat (black) is defined as all pAMF species and subspecies contained in any biocatalytic sample. The sample can be subdivided into four species (grey): one pAMF-incorporated (pAMF-TK incorporated ) species and three pAMF-misincorporated (pAMF-TK misincorporated ) species (385Q, 385F and 385Y), where the natural amino acids glutamine (Q), phenylalanine (F) or tyrosine (Y) have been misincorporated at residue 385. Each of these four species can be subdivided into three subspecies (blue): low-activity TK (TK low ), active TK (TK active ) and inactive TK (TK inactive ). However, the proportion of variant pAMF that existed as the TK inactive subspecies of 385Q, 385F and 385Y was negligible and hence excluded from all calculations and from Figure 2. each mass spectra since they were dwarfed by the larger TK low peaks. However, formation of TK high is a redox-regulated cellular process and the ratio of TK high :TK low is relatively constant but variant-specific [40]. We therefore assumed that a) the two TK high peaks of each TK misincorporated species were present and still contributed towards the overall peak area of the mass spectra; and b) the ratio of TK high :TK low was equivalent in naturally incorporated and misincorporated natural amino acid species. . Each mass spectrum (n = 3; average spectra in black) was fitted to the sum of multiple Gaussian functions (smaller individual peaks) corresponding to unmodified (TK low ), singleand double-oxidised forms (TK high or TK active ) of the misincorporated species 385Q (red) and 385F (magenta); and TK low , TK high and the overoxidised, inactive forms (TK inactive ) of the ncAA-incorporated species 385pCNF (blue). These same species have been observed previously in both wild-type TK [8] and variant S385Y/D469T/R520Q [40]. The ratio of the peak area of unmodified, singly-and doubly oxidised peaks of Y and F was determined from their mass spectra (Fig. S1B,C), and applied to the mass spectra of 385pCNF. The cumulative fit of all peaks is shown in wine-red. The proportion of active 385pCNF relative to all TK species (both incorporated and misincorporated), %pCNF active , was determined from the peak area of singly-and doubly oxidised 385pCNF-TK high relative to the cumulative peak area ( Table 1). The %TK incorporated , %TK misincorporated and %TK active were determined for 385pAMF, 385Y, 385F and 385pNTF (Table S1) from their respective mass spectra (Fig. S1A,D) using the same methodology. The expected and observed molecular weights are listed in the Supporting Information (Table S2). We estimated the peak area of the two TK high peaks of each TK misincorporated species by applying the peak area ratio of peak1:peak2 and peak1:peak3, obtained from the mass spectra ( Fig. S1B-C), to peak1 of each TK misincorporated species within the mass spectra of each ncAA-TK incorporated variant (Fig 3; Fig. S1A & D). Therefore, the peak centre and peak area of the two oxidised TK high peaks of each misincorporated subspecies were fixed and the function refitted to obtain peak areas for peaks 1-3 of each misincorporated subspecies and peaks 1-7 of the ncAA-incorporated species within a single mass spectra. The fidelity of incorporation and the terms defined above were subsequently determined from the peak areas derived from the mass spectra of each ncAA variant, and were also applied to the experimentally determined activity and stability data below.

Strain choice has negligible impact on incorporation fidelity
Comparison of the mass spectra of 385pCNF when coexpressed with pUltra in either C321.ΔA.exp or BL21 (DE3) confirmed that in the case of TK, strain choice made little difference to the incorporation fidelity (76.8 AE 4.0 % and 73.8 AE 1.9%, respectively) ( Fig. 3; Table 1). Since there was negligible difference between the incorporation fidelity of the two expression strains, all subsequent variants were expressed in Fig. 4. Incorporation efficiency (grey), fidelity (red) relative to 385Y and the %TK active (blue) at residue 385. The incorporation efficiency, defined as the relative yield of TK (incorporated and misincorporated) relative to 385Y, was determined from densitometry analysis of SDS/ PAGE (n = 3) (Fig. 2). The incorporation fidelity, defined as the proportion of ncAA-incorporated species relative to all misincorporated species (385Q, 385F and 385Y), was determined from analysis of mass spectra (n = 3) ( Fig. 3; Table 1; Fig. 1). The proportion of active TK incorporated relative to all TK species (both incorporated and misincorporated), %TK active , was determined from the peak areas of the singlyand doubly oxidised TK high species relative to the cumulative peak area ( Fig. 3; Table 1; Fig 1). Error bars represent the standard error of the mean (SEM). The fidelity and efficiency of incorporation is ncAAdependent Figure 4 summarises the incorporation efficiency and fidelity as observed by SDS/PAGE and LC-ESI-MS, respectively (for full analysis and the supporting SDS/ PAGE gels and deconvoluted mass spectra, see SI Text 1, SI Fig. S1 and SI Fig. S2). In all ncAA variants, we observed a low but significant level of misincorporation of glutamine, phenylalanine and sometimes tyrosine, in addition to the TK low , TK high and TK inactive forms [8] of both ncAA-incorporated and misincorporated species. The proportions determined by mass spectrometry were subsequently used to extract the catalytic and stability parameters of the ncAA-incorporated species from our enzyme kinetic measurements. Overall, the M. jannaschii aminoacyl-tRNA synthetase/tRNA incorporation system was found to prefer ring-substituted phenylalanine derivatives with strong electron-withdrawing groups at the para position over those with electron-donating groups.

How heterogeneous is too heterogeneous?
The existence of multiple TK species in any given sample raised intriguing questions that were relevant not only to this study but a large number of proteins. Nature has evolved complex regulatory mechanisms and networks to facilitate dynamic intracellular responses to exploit or survive changes in external conditions. Many regulated proteins will therefore exist in more than one noncovalently or covalently modified state, which can be in reversible and/or irreversible equilibria. Other protein modifications are an often-unavoidable result of cellular conditions, such as the overoxidation of solvent-exposed cysteine residues. In most studies, these species cannot, and are not, readily separated during purification. The same is true in ncAA incorporation, where heterogeneous samples of ncAA-incorporated and misincorporated natural amino acid species are frequently, if not always, observed. Where we should draw the line between homogeneous vs heterogeneous samples is nontrivial, but ultimately depends on the context. Industrial uses of enzymes in biocatalysis and proteins as therapeutics evaluate the overall performance of the heterogeneous mixture in processes or clinical trials. However, understanding and potentially controlling the underlying heterogeneity and its impact on performance is still important. For example, we now know that TK exists as multiple homodimeric species [8,40], and yet all previous work evaluating the performance of TK variants in biocatalytic processes is no less useful or valid. Instead, the performance measured characterises the 'apparent' activity and stability of that variant as a biocatalyst, including all unmodified and modified species within the sample, obtained using a particular production process. However, to increase our understanding of the relative contributions of subspecies, and to potentially modify these for improving biocatalysis, there is value in attempting to deconvolve the 'true' kinetic parameters for individual species, from the total 'apparent' kinetic parameters, using simultaneous equations. We therefore aimed to deconvolve the relative contributions from natural amino acid misincorporation during ncAA incorporation, to reveal variants with significantly improved 'true' parameters, that would otherwise be overlooked.
The catalytic performance of 385pAMF, 385Y, 385F, 385pCNF and 385pNTF Previously, the K m and k cat of the 3M triple-mutant towards 3-HBA were reported as 390 AE 10 mM and 2.1 AE 0.2 s -1 , respectively, using a substrate concentration up to 25 mM 3-HBA owing to insolubility at higher concentrations [14]. As a result, the reported K m and k cat were overestimated because saturating concentrations of substrate were never reached. By sonication and careful pH adjustment, we have now improved the solubility limit of 3-HBA to at least 50 mM, allowing a more accurate calculation of the true kinetic parameters. Initially, the specific activity of each variant was determined at 50 mM 3-HBA. The catalytic performance of each variant was then investigated further over a range of [3-HBA] (Fig. 5).

The sliding-substrate inhibition Model
The significant substrate inhibition observed at above 25 mM for all variants indicated the binding of multiple substrate molecules or inhibitory poses, given the 'cooperative' shape of the curve at high [S] (Fig. 5). The experimental data were first fitted to the standard Michaelis-Menten-derived function describing a sequential-order noncompetitive inhibition (NCI) model ( Fig. 5A; Equation 1, Materials and Methods). In this model, enzyme activity decreased when the enzyme was bound to a single inhibitor molecule (in this case also the substrate) regardless of whether the enzyme was bound to the substrate or inhibitor first. With the exception of 385pNTF (Fig. 5A, magenta), the function gave an inadequate fit to the experimental data, which implied the observed acceptor substrate inhibition could not always be explained by the standard noncompetitive inhibition model alone.
Subsequently, a modified Michaelis-Menten function, the sliding-substrate inhibition (SSI) model, was derived (Equation 2, Materials and Methods) that describes an initial catalytically productive enzyme-substrate binding event with one substrate molecule, n 1 = 1, that is completely inhibited at high [S] in a second collective binding event, but allowing for multiple inhibitory enzymesubstrate interactions, n i ≥ 1. The function assumed that binding is sequential, with the catalytically productive substrate binding preceding subsequent inhibitory binding. The inhibitory poses are likely to have different binding constants (K i ) and so in theory, the substrate could 'slide' between different inhibitory binding poses. 'Sliding' is significantly more likely to be observed in catalytic reactions that are dominated by hydrophobic enzyme-substrate interactions, such as the interactions between the hydrophobic S385Y/D469T/R520Q active site and 3-HBA, which result in much less-well defined enzyme-substrate orientations compared to electrostatic and hydrogen-bonded interactions. For simplicity, the n i number of inhibitory poses were assumed to be represented by a single average inhibitory dissociation constant, K i .
The experimental data of all variants fitted extremely well to the SSI model (Fig. 5B) and were therefore considered to be a good kinetic model for TK. While the kinetic data of 385pNTF also fitted well to the SSI model, we observed a superior fit to the NCI model, which was equivalent to fixing n = 2 (and n i = 1) in the SSI model (Equation 1, Materials and Methods). The n i value was fixed at 1 in the SSI model when determining the kinetic parameters of 385pNTF, while the SSI model was used for all other variants. The experimental data of each variant fitted equally well with identical parameters to the randomly ordered SSI model (Fig. S3, Equation S1), below, which indicated that the additional random-order term did not contribute to the overall fit, thus validating the prior assumption that the reaction was sequential in order.
Determination of kinetic parameters for (nc)AA-TK biocat , (nc)AA-TK incorporated and (nc)AA-TK active The analysed mass spectrometry data were combined with the experimentally determined activity data and then globally fitted to the weighted sum of three SSI model functions (Equation 3, Materials and Methods).
For each variant, the apparent kinetic parameters derived from the experimental data were a convolution of the contributions from the various (nc)AA-TK incorporated and (nc)AA-TK misincorporated subspecies (Fig. 2). By contrast, the kinetic data for natural incorporation of F and Y at residue 385 were not convoluted and so were obtained directly, thus also allowing their contributions to the other variant kinetics to be accounted for. Three variations of this function were used to analyse the kinetic parameters of (nc)AA-TK biocat , (nc)AA-TK incorporated and (nc)AA-TK active of each TK variant ( Table 2). The pros and cons of each method are discussed in turn below.

Analysis 1: No deconvolution (TK biocat )
The first analysis compared the kinetic parameters of each transketolase variant as an overall biocatalytic sample, TK biocat , without attempting to deconvolve the parameters of each species and subspecies. The activity data were fitted to a single SSI function (Fig. 5B).
Analysis 1 provided information on the performance of each variant (and all subpopulations) as an overall biocatalyst. However, it was the highest level analysis and provided only 'apparent' activity and stability parameters of TK biocat , but did not reflect the 'true' catalytic performance of the ncAA-incorporated subpopulations, especially in cases where the ncAA is incorporated with low fidelity (i.e. high % misincorporation). This analysis indicated 385pCNF to be the best overall biocatalyst. Incorporation of pCNF into the active site of S385pCNF/D469T/R520Q improved the k cat , K i , n i and k cat /K m by 57%, 74%, 57% and 64%, respectively, relative to variant 385Y. The K m of 385pCNF was within error of that for variant 385Y. Incorporation of pAMF at residue 385 gave a 140% improvement in K m but decreased the k cat by 38%, resulting in a 45% improvement in catalytic efficiency (k cat /K m ) relative to 385Y. Finally, incorporation of F and pNTF at residue 385 significantly improved their K m , while the catalytic turnover was severely impaired, resulting in overall lower catalytic efficiencies than for 385Y. Table 2. The kinetic parameters of the TK biocat , TK incorporated and TK active species and subspecies of variants 385pAMF, 385Y, 385F, 385pCNF and 385pAMF towards 3-HBA. Associated errors are the fitting error for the weighted sum of three sliding-substrate inhibition (SSI) model functions. The %(nc)AA incorporated , %(nc)AA misincorporated and %(nc)AA active used to weight the function were determined from the peak areas from LC-ESI-MS data as described in Fig. 3 and the Materials and Methods. Note that 385pNTF-TK incorporated and 385pNTF-TK active fitted best to the NCI model; hence, n i = 1 and has no associated error.

Variant
Rel. spec activity (50 mM) The 'true' kinetic parameters of each (nc)AA-TK incorporated variant were deconvoluted from the experimentally determined 'apparent' data, by accounting for the presence of (nc)AA-TK misincorporated species, and their contribution towards the total activity/ stability (Fig. 5C). The three SSI functions were weighted to the %ncAA incorporation %F misincorporation and %Y misincorporation determined from the mass spectra of each variant ( Fig. 3; Equation 3, Materials and Methods; Fig. 1, supplementary information). The 'true' catalytic parameters of naturally incorporated 385F-TK incorporated and 385F-TK incorporated were determined from experimental activity data ( Fig. 5; Table 2). The specific activity of 385Q at 50 mM 3-HBA was negligible relative to 385Y, and most variants had a small %Q misincorporation , and therefore, the contribution of 385Q towards overall activity was negligible. Two of the three weighted SSI model functions were populated with the catalytic parameters of 385F-TK incorporated and 385Y-TK incorporated . The K m of the third SSI function, corresponding to the ncAA-TK incorporated species, was fixed at the TK biocat K m value, while the other parameters were floated to extract the 'true' kinetic parameters of the ncAA-TK incorporated species of 385pAMF, 385pCNF and 385pNTF (visualised in Fig. 5C). Analysis 2 had the largest relative impact on 385pAMF because a high level of misincorporation was observed in this variant. Consequently, after accounting for misincorporation, the incorporation of pAMF at residue 385 improved the k cat by 22% relative to 385Y, rather than the 38% decrease observed in Analysis 1 ( Table 2). This improvement in k cat also resulted in an impressive 186% improvement in the k cat /K m relative to 385Y. Thus, strategies to minimise misincorporation in 385pAMF would be greatly beneficial for this variant as a biocatalyst. The major caveat to Analysis 2 is that it averages out effects that may have arisen from any population of mixed dimers between the monomeric subunits of different subspecies. It also ignores any potential influence of the variant upon the ratio of TK high :TK low [40], which could in turn affect the overall activity, as well as the inhibition of TK high by TK low in the resulting mixed dimer species.
Analysis 3: Deconvolution of (nc)AA-TK active subspecies parameters Analysis 3 provided the most detailed information about the 'true' parameters of only the active subspecies within each (nc)AA-incorporated TK variant, (nc)AA-TK active (visualised in Fig. 5C). The same methodology was employed as in Analysis 2, except the three SSI functions were weighted using % ncAA active of the TK incorporated and %F active /%Y active of the TK misincorporated species, all derived from mass spectra. The methodology therefore normalised for variations in the active proportion of TK incorporated and TK misincorporated species, which varied from 11.5% to 28.3%, and hence provided even more clarity on the best-performing active variant subspecies. The only caveat remaining is that this analysis again averages out any specific effects of mixed dimer species. Analysis 3 therefore makes the most assumptions for deconvolution, but gave the best comparative analysis of TK incorporated variant subspecies. The kinetic parameters for TK active subspecies of each variant were also most amenable to interpretation through in silico molecular docking simulations, as discussed below.
Taking the discussed caveats into account, we were able to determine the kinetic parameters of (nc)AA-TK biocat , (nc)AA-TK incorporated and (nc)AA-TK active species of each variant in order to determine, or at the very least compare, the relative catalytic performance of each variant. While, as expected, their absolute values varied significantly between TK biocat and the TK active subspecies, the relative changes in parameter trends were relatively constant (with the exception of 385pAMF) because the % TK misincorporated and % TK high remained comparable between variants. Analyses 2 and 3 revealed the 'true' catalytic potential of 385pAMF that was otherwise masked by low-activity TK misincorporated and TK low /TK inactive subspecies. All kinetic parameters referred to henceforth correspond to only the TK active subspecies of each variant (Table 2), unless stated otherwise.

Analysis of the kinetic parameters of (nc)AA-TK active
While we expected a change in the catalytic performance upon tweaking the electronics of an important residue in the active site, it was difficult to predict what trend, if any, may emerge given the complexity of substrate binding and catalytic turnover.
The specific activity of 385pCNF was an impressive 43-fold greater than that of 385Y, while those of 385pAMF, 385pNTF and 385F increased 13-fold, 4-fold and 1.2-fold, respectively ( Fig. 6A; Table 2). However, no clear trend was observed in the specific activity, K m nor k cat (Fig. 6B; Table 2) as a function of the aromatic ring electron density of residue 385, which confirmed that the altered active-site interactions were more complex than simply relating to ring electron density. Interestingly, the catalytic improvement of the two best-performing variants, 385pAMF and 385pCNF, relative to 385Y, was driven by two divergent evolutionary mechanisms. Incorporation of pAMF improved the catalytic efficiency, k cat /K m , by 240%, primarily via a 140% improvement in K m ( Table 2; Fig. 6B). Conversely, incorporation of pCNF at residue 385 instead improved k cat /K m by 110% almost exclusively via a 100% improvement in k cat . The other two mutations, 385F and 385pNTF, resulted in large improvements in K m (290% and 200% improvements, respectively), but even greater decreases in k cat .
The theoretical total number of inhibitory substrate orientations and/or binding events, n i , was correlated with the aromatic ring electron density of residue 385 ( Fig. 6C; Table 2). 3-HBA could bind 385pAMF in as many as 10 inhibitory orientations, while that was reduced to one in 385pNTF. Variant 385pCNF had a higher than expected n i and was a slight anomaly to the general trend. There appeared to be no obvious trend in K i as a function of aromatic ring electron density as variants 385Y, 385F and 385pNTF had similar K i -values, while the introduction of pAMF and pCNF at residue 385 improved the K i by 42% and 76%, respectively.
Overall, the substantially improved specific activity of pCNF at 50 mM 3-HBA relative to 385Y was explained by a combination of a 100% improvement in k cat and substantially reduced substrate inhibition (through both a decrease in n i and an increase in K i ).

Thermal stability of holo-transketolase variants
Previously, introduction of mutations that conferred rigidity into the cofactor-binding loop (e.g. H192P/ H282P) led to improved thermal stability of wild-type transketolase by 5°C [41]. We postulated that a decrease in aromatic ring electron density via electronwithdrawing substitutions may also strengthen π-π stacking interactions between residue 385 and activesite nonpolar aromatic residues, with the potential to increase active-site rigidity and hence thermal stability.
The methodology of Analyses 1 and 2 was reapplied to the experimental thermostability data of the five variants (Fig. 7) by similarly weighting a sum of twostate model functions for thermal denaturation [42][43][44] of each variant (Equation 4, Materials and Methods). In Analysis 2, the variants were globally fitted, with the stability parameters of 385Y and 385F shared, and the %(nc)AA-TK incorporated , %385F-TK misincorporated and %385Y-TK misincorporated weighting parameters fixed for each variant. TK active cannot currently be isolated from TK low , and their relative stabilities are unknown; therefore, the stability of (nc)AA-TK active cannot be calculated using the methodology of Analysis 3. The T m values from Analyses 1 and 2 (Table 3) were within error for all variants, which indicated (nc) AA-TK biocat stability was largely dominated by the most populous species, (nc)AA-TK incorporated . The ΔH vh values varied considerably more between Analyses 1 and 2, with the values from Analysis 2 generally higher than those of Analysis 1. This was potentially a consequence of the interactions between species and subspecies in heterodimers that have not been accounted for in the model. Since fewer caveats apply to Analysis 1, and Analysis 2 gleans little more information than Analysis 1, all stability parameters referred to henceforth correspond to those of only the TK biocat species of each variant (Table 3), unless stated otherwise.
The T m did not correlate linearly to any properties of the variants or mutation types, with the best observed for n i of the variants (R 2 = 0.73). However, the thermal transition mid-point, T m , increased nonlinearly with aromatic ring electron density (Fig. 6D), in the opposite direction to that predicted based on the strength of π-π stacking. The most stable variant, 385pAMF, was 7.8°C more stable than the least Fig. 7. The thermal unfolding of all TK subspecies (ncAA biocat or AA biocat ) of variants 385pAMF (black), 385Y (red), 385F (blue), 385pCNF (green) and 385pNTF (magenta). The apparent thermal stability was determined by the change in the fluorescence emission ratio (350/330) as a function of time, and fitted to a two-state model of thermal denaturation, as described previously [42][43][44]. Error bars represent the standard error of the mean (SEM), n = 3. Table 3. Summary of the thermal stability of TK biocat and TK incorporated of wild-type, 385pAMF, 385Y, 385F, 385pCNF and 385pNTF. Parameters were determined by fitting the change in the fluorescence emission ratio (350/330) as a function of temperature to a two-state model of thermal denaturation [42][43][44] (Table 3). This was the first time that the aromatic substrate-accepting 3M variant had been engineered to be more stable than the wild-type [45]. The thermal denaturation of 385pAMF had increased cooperativity, as indicated by the high value of ΔH vh (Fig. 7; Table 3), which is itself indicative of tight overall packing and increased rigidity. The additional rigidity in 385pAMF was not expected from the increased aromatic ring electron density, which would in theory form weaker hydrophobic interactions than the other four aromatic amino acids. Molecular modelling instead suggested that the shift in the para-substituted functional group, from an H-bond acceptor (pCNF and pNTF), to nonpolar (F), to an H-bond donor (pAMF and Y), increased the active-site and cofactor-loop rigidity. Analysis of the energy-minimised structures of 385pAMF and 385Y showed the carbonyl backbone of G262 was within H-bonding distance of the -NH 2 and -OH side chain groups, at 3.6Å and 4.2Å, respectively (Fig. 8A, C & D). The latter distance is large for an H-bond, and, if formed, such an interaction would be extremely weak. In both energy-minimised structures, a helix-turn was formed between residues D259-H261, as a result of stabilisation by the interaction between residue 385 and G262. Neither an H-bond nor a helix-turn was observed in the other three less-stable variants (Fig. 8  B, E, F & G). G262 is located in the opposite subunit of the transketolase dimer and therefore provides both an anchor to increase the rigidity of the cofactor-binding loop, increasing ΔH vh , and an additional intersubunit interaction, to increase T m . The loss of secondary structure in at least three variants may explain why the ΔH vh was much lower and roughly equivalent across these variants. Furthermore, the weakened or absent intersubunit H-bond, in addition to the introduction of bulkier, and polar, ring-substituted functional groups, may explain the gradual decrease in T m observed from 385pAMF to 385pNTF.

Computational docking of 3-HBA into triplemutant variants
The structure-function relationships driving the two divergent evolutionary mechanisms of pAMF and pCNF, and the trend in changing substrate inhibition across the variants were subsequently interrogated further. In silico molecular docking of 3-HBA into the active site of variant Y, and also into the computationally mutated, energy-minimised active sites of For all analyses, the highest energy substrate subcluster, energy cluster 1 (EC1), shown in Fig. 9, was considered catalytically productive when the aldehyde moiety was oriented 0°or 180°, and catalytically inhibitive when oriented 90°or 270°, relative to the DHE-TPP intermediate. All subclusters of EC2 or lower were defined as inhibitory. Active-site interactions with the catalytically productive substrate poses were subsequently analysed to rationalise the experimentally determined catalytic data, in terms of: the number of productive, unproductive and inhibitory poses; binding pocket preference (% Pockets A and B); and predicted active-site interactions. The results are summarised in Table 4, while the full results can be found in Supporting Information (Supporting Information Appendix, Supporting Information Text 3).

Comparison of variant Y and pAMF
In silico analysis predicted 3-HBA to bind to the active site of 385pAMF and 385Y in near-identical orientations (Fig. 9A), including two that were catalytically productive and three that were inhibitory. The high number of inhibitory subclusters in both 385pAMF and 385Y was consistent with the high cooperativity of inhibition, and high theoretical number of inhibitory binding orientations (n i ) observed experimentally. The two catalytically productive subclusters of each variant were split between the two binding pockets that had been previously identified [15] (Fig. 9A). The decreased aromatic ring electron density in 385Y relative to 385pAMF shifted the binding pocket preference from Pocket A to Pocket B, with 52% of productive poses in Pocket B of 385pAMF compared to 82% in 385Y. The observed shift to Pocket B may have been due to the loss of two interactions between 3-HBA and Pocket A of 385Y compared to that of 385pAMF (Fig. S4). The additional active-site interactions in 385pAMF were consistent with the decreased K m , observed experimentally.

Analysis of 385pCNF
3-HBA was predicted to bind 385pCNF in two catalytically productive and one inhibitory subcluster of poses that were unique to 385pCNF (Fig. 9B) and hence could not be directly compared to the sets of subclusters that were common between variants 385pAMF/385Y and 385F/385pNTF (Fig. 6A). The  low number of in silico inhibitory poses agreed with the relatively low n i determined experimentally. The presence of two catalytically productive subclusters likely contributed to the higher n i value relative to 385F and 385pNTF, and may also have contributed to the low substrate affinity as a result of a less-favourable enthalpic contribution. The further decrease in aromatic ring electron density completely shifted the binding pocket preference from Pocket A to Pocket B, with 54% of productive poses occurring in a newly created extended region of Pocket B (Pocket B*). The catalytic subcluster in Pocket B* was predicted to bind weakly with few interactions between the active-site and 3-HBA, and an unfavourable interaction between the polar aldehyde moiety of 3-HBA and the nonpolar ring of F434 (Fig. S5). Nevertheless, the aldehyde moiety was positioned very close to the DHE-TPP intermediate in the extended pocket and had a Bürgi-Dunitz angle (the geometric angle between the carbonyl plane and the nucleophile-carbon line) closest to the theoretically best Bürgi-Dunitz angle of (124°versus 107°, respectively). The prediction of one low-affinity, highly productive and one high-affinity, lowly productive binding pocket within the active site of 385pCNF is in agreement with the high K m , k cat and higher than expected n i determined experimentally.

Comparison of 385F and 385pNTF
The single catalytically active subcluster of 385F and 385pNTF was predicted to bind exclusively in Pocket B in identical orientations (Fig. 9C), while each variant had two and one additionally inhibitory subclusters, respectively. The inhibitory subclusters were low in energy and number, again supporting the low number of inhibitory substrate orientations observed experimentally. Furthermore, the higher number of inhibitory subclusters and a higher overall proportion of inhibitory poses (20%) also explains the increased substrate inhibition observed in 385F compared to 385pNTF (0.5% of poses were inhibitory).
385F and 385pNTF were predicted to have nearidentical interactions with 3-HBA in their catalytically productive orientations, the only difference being a stronger π-π stacking interaction with residue F385 compared to only weak Van der Waal's forces with residue pNTF385 (Fig. S6). This difference was not observed in the experimentally determined k cat and K m , however, highlighting the caution that must be taken in the interpretation of computational docking, which at best provides only guidance for possible mechanistic explanations, and to inform future experimental work. Finally, the aldehyde moiety of 3-HBA was in close proximity to H473, potentially explaining the low K m values for variants 385F and 385pNTF.

Discussion
A comparison of the kinetic and stability parameters of each variant before and after accounting for misincorporation and the oxidised TK subspecies demonstrated the impact that this can have on the kinetic parameters derived directly from the experimental data. Previously, studies into the activity and stability of ncAA-incorporated variants have confirmed, but often not quantified, incorporation and misincorporation by mass spectrometry. We propose that determination of the fidelity of incorporation should be a standard procedure in studies similar to ours to allow a more accurate comparison of the catalytic performances of two or more ncAA-incorporated variants.
Through evolution and expression optimisation of tRNA/synthetase pairs, the fidelity of ncAA incorporation has improved greatly over the years. It is relatively unsurprising that the ncAA-incorporation fidelities obtained for GFP are often high, given that most tRNA/synthetase pairs have been evolved using GFP as the target protein. However, incorporation efficiency and fidelity can be protein-/sequence-dependent because of a range of factors, such as mRNA secondary structure, promoter sequence, rate of translation, protein length, amino acid sequence and the interaction between sequential amino acids and their cognate tRNA/synthetase pairs as the nascent polypeptide chain is synthesised. Differentiating between structurally similar natural and noncanonical amino acids is often challenging, even for evolved orthogonal tRNA/synthetase pairs. In these cases, higher levels of misincorporation are unavoidable and must be accounted for in kinetic comparisons.
This study began with a highly evolved variant that had already been optimised for acceptance of aromatic aldehyde substrates through saturation mutagenesis with natural amino acids at residue 385 [15]. However, the improved function had led to a trade-off in stability, which had been recovered through the engineering of molecular dynamics in a previous study, but never improved beyond the stability of wild-type TK [45]. The expansion of saturation mutagenesis to include ncAAs has now led to observed improvements in both catalytic activity and stability simultaneously, which demonstrates the benefits of their inclusion in designer library approaches. Even a small library of variants at a single active-site residue generated a broad range of catalytic and stabilising properties. Variant 385pAMF had the highest thermal stability, a relatively high maximal activity and strong substrate binding, but high susceptibility to substrate inhibition. Variant 385pCNF gave high catalytic turnover and lower susceptibility to substrate inhibition, but weaker substrate affinity. Variants 385F and 385pNTF had lower maximum activities, but strong substrate binding and low susceptibility to substrate inhibition.
To our knowledge, S385pAMF/D469T/R520Q is the first example in which both catalytic activity and stability are simultaneously improved via site-specific ncAA incorporation into an enzyme active site, and demonstrates the benefits of both fine-tuning preevolved residues using ncAAs, but also expanding directed evolution or designer libraries to include ncAAs in general.

Chemicals and reagents
Tris/HCl was purchased from VWR International (Lutterworth, UK), and guanidine-HCl was purchased from Life Technologies Ltd. HPA was synthesised by reacting bromopyruvic acid with LiOH, as described previously [46]. Para-cyanophenylalanine was purchased from Bachem (California, USA). All other chemical reagents were purchased from Sigma-Aldrich (Poole, UK).

Mutagenesis at residue 385
Mutagenesis was carried out using QuikChange site-directed mutagenesis using the manufacturer's protocol (Stratagene, Cambridge, UK) and the following primer and its reverse complement, designed for specific mutations at residue 385: The dpnI-digested PCR product was transformed into XL10-gold competent cells and the plasmid subsequently isolated using a Qiagen Miniprep Kit (Stratagene, Cambridge, UK).

Preparation of and cotransformation into competent C321.ΔA.exp 'amberless' cells
The 'amberless' E. coli strain, C321.ΔA.exp, a gift from George Church (Addgene plasmid #49018), was used as the expression strain for all variants. A 50 mL culture of C321.ΔA.exp in LB was grown to an OD 600 ≈ 0.5 in a 250-mL shake flask and subsequently transferred to two prechilled 50-mL falcon tube and cooled for 10 min on ice. All consumables required were cooled on ice for the duration of the procedure. The cells were centrifuged at 2700 g for 10 min at 4°C, the supernatant discarded and the pellet resuspended in 1.6 mL precooled 100 mM CaCl 2 for 30 min on ice. Centrifugation followed by resuspension and incubation was repeated for each falcon tube. Finally, the cells were combined into a single tube and 0.5 mL prechilled 80% glycerol was added. The resulting chemically competent cells were frozen in liquid nitrogen and stored at −80°C until required. The chemically competent C321.ΔA.exp cells were cotransformed with pUltra, a gift from Peter Schultz (Addgene plasmid #48215), encoding the ncAA-incorporation system [20], and pQR791, encoding a constitutively expressed transketolase variant.

Enzyme preparation and enzyme kinetics
Variants of the transketolase mutant S385Y/D469T/R520Q were coexpressed with the ncAA-incorporation machinery from the pUltra plasmid for eight hours in C321.ΔA.exp cells in the presence of 1 mM ncAA and 1 mM IPTG. The resulting cell pellet was lysed and purified as described previously [47]. Purified transketolase was ultrafiltrated four times using Amicon Ultra-4 10k MWCO centrifugal filter to remove excess imidazole and cofactors. Protein concentration was determined by absorbance at 280 nm in 6 M guanidine-HCl and 20 mM sodium phosphate, pH 6.5. Absorbance was measured using a NanoDrop spectrophotometer; the molecular weight of each variant was based on the wild-type monomeric molecular weight of 73035.5 gÁmol −1 and an extinction coefficient (ϵ) of 92630 LÁmol −1 Ácm −1 , modified for each variant.
Kinetic parameters were obtained at saturating 50 mM HPA and 3-50 mM 3-HBA. About 80 µL 0.6-1.0 mgÁmL −1 TK was incubated with 20 µL of 10 × cofactor solution (24 mM ThDP, 90 mM MgCl 2 ) for 30 min, and the reaction initiated with 100 µL 2x 3-HBA in 50 mM Tris/HCl, pH 7.0. All reactions were carried out in triplicate in glass vials at 22°C. Samples of each reaction were quenched every 30 min for 180 min by addition of 380 µL 0.1% TFA to 20 µL sample and centrifuged at 13 000 rpm for 3 min, and the supernatant analysed by HPLC, as described previously [14]. The TK concentration in each reaction was between 0.07 and 0.3 mgÁmL −1 . Higher TK concentrations were used for the 3-HBA reaction due to the slower conversion. All data were fitted by nonlinear regression to either the sequential noncompetitive inhibition (NCI; Equation 1) or sliding-substrate inhibition (SSI; Equation 2) model to determine the kinetic parameters of each variant.

Derivation of the modified Michaelis-Menten equation
The modified Michaelis-Menten function was derived in a similar way to the standard Michaelis-Menten function, except using the following chemical equilibria, left, that describe a single catalytically productive enzyme-substrate binding event, n 1 = 1, that is completely inhibited at high [S] by multiple (n i ) inhibitory enzyme-substrate interactions. The function assumes that the catalytically productive binding event occurs prior to the inhibitory binding event in a sequential reaction order. The chemical equilibria marked with a cross are those which become obsolete once the reaction is defined as kinetically sequential. The respective dissociation constants, right, are derived using the law of mass action. where [E] is the enzyme concentration, [S] is the substrate concentration, n 1 and n i are the theoretical number of molecules, or number of orientations, of substrate binding in the first (catalytically productive) and second (inhibitory) binding event, n is the theoretical total number, or total orientations, of productive and inhibitory substrate molecules that can bind the enzyme and K d1 and K d2 app are the apparent dissociation constants of the first catalytically productive binding event and all subsequent inhibitory binding events, respectively. The inhibitory binding events are defined as either multiple substrate molecules that simultaneously bind to the enzyme active site, or as a single substrate molecule that interacts with the enzyme active site in multiple inhibitory binding poses. The n i number of inhibitory poses are assumed to have slightly different binding inhibition constants, K i , which are represented by a single inhibitory dissociation constant, K d2 app . In other words, the individual inhibitory dissociation constants for each binding pose are convoluted or 'averaged' by multiplying and dividing the binding constants together to give a single representative inhibitory binding constant, such that.
In theory, the substrate can 'slide' between inhibitory binding poses. 'Sliding' is significantly more likely to be observed in catalytic reactions that are dominated by hydrophobic enzyme-substrate interactions, which result in much less-well defined enzyme-substrate orientations compared to electrostatic interactions.
The total enzyme concentration, [E T ], can be expressed as the following: i which rearranges and simplifies to the following: The SSI model above contrasts to the noncompetitive inhibition (NCI) model (Equation 1), which describes inhibition by a single substrate molecule in a single orientation. In this case, the SSI model simplifies to the NCI model when n = 2, and therefore n i = 1.
Determination of kinetic parameters of (nc)AA-TK biocat , (nc)AA-TK incorporated and (nc)AA-TK active The analysed mass spectrometry data were combined with the experimentally determined activity data and then

Supporting information
Additional supporting information may be found online in the Supporting Information section at the end of the article. Supplementary text. Full analysis of ncAA-incorporation.
Supplementary equation S1. The random-order SSI model. Fig. S1. The mass spectra of A) 385pAMF, B) 385Y C) 385F and D) 385pNTF. Fig. S2. SDS-PAGE of 1 mg/mL lysate of A) 385pCNF and 385pNTF, B) 385pAMF, 385Y and 385F; and C) 0.2 mg/mL of purified 385pCNF. Fig. S3. A fit of the experimental kinetic data of variants 385pAMF (black), 385Y (red), 385F (blue), 385pCNF (green) and 385pNTF (magenta) towards 3-HBA to the randomly-ordered SSI model. Fig. S4. The predicted enzyme-substrate interactions of Pocket A (A and C) and Pocket B (B and D) of variants pAMF and Y, respectively. Fig. S5. The predicted enzyme-substrate interactions of Pocket B (A) and the extended region of pocket B, B*, (B) of variant pCNF. Fig. S6. The predicted Pocket B enzyme-substrate interactions of variants F and pNTF, respectively. Table S1. The random-order SSI model. Table S2. The expected (Q exp , F exp , Y exp and ncAA exp ) and observed (Q obs , F obs , Y obs and ncAA obs ) peak centres for variants 385pAMF, 385Y, 385F, 385pCNF and 385pNTF.