Introduction

Protein engineering enables biocatalytic enzymes to be evolved in vitro for improved properties under process conditions, or for altered substrate specificity or enantioselectivity (Turner 2003; Hibbert et al. 2005). Advances in process miniaturisation and automation are also enabling process options to be rapidly assessed prior to scale up (Duetz et al. 2000, 2001; Lye et al. 2002; Doig et al. 2002; Nealon et al. 2005; Micheletti et al. 2006). These approaches require generic microplate-based assays either for the screening of libraries of enzyme variants or for a wide range of process conditions. An ideal assay would be: rapid; sensitive; amenable to organic solvent or extremes of temperature and pH; useful for a range of substrates; and capable of direct quantification of reactants and products. Here we present the development, optimisation and characterisation of a generic microplate-based assay for transketolase (TK).

The thiamine diphosphate (ThDP) and Mg2+-dependent enzyme transketolase (EC 2.2.1.1) catalyses the transfer of a 1,2-dihydroxyethyl group from a ketose to an aldose sugar, forming a new carbon-carbon bond in a stereospecific and stereoselective manner. Use of β-hydroxypyruvate (β-HPA) as the donor substrate, renders the reaction irreversible (de la Haba et al. 1955), and improves its potential as a catalyst for large-scale biotransformations (French et al. 1995; Hobbs et al. 1996; Brocklebank et al. 1999; Shaeri et al. 2006). Several TK activity assays could potentially be implemented in microplates. A fluorogenic assay (Sevestre et al. 2003) allows the sensitive detection of activity, useful for identifying enzyme variants with poorly accepted substrates. However, the substrate contains a fluorogenic leaving group which can affect the K m (Sprenger et al. 1995), substrate inhibition (Brocklebank et al. 1999) and rate of reaction (Wahler et al. 2001; Goddard et al. 2004). The fluorogenic and enzyme-linked (de la Haba et al. 1955; Sprenger et al. 1995) assays are not amenable to non-physiological conditions, such as organic solvents, or extremes of temperature and pH, as they require secondary enzymes to generate the signal change.

HPLC based assays are more generic than fluorogenic or colorimetric assays (Smith et al. 2006), as all reaction components can potentially be quantified, independent of reaction conditions, and without the need for labelled substrates. We have modified an existing 35 min HPLC assay (Mitra et al. 1996) for increased throughput, then used it to measure the response curves of both cofactors for the biocatalytic process, and also to determine their dissociation constants. Conditions were also optimised for improved transketolase activity obtained in microplate-based fermentations, cell lysis, cofactor addition and bioconversion, thus enabling its use in the screening of directed evolution libraries under biocatalytic process conditions. The well-to-well variation is determined for each step and also the complete screening process, including where appropriate, the Z-factor which estimates the confidence level at which an enzyme activity improvement can be identified.

Materials and methods

Transketolase enzyme reactions in 96-well microplates

To 150 μl of each 0.5 ODU culture sonicate (see below) in 96-well microplates, was added 30 μl cofactor solution (90 mM MgCl2.6H2O, 25 mM ThDP, pH 7.5). After 15 min, 120 μl substrate solution (250 mM β-HPA, 250 mM glycolaldehyde, 125 mM Tris/HCl, pH 7.5), was added. Reactions were quenched after 10 min at 25°C, with four volumes of 0.2% (v/v) trifluoroacetic acid (TFA), and l-erythrulose determined by standard or rapid HPLC as below.

Standard HPLC for l-erythrulose and β-HPA

Samples (10 μl) were injected onto a 300 mm Aminex HPX-87H ion-exclusion column (Bio-Rad Laboratories) maintained at 60°C in a Dionex GP50-LC30 HPLC system. The mobile phase was 0.1% (v/v) TFA at 0.6 ml min−1. Detection was by absorbance at 210 nm.

Rapid HPLC assay for transketolase

As above except that samples (10 μl) were injected onto a 50 mm PL Hi-Plex H guard column (Polymer Laboratories Ltd.) maintained at 30°C. The mobile phase was 0.1% (v/v) TFA at 2 ml min−1. Detection was by absorbance at 210 nm.

Comparison of transketolase activity from each strain

The high copy number plasmid pQR711 contains the promoter, ribosome binding site and transketolase open reading frame of the tktA gene from E. coli and will therefore express TK to high levels in any E. coli strain (French et al. 1995). It was purified from the E. coli K12 bacterial strain JM107 (endA1 gyrA96 thi hsdR17 supE44 relA1 Δ(lac-proAB) [F′ traD36 proAB+ lacIq lacZΔM15] hsdR17(R −K m +K ) λ−) and transformed by heat-shock into three further E. coli strains: BL21-Gold(DE3) (Novagen); JM109 and XL10-Gold (Stratagene, Inc.). Shake-flask cultures of all four strains, with and without pQR711, were prepared in LB medium (with 150 μg ampicillin ml−1 for pQR711 strains). Aliquots were diluted with LB medium to an OD600 of 0.5 (0.5 ODU) and sonicated. Transketolase activities for each strain were measured in triplicate from the concentration of l-erythrulose determined by standard HPLC.

Microplate-based fermentations

Wells of a standard 96-deep-square-well (DSW) plate, 96-round-well (RW) plate or 384-square-well (SW) plate containing respectively 900 μl, 150 μl or 60 μl of LB-ampicillin medium, were inoculated from single colonies of E. coli XL10-Gold pQR711 by a QPix2 colony picker (Genetix Ltd., UK). Each plate was sealed by inverting and taping an identical plate over the top, then incubated for 16 h at 600, 1,000 or 1,400 r.p.m. for RW and SW plates, or at 900 r.p.m. for DSW plates, on a Variomag Teleshake unit (Camlab Ltd.) in a 37°C incubator. After incubation, the cell density in each well was measured at 600 nm using a FLUOstar Optima plate reader (BMG Labtechnologies GmbH.) after 10-fold dilution with RO-water into either 384-SW or 96-RW plates. For each plate, wells were also pooled to determine the average cell density as OD600.

Optimisation of cell lysis

A shake-flask culture of E. coli XL10-Gold pQR711 was diluted with LB medium to an OD600 of 3.4. Aliquots of 1 ml were either lysed by sonication, the addition of Bugbuster (Novagen), freeze-thaw cycles from −80°C, or left untreated. Each sample, prepared in triplicate, was diluted with LB medium to the equivalent OD600 of 0.5, and 150 μl used to measure transketolase activity as determined by standard HPLC.

Cofactor optimisation and determination of dissociation constants

A shake-flask culture of E. coli XL10-Gold pQR711 was diluted with LB medium to an OD600 of 0.5 and sonicated. A 96-well plate containing 30 μl of increasing concentrations of ThDP across one axis and increasing concentrations of MgCl2 across the other, and 150 μl of cell sonicate in each well, was incubated for 15 min. Two control wells contained LB medium with either maximum cofactor concentration or no cofactors. Transketolase reactions were initiated by addition of 120 μl of substrate solution, to give final cofactor concentrations of 0–32 mM MgCl2, and 0–8 mM ThDP. l-Erythrulose was determined by rapid HPLC. Cofactor dissociation constants (K d) for Mg2+ and ThDP were determined at each concentration of the other cofactor, from the slope of a linear Eadie plot of TK activity vs activity/[cofactor].

Evaluation of the whole microplate-based assay process

Wells of a standard 96-deep-square-well (DSW) plate containing 900 μl LB-Amp medium were inoculated with a single colony of E. coli XL10-Gold pQR711 as above. Fermentation in the microplate was for 16 h at 900 r.p.m. at 37°C as above, and the final cell density in each well determined as previously in 96-RW plates. Plates were freeze-thawed once from −80°C to lyse the cells and 30 μl of each sample added to 120 μl of 0.25 × cofactor solution in each well of a 96-RW plate, incubated for 15 min and bioconversion reactions initiated by addition of 120 μl substrate solution. Final concentrations were 10% (v/v) culture lysate, 50 mM glycolaldehyde, 50 mM β-HPA, 50 mM Tris/HCl pH 7.5, 9 mM MgCl2, and 2.5 mM ThDP. After 10 min at 25°C, samples were quenched with four volumes of 0.2% (v/v) TFA and l-erythrulose determined by rapid HPLC. For saturation mutagenesis library mutants at residue A29X (obtained from Edward Hibbert, unpublished), the assay was carried out as above except that reactions were allowed to proceed for 1 h prior to quenching.

Results and discussion

The steps for the microplate-based HPLC assay of transketolase (TK) occur in the following order: colony picking; fermentation in microplates; cell lysis; addition of cofactors and substrates; HPLC analysis. Each step has been optimised sequentially to maximise the total activity per well for an improved biocatalytic process and also for sensitivity when assaying potentially less reactive substrates. Throughout the discussion, we indicate SD for the error associated with single value measurements, and the signal improvement thresholds required to achieve Z-factor scores (Zhang et al. 1999) of 0 and 0.5 which indicate increasing statistical significance.

Rapid HPLC method

A previous 35 min HPLC method for TK activity (Mitra et al. 1996) was modified in our ‘standard HPLC assay’ with a single 300 mm column for a cycle time of 11 min. For an even faster method we used an equivalent 50 mm guard column (Polymer Laboratories Ltd.) containing a similar stationary phase. A flow rate of 2 ml min−1 permitted 25 mM β-HPA and 25 mM l-erythrulose to be determined (Fig. 1) with retention times of 0.53 min for β-HPA and 0.66 min for l-erythrulose. The other substrate used in the reaction, glycolaldehyde, has a similar retention time to l-erythrulose, but has a negligible absorbance signal at 210 nm and so does not affect the determination of l-erythrulose. The cycle time of 1.2 min obtained by this method enables up to 1200 samples to be screened each day. This is almost 10-fold faster than our standard 11 min protocol (Micheletti et al. 2006) and 30-fold faster than the previously published 35 min HPLC protocol and is, therefore, suitable for the screening of mutant libraries generated by directed evolution, or for the rapid assessment of a matrix of biocatalytic process conditions.

Fig. 1
figure 1

Rapid HPLC analysis of components of the β-HPA and glycolaldehyde to l-erythrulose reaction. Chromatograms of 25 mM β-HPA (solid), 25 mM glycolaldehyde (dotted), and 25 mM l-erythrulose (dashed) are superimposed. The retention times of β-HPA and l-erythrulose are 0.53 min and 0.66 min, respectively. The retention time for glycolaldehyde is also 0.66 min, but has negligible absorbance at 210 nm

Full size image

The proportionality and accuracy of the rapid HPLC assay was determined with calibration curves for l-erythrulose, β-HPA, and an alternative substrate, pyruvate (Fig. 2). R 2 values of 0.9960, 0.9999, and 0.9996 for linear curve fits of 0 to 25 mM l-erythrulose, β-HPA, and pyruvate, respectively, demonstrate that the method is highly accurate even at low concentrations. The SD achieved at the lowest concentrations measured using this method was 1.9%. As expected, this is greater than the 0.1% SD previously obtained with the standard HPLC assay (Micheletti et al. 2006). Overall, a 148-fold range of activities from 0.0052 to 0.77 mM min−1 were measured (0.05–8 mM product) with the rapid HPLC assay as discussed below.

Fig. 2
figure 2

Calibration curves for transketolase reaction components separated by rapid HPLC. The relationships between concentration and UV detector response (210 nm) are linear for l-erythrulose (●), β-HPA (◯), and pyruvate (▾) with R 2 values of 0.9960, 0.9999, and 0.9996, respectively

Full size image

Selection of host strain

Transketolase from E. coli has been previously overexpressed at high levels in E. coli strain JM107, reaching 23–40% of the total protein (French et al. 1995; Hobbs et al. 1996). We have tested the plasmid pQR711 also in the alternative E. coli host strains: BL21-Gold(DE3) designed to decrease proteolysis of foreign proteins; and both JM109 and XL10-Gold, designed to improve plasmid stability. Initial transketolase reaction velocities, from triplicate samples in a 96-well microplate, for sonicated cultures (at OD600 = 0.25) of E. coli JM107, BL21-Gold(DE3), JM109, and XL10-Gold, each with and without pQR711 are shown in Fig. 3. Using 100 mM β-HPA and 100 mM glycolaldehyde as substrates, BL21-Gold(DE3) pQR711 had transketolase activity of approx 6-fold lower (0.063 ± 0.001 mM min−1) than JM107 pQR711 (0.345 ± 0.004 mM min−1), and JM109 pQR711 (0.337 ± 0.005 mM min−1), whereas XL10-Gold pQR711 produces over twice the total activity (0.78 ± 0.01 mM min−1). As expected, the wild-type strains produced a low activity (0.0052–0.0082 ± 0.0001 mM min−1) due to the small amounts of endogenous TK. BL21-Gold(DE3) is designed to maximise expression of foreign proteins through decreased proteolysis (ompT & lon genotypes). However, this is unlikely to improve the expression of E. coli TK as it is a housekeeping enzyme tolerant to proteolysis in its native host. The improved expression in JM107, JM109 and XL10-Gold reflects the greater stability and copy number of the plasmid in these strains (French et al. 1995) relative to BL21-Gold(DE3), resulting from the endA1 genotype in all three, and the recA1 genotype in JM109 and XL10-Gold.

Fig. 3
figure 3

Transketolase activity obtained in sonicates of various host strains. Cultures of E. coli JM107, BL21-Gold(DE3), JM109, XL10-Gold, and the corresponding pQR711 harbouring strains were sonicated and analysed for transketolase activity at 0.25 ODU culture, 100 mM β-HPA, 100 mM glycolaldehyde, 9 mM MgCl2, 2.5 mM ThDP, in 50 mM Tris/HCl, pH 7.5, 25°C. l-Erythrulose was determined after 10 mins reaction by standard HPLC. Wild-type host strains (black columns) and pQR711 containing strains (white columns) are compared. Error bars show the SD of the mean from three replicates

Full size image

The 2-fold greater expression level of TK in XL10-Gold compared to JM107 potentially allows us to detect twice as much product from a low turnover reaction in enzyme library screens, increasing the overall sensitivity of any rapid TK screen using this host strain. As XL10-Gold is designed for library cloning with high transformation efficiency and plasmid stability, it also provides an ideal host for creating directed evolution libraries of TK, and potentially also for large-scale bioconversions. A typical SD of 2.5% was obtained from these measurements, which include errors in reaction sample preparation and activity measurement by standard-HPLC.

Microwell fermentations

To take maximum advantage of the increased TK expression level obtained in XL10-Gold, we have evaluated a number of parameters that influence the growth of cells in microplate fermentations. Previous studies have shown that for 96-well microplate fermentation of E. coli strains, growth kinetics similar to those in shake-flasks can be obtained when using the deep-square well (DSW) geometry for maximum agitation and aeration (Micheletti et al. 2006). For improved library culture handling, minimised sample requirements, and increased throughput we have also evaluated these parameters for 384-square well (384-SW) microplates and standard 96-round well (96-RW) microplates, for comparison. The effects of plate geometry and agitation speed on the average cell densities as determined by the OD600, are shown in Fig. 4. For comparison, the cell density obtained in a 50 ml shake-flask culture agitated at 220 r.p.m. at 37°C is also shown. The highest cell densities were obtained in 96-DSW, or 96-RW plates, where the final OD600 was 4.09 ODU for 96-DSW plates at 900 r.p.m. For 96-RW plates, the final OD600 increased with agitation speed from 4.14 ODU at 600 r.p.m., to 4.15 ODU at 1000 r.p.m., and 4.53 ODU at 1400 r.p.m. This latter value was close to that achieved in a shake-flask (4.86 ODU). For the 384-SW plates, lower final cell densities were obtained than for the 96-well plates, though it also increased with agitation speed from 1.48 ODU at 600 r.p.m., to 1.91 ODU at 1000 r.p.m., and 3.41 ODU at 1400 r.p.m. (the maximum speed available using the Variomag plate shaker). The lower cell densities in 384-SW plates, most likely results from the small volume and high surface tension at the relatively large surface contact area giving rise to decreased liquid agitation and aeration at the air-liquid interface. The level of expression of TK obtained in the 384-SW plates is sufficiently close to that of shake-flasks to justify using this format for cases requiring improved throughput and minimised sample volume, though for the highest sensitivity 96-DSW plates are the format of choice.

Fig. 4
figure 4

The effects of plate format and agitation speed on the final OD600 of pooled microplate fermentations of E. coli XL10-Gold pQR711 in LB-Amp medium, pH 7, 37°C. Plate formats used: 96-deep-square-well (96 DSW); 96-round-well (96 RW); 384-square-well (384 SW)

Full size image

The benefits of robotic systems for reduced human errors, improved reproducibility and lower data scattering are already known (Lamsa et al. 2004; Nealon et al. 2005). It is useful, however, to determine the well-to-well variability and the presence of any so-called edge effects, to identify the confidence limits of measurements taken from single wells. We have assessed the reproducibility and data scattering obtained in the 96-DSW, and 384-SW format, under the conditions above that result in the highest observed cell density. The cell densities obtained in each well of a 384-SW plate, agitated at 1400 r.p.m. under the same conditions, are shown in Fig. 5. It can be seen that a similar cell density is obtained independently of the position of the well on the plate, and that any variation is due to random error, except wells A10 and I9 which did not grow at all. The data approximate to a random normal distribution with a SD of 0.3 ODU, or 8.8% (Table 1). Their mean value OD600 was 3.41 ± 0.015 ODU, where the error value refers to the SD of the mean. The variation of cell densities obtained using 96-DSW plates followed a similar distribution with a SD of 12% (Table 1). While these random errors will contribute to the error associated with the measurement of biocatalyst activities from a single well, they could in theory be reduced by normalising the activity measurements with the optical densities obtained in each well. This assumption will be analysed in more detail later.

Fig. 5
figure 5

Well-to-well variation of final fermentation cell densities measured by OD600, for E. coli XL10-Gold pQR711 grown in 60 μl LB-Amp medium, at pH 7, 37°C, in a 384-well microplate agitated at 1400 r.p.m. There is no discernable pattern to the growth. Wells A10 and I9 failed to inoculate

Full size image
Table 1 Formats, volumes, methods of analysis used and SD for each process step
Full size table

Evaluation of cell lysis methods

Cell lysis is normally required to release TK from the cell and allow access to the substrates. Maximising the availability of the expressed TK increases the sensitivity of the rapid HPLC assay to low turnover reactions. The effects of various cell lysis methods on the final activity of transketolase are shown in Fig. 6 for an overnight culture of E. coli XL10-Gold pQR711. Samples were analysed at 1 ml scale, equivalent to 96-DSW cultures. Sonication is the most effective method for obtaining TK activity (0.74 ± 0.01 mM min−1). Interestingly, an untreated culture of XL10-Gold pQR711 had approximately two-thirds (0.45 ± 0.01 mM min−1) of the activity of the fully sonicated culture. This background activity possibly results from cell lysis during the fermentation, or is potentially due to the substrates entering the intact cells.

Fig. 6
figure 6

Comparison of lysis methods for E. coli XL10-Gold pQR711, at 1 ml scale. Aliquots of a 3.4 ODU culture were lysed by different methods, and analysed for transketolase activity at 0.25 ODU culture, 100 mM β-HPA, 100 mM glycolaldehyde, 9 mM MgCl2, 2.5 mM ThDP, in 50 mM Tris/HCl, pH 7.5, 25°C. l-Erythrulose was determined after 10 min reaction by standard HPLC. (x% BB): x% (v/v) BugBuster. (F/T × y): y cycles of freezing at −80°C and thawing at 25°C. Error bars show the SD of the mean for three samples. Standard deviations are given in Table 1

Full size image

Addition of a chemical lysis reagent to the cells provides another potential option. The lysis reagent, Bugbuster, should be used neat to resuspend cells according to its manufacturers, but this would require a centrifugation or filtration step that is best avoided for the simplest automation and maximum reproducibility. Therefore, we investigated its direct addition to cultures. Such an addition requires a compromise to be reached between increasing the amount of total TK released upon addition of BugbusterTM, and dilution of the final enzyme activity. Assuming that sonication releases 100% of available TK activity, then the maximum activity expected after dilution at 20% (v/v) is 0.6 mM min−1, and 0.56 mM min−1 at 25% (v/v). In fact, a 20% (v/v) addition of Bugbuster achieved an activity of 0.51 ± 0.01 mM min−1 after 30 min and 0.54 ± 0.01 mM min−1 after 60 min. Consequently, this was the maximum concentration of Bugbuster studied. Incubation with 10% (v/v) Bugbuster for 30 min yielded a TK activity of 0.49 ± 0.01 mM min−1, rising slightly to 0.54 ± 0.01 mM min−1 after 1 h.

While it is simplest to use the cultures without treatment, it is also convenient to store microplates at −80°C until they are thawed when required, and so the use of one freeze-thaw cycle is preferred as the standard procedure. A single cycle of freeze-thaw from −80°C increased the TK activity to 0.51 ± 0.01 mM min−1. This process was repeated but the gains were not very large for the second cycle, which increases the activity to 0.57 ± 0.01 mM min−1, and no further gains were obtained by a third cycle which gave an activity of 0.56 ± 0.01 mM min−1.

Overall, the freeze-thaw cycles gave similar final activity results to incubation with Bugbuster. Well-to-well variations represented by SD for all lysis steps averaged 3.3%, though these also include variation that is obtained for the bioconversion steps and measurements using the standard HPLC method. This latter variation was found to be 2.5% during the analysis of different host strains (Table 1). However, the use of Bugbuster addition gave an average SD of 3.7% on the final activity, whereas the use of freeze-thaw cycles gave an average SD of 2.8%. Overall, freeze-thaw cycling is the least invasive method for cell lysis, does not require a liquid-volume addition which would otherwise contribute an additional random error to the final activity, and is convenient for microplate storage.

Process screening for optimisation of cofactors

The effects of buffer composition, pH, cofactor concentration and holo-enzyme reconstitution time have all been studied previously for the purified transketolases from yeast (de la Haba et al. 1955; Datta et al. 1961; Kochetov et al. 1975) and from E. coli (Sprenger et al. 1995). The dissociation constants for ThDP and Mg2+ have been previously observed to be co-dependent (Sprenger et al. 1995; Selivanov et al. 2004), where that for ThDP was 8 μM at 0.1 mM Mg2+, and that for Mg2+ was 8.7 μM at 0.1 mM ThDP, when using 1.9 nM E. coli TK at pH 8.5. Saturation is typically considered to be achieved by the addition of 2.5 mM ThDP and 9 mM Mg2+ for E. coli TK (Kobori et al. 1992; Sprenger et al. 1995; Bongs et al. 1997; Brocklebank et al. 1999). However, the concentration of TK used in biocatalytic conversions is considerably higher than the 1.9 nM studied previously, with a typical concentration of 2 mg ml−1 (25 μM) obtained in cultures of JM107 (Hobbs et al. 1993, 1996), and our results indicating that the XL10-gold strain produces up to 2.3-fold greater concentrations of TK relative to JM107. This observation along with the possibility of widely varying endogenous concentrations of Mg2+ and ThDP between the lysates of different E. coli strains, and the potential for other lysate components to sequester the added cofactors, led us to re-evaluate the cofactor requirements for achieving saturation of TK, and hence maximum activity under biocatalytic process conditions.

The SD for activities measured in this cofactor screen was only 4%. This equates to a Z-factor of 0.5 with only a 1.5-fold improvement in activity over that at which no cofactors were added, demonstrating the sensitivity of the screen to process changes. The effects of adding various concentrations of the cofactors Mg2+ (0–32 mM) and ThDP (0–8 mM) upon the total activity of transketolase are shown in Fig. 7, for a sonicated culture of E. coli XL10-Gold pQR711 at a final cell density equivalent to an OD600 of 0.25. For the reaction between 100 mM of both β-HPA and glycolaldehyde at 25°C and pH 7.5, transketolase activity was dramatically enhanced by the addition of ThDP. The activity at 0 mM Mg2+ increased by 579% from 0.067 ± 0.003 mM min−1 to 0.46 ± 0.02 mM min−1 upon addition of just 0.5 mM ThDP, and then to 0.61 ± 0.02 mM min−1 with 8 mM ThDP. This pattern of behaviour was observed at all concentrations of Mg2+, and is consistent with previous observations for both yeast TK (Datta et al. 1961; Selivanov et al. 2004) and E. coli TK at much lower concentrations of enzyme (Sprenger et al. 1995).

Fig. 7
figure 7

The effect of varying cofactor concentrations on transketolase activity. A culture of XL10-Gold pQR711 was sonicated and analysed for transketolase activity at 0.25 ODU culture lysate, 100 mM β-HPA, 100 mM glycolaldehyde, 50 mM Tris/HCl, pH 7.5, 25°C, with various concentrations of ThDP and Mg2+ arrayed in a 96-well microplate. l-Erythrulose was determined after 10 min reaction by rapid HPLC

Full size image

The addition of Mg2+ also increased TK activity, though less dramatically than for ThDP, with initial increments in concentration having the most profound effect. Activity at 0.5 mM ThDP was increased by 24% from 0.46 ± 0.02 mM min−1 to 0.57 ± 0.02 mM min−1 upon addition of 0.5 mM Mg2+ to the reaction, and to 0.69 ± 0.03 mM min−1 with 32 mM Mg2+. However, at zero ThDP the addition of 32 mM Mg2+ only increased the activity by 4.5%. These observations are consistent with previous results (Datta et al. 1961) in which addition of ThDP alone partially activated the enzyme, whereas addition of Mg2+ alone did not. The highest transketolase activity (0.77 ± 0.03 mM min−1) was observed at the greatest concentrations used of 8 mM ThDP and 32 mM Mg2+. By contrast, biocatalytic transformations with transketolase typically use up to 2.5 mM ThDP and 9 mM Mg2+ (Kobori et al. 1992; Sprenger et al. 1995; Bongs et al. 1997; Brocklebank et al. 1999). In Fig. 7 we can see that these concentrations are sufficient to obtain 96.8% of the maximum observed activity.

The data obtained from this microplate-based cofactor study can be used to obtain apparent dissociation constants. True dissociation constants cannot be obtained because there is already some ThDP and Mg2+ present in the cell lysate resulting in an initial quantity of holo-TK. Furthermore, the added cofactors can potentially be sequestered by other components of the cell lysate. However, the apparent dissociation constants are still a useful way of quantifying the association of cofactors under biocatalytic process conditions. The apparent dissociation constants obtained for ThDP at each concentration of Mg2+ are shown in Fig. 8a, whereas those for Mg2+ at each concentration of ThDP are shown in Fig. 8b. In both cases the apparent dissociation constant decreases as the concentration of the other cofactor is increased. The lowest apparent dissociation constant for ThDP (at above 2 mM Mg2+) is 0.057 ± 0.003 mM, which is 7.5-fold higher than that reported previously for purified TK (Sprenger et al. 1995). Similarly, the lowest apparent dissociation constant for Mg2+ (at above 4 mM ThDP) is 0.078 ± 0.009 mM, which is 9-fold higher than reported previously for purified TK. In both cases the increase in apparent dissociation constant over that determined for purified enzyme, highlights the need to re-evaluate cofactor binding under the biocatalytic process conditions to be used which potentially allow sequestration of the cofactors by other cellular components.

Fig. 8
figure 8

Dissociation constants of (a) ThDP determined at each concentration of Mg2+ using the rapid HPLC assay. (b) Mg2+ determined at each concentration of ThDP using the rapid HPLC assay. Both sets of data were obtained using sonicated E. coli XL10 pQR711 at 0.25 ODU, 100 mM β-HPA and 100 mM glycolaldehyde in 50 mM Tris/HCl, pH 7.5, 25°C

Full size image

The cofactor analysis in Fig. 7 also enabled us to determine the relative levels of active holo-TK and inactive apo-TK found in the lysate of the XL10-gold strain. Previously in JM107, 15% of the possible TK activity was observed before the addition of cofactors suggesting 85% of the enzyme was in the apo form (Mitra et al. 1996). Comparison of the activity of TK in the lysate of XL10-gold before cofactor addition (0.067 mM min−1), to the maximum activity of 0.77 mM min−1 observed at 8 mM ThDP and 32 mM Mg2+, suggests that the XL10-gold lysate contains just 8.7% of TK in the holo form.

Evaluation of the whole microplate-based assay process

The individual steps of cell fermentation, cell lysis, bioconversion, and rapid HPLC were each examined above, along with their SD which reflect well-to-well variation. Standard deviations and Z-factors are important factors to consider in microplate-based applications where decisions are often made using single sample measurements for each enzyme variant or process condition. The SD obtained for each step of the microplate-based process are 12% for 96-DSW fermentation and 1.9% for 96-RW rapid HPLC (Table 1). Standard deviations were also obtained from alternative starting points. For example, the SD of 3.3% for 96-DSW (1 ml) cell-lysis also includes random errors from preparation of 96-RW bioconversions and the standard HPLC analysis. Likewise, the SD of 4% for 96-RW cofactor analysis (bioconversion) also includes the random error (1.9%) from the rapid HPLC analysis. It is useful to compare these random errors to those accumulated through the entire microplate-based process.

The whole process was performed using the 96-DSW format for fermentation (obtained from above) and cell lysis by a single freeze-thaw step. Bioconversions for lysed cells, cofactor and substrate in the 96-RW format were analysed using the rapid HPLC method. Whereas the previous analyses of host cells, cell lysis and cofactor concentrations were all performed using 150 μl of a 6.8-fold diluted cell culture or lysate in a final volume of 300 μl, the microplate-based process uses just 30 μl of undiluted cell lysate in the same final volume. The smaller sample volume was necessary to achieve a similar overall dilution of TK activity in the bioconversion step and ensure that the reaction kinetics can still be accurately determined at the 10 min time-point, even for conditions or potential enzyme variants which significantly improve or reduce the rate.

The well-to-well SD for the fermentation was 12% as discussed above. However, the bioconversion and measurement by rapid HPLC on these samples gave an accumulated well-to-well SD of 35%. Such an increase in random error is perhaps unexpected when the previous bioconversion steps gave a random error of only 4%. This dramatic increase is most likely due to a variation in enzyme expression between the cultures in each well. As discussed above, activity data are often normalised by dividing by the individual cell-density measurements, under the assumption that the resulting specific activity will have greatly reduced associated errors. However, the SD for our normalised activity data (activity/OD600) actually increased to 42% (Table 1), which further implies a poor correlation between the cell density and the expressed level of TK enzyme within the cell culture. Until a suitable rapid method can be devised for specifically determining each TK enzyme concentration, the total activity data are the most useful for rapid screening purposes. A well-to-well variation on total activity of 35% equates to a Z-factor of 0 for a 3-fold improvement over the average wild-type activity. This is useful only for elimination of the lowest performers prior to a more accurate analysis of the activity and enzyme concentration of selected variants to positively identify improved specific activities. However, a Z-factor of 0.5 represents statistical significance, and is obtained where improvements over wild-type activity are at least 5-fold.

Application of the screen towards a transketolase mutant library

The whole microplate-based process was applied to a saturation mutagenesis library of transketolase containing 90 randomly picked colonies representing 20 possible protein variants, three blank wells and three wild-types. The fraction of l-erythrulose formed from 50 mM β-HPA and 50 mM glycolaldehyde for each colony is shown in Fig. 9. The wild-type colonies were shown to group well at a fractional conversion of 0.31 ± 0.008, while the variants were distributed from 0 to 0.44. The top performing mutants were sequenced and found to contain essentially only one mutant type, shown as triangles, which clustered well at a fractional conversion of 0.42 ± 0.01. Using this mutant as the positive control, gave a Z-factor of 0.49 which is statistically significant. The improvement in Z-factor for this screen over the previous example is likely to have resulted from increasing the reaction time from 10 min to 1 h, yielding a more accurate measurement of the conversion of substrates to product.

Fig. 9
figure 9

Rapid HPLC screening of a library obtained by saturation mutagenesis of a single residue in TK. Variants were expressed in E. coli XL10 pQR711 and assayed after 1 h reaction with 50 mM β-HPA and 50 mM glycolaldehyde, in 50 mM Tris/HCl, pH 7.5, 25°C. (♦) wild-type negative control and (Δ) positive control mutant gives Z-factor of 0.49

Full size image

Conclusions

A rapid microplate-based HPLC assay for transketolase was established that provides a direct determination of the β-hydroxypyruvate substrate and erythrulose product suitable for directed evolution, and rapid screening of biocatalytic process conditions. The assay can be used to determine TK activity with a throughput of up to 1,200 samples per day, and the well-to-well variation from the HPLC measurement was just 1.9% for the lowest activities measured.

For application to process screening in microplates, the assay consists of addition of a common cell lysate to each well of a microplate, followed by addition of cofactors and substrates, and the determination of activity by rapid HPLC. The well-to-well variation measured as SD in 96-well microplates for this method was found to be only 4% (Z-factor of 0.5 for 1.5-fold improvement in activity) making it highly suitable for rapid preliminary screening of process conditions with enzyme obtained from a single culture, prior to a more detailed characterisation of selected conditions. For example, the assay was used to determine the apparent dissociation constants for each of the two cofactors required by transketolase to within 5–11% SD of the mean.

For application to the directed evolution of transketolase, the assay also includes colony picking and microplate-based fermentation. The well-to-well variation of the complete assay process gives a SD of 35% on erythrulose product concentration, which enables single measurements to identify enzyme variants with at least three- and 5-fold improvements over the wild-type activity, for Z-factors of 0 and 0.5 respectively. The Z-factor of 0 would require at least one repeat of the screen, whereas a Z-factor of 0.5 already represents a statistically significant result. Improvement of the method by increasing the reaction time to 1 h led to an increase in the Z-factor to 0.49 for a positive control that has only 1.4-fold greater activity than the wild-type enzyme.

Each step of the microplate-based assay was optimised for obtaining the maximum transketolase activity per well. The host strain producing the greatest amount of TK activity was XL10-gold, which produces 2.3-fold more TK than the previously used JM107 strain. Conveniently, XL10-gold has greater plasmid stability than the other strains, suitable for large scale biotransformations and long-term storage, and also a high transformation efficiency which makes it a good choice for directed evolution library construction. The use of a single freeze-thaw cycle was sufficient for cell lysis and minimises sample-to-sample variation by avoiding additional liquid transfer steps. Freeze-thawing also provides the convenience of freezing samples until it is possible to measure their activity.