Engineering thermal stability and solvent tolerance of the soluble quinoprotein PedE from Pseudomonas putida KT2440 with a heterologous whole‐cell screening approach

Summary Due to their ability for direct electron transfer to electrodes, the utilization of rare earth metals as cofactor, and their periplasmic localization, pyrroloquinoline quinone‐dependent alcohol dehydrogenases (PQQ‐ADHs) represent an interesting class of biocatalysts for various biotechnological applications. For most biocatalysts protein stability is crucial, either to increase the performance of the protein under a given process condition or to maximize robustness of the protein towards mutational manipulations, which are often needed to enhance or introduce a functionality of interest. In this study, we describe a whole‐cell screening assay, suitable for probing PQQ‐ADH activities in Escherichia coli BL21(DE3) cells, and use this assay to screen smart mutant libraries for increased thermal stability of the PQQ‐ADH PedE (PP_2674) from Pseudomonas putida KT2440. Upon three consecutive rounds of screening, we identified three different amino acid positions, which significantly improve enzyme stability. The subsequent combination of the beneficial mutations finally results in the triple mutant R91D/E408P/N410K, which not only exhibits a 7°C increase in thermal stability but also a twofold increase in residual activity upon incubation with up to 50% dimethyl sulfoxide (DMSO), while showing no significant difference in enzymatic efficiency (k cat/KM).


Introduction
Dehydrogenases are important industrial enzymes with biotechnological applications ranging from the selective oxidation of alcohols and aldehydes to the regioselective oxidation of polyols, the production of enantiopure alcohols via kinetic resolution or the oxidation of prochiral compounds (Patel, 2004;Liese and Seelbach, 2006;Yakushi and Matsushita, 2010;Hollmann et al., 2011;Romano et al., 2012). Among alcohol dehydrogenases (ADHs), the pyrroloquinoline quinone (PQQ)-dependent ADHs are of particular interest for several reasons. First, they can be coupled to electrodes, rendering them suitable for biosensor and fuel-cell applications (Igarashi et al., 2004;Treu and Minteer, 2008;Takeda et al., 2013;Scherbahn et al., 2014;Guo et al., 2016). Second, these enzymes are localized within the periplasm of Gram-negative cells, thus eliminating the need for energy-driven transport of target chemicals into the cytoplasm. Together with the irreversible nature of the catalyzed reaction, this leads to unhindered accumulation of oxidized products outside of the cells. For living biocatalysts, this can be of great advantage as it minimizes toxic effects of substrates and corresponding oxidation products (Acharya and Manning, 1983;Huang et al., 2005). Third, PQQ-dependent enzymes do not interfere with the intracellular NAD + /NADH and NADP + /NADPH homeostasis, which is beneficial to maintain metabolic functionality of the biocatalyst (Matsushita et al., 2002;Chen et al., 2014). Last, the recent identification of specific PQQ-ADHs which utilize lanthanides as metal cofactor has fuelled the idea of using these enzymes or the corresponding bacterial hosts in biomining and recycling processes of rare earth metals (Skovran and Martinez-Gomez, 2015;Martinez-Gomez et al., 2016;Wehrmann et al., 2017).
For many biocatalytic applications, enzyme stability is a key criterion (Polizzi et al., 2007;Woodley, 2013). This is particularly true for enzymes located in the periplasmic space or enzymes used as purified entities outside of the protective context of the cell. Additionally, enzyme stability has also been associated with increased evolvability and mutational robustness, which is important for the development of biocatalysts with new and/or specific functionalities (Bloom et al., 2006;Bloom and Arnold, 2009;Tokuriki and Tawfik, 2009).
As a first step towards establishing soluble PQQ-ADHs as novel tools for biocatalysis, the main objective of this study was hence to increase the stability of our model quinoprotein PedE (PP_2674) from Pseudomonas putida KT2440. Different methods have been described to rationally engineer increased stability of proteins. These include the introduction of disulphide bonds (Yin et al., 2015), identifying mutations based on free energy calculations (Schymkowitz et al., 2005;Kellogg et al., 2011), protein surface charge optimization (Gribenko et al., 2009), B-factor analysis (Reetz et al., 2006), homologous comparison (Lehmann et al., 2000) and combinations of the before-mentioned methods (Jochens et al., 2010;Cerdobbel et al., 2011;Blum et al., 2012;Wijma et al., 2014).
To engineer proteins for a desired function, suitable whole-cell assays are highly beneficial, as they allow efficient screening of mutant libraries without prior purification of the different protein variants. Notably, two recent studies reported whole-cell assays, allowing activity screens of native membrane-associated PQQ-ADHs using Gluconobacter and Methylobacterium strains (Peters et al., 2013;Vemuluri et al., 2017). However, to the best of our knowledge, a heterologous system suitable for efficient production, purification and engineering of soluble PQQ-ADHs has not been developed thus far.
With this study, we describe the engineering of the calcium-dependent quinoprotein PedE from P. putida KT2440 towards increased thermal stability. For this, a heterologous whole-cell screening assay based on the well-characterized protein expression host Escherichia coli BL21(DE3) was developed, which allows the functional expression, purification and comparable quantification of enzyme activities for different quinoproteins. Increased stability was achieved using a combinatorial strategy targeting flexible residues with smart libraries of naturally occurring amino acids at the corresponding positions. With this approach, several beneficial mutations were identified and, together with combinatorial variants thereof, biochemically characterized for their stability towards elevated temperatures and the presence of the model organic solvent dimethyl sulfoxide (DMSO).
With these experiments, we found activities for all tested PQQ-ADHs when 10 mM of ethanol was used as substrate (Fig. 1A). In contrast, control cells producing eGFP from the same expression vector showed no detectable enzyme activity. To further demonstrate the robustness and versatility of the whole-cell assay, PedE activities on four different substrates, namely methanol, ethanol, 2-phenylethanol and acetaldehyde, were determined ( Fig. 1B) and compared with activities obtained with the purified enzyme (Fig. 1C). Under both conditions, the highest specific activity was found with ethanol as substrate (413 AE 2 U gCDW À1 and 6.7 AE 0.3 U mg À1 ). When using purified enzyme, 2-phenylethanol was converted with 76 AE 6% followed by acetaldehyde with 71 AE 9% of the activity determined with ethanol. For methanol, the activity was less than a tenth of that determined for ethanol (5 AE 1%). Notably, a similar trend in enzyme activities with all substrates was observed when using the whole-cell assay, however, with slightly higher relative activities, namely 92 AE 2%, 84 AE 3% and 16 AE 1% for 2-phenylethanol, acetaldehyde and methanol, respectively. To show that assay activity measurements correspond to product formation, additional biotransformations with PedE were carried out in a whole-cell set-up and with the purified enzyme. These experiments revealed that phenylacetaldehyde is the main oxidation product of 2-phenylethanol in both experimental set-ups and that product formation only occurred in the presence of the electron mediator PMS and the terminal electron acceptor DCPIP (Table S3).

Identification of amino acid residues in PedE to increase thermal stability
To engineer increased thermal stability of the quinoprotein PedE, we aimed for a combination of the B-FITTER approach and the consensus method (Lehmann et al., 2000;Reetz et al., 2006), similar to a recent study by Jochens et al. (2010). The B-FITTER strategy uses the concept that temperature-mediated unfolding of protein structures (denaturation) is initiated at flexible residues within the protein scaffold and thus, rigidifying those positions can improve the stability of an enzyme (Reetz et al., 2006(Reetz et al., , 2010Yu and Huang, 2014). Consequently, we initially determined the most flexible residues within PedE. As no crystal structure of PedE was available, the B-values from the crystal structure of the PQQ-ADH ExaA of P. aeruginosa (sequence identity 88%; PDB entry: 1FLG) were derived using the B-FITTER tool (Reetz et al., 2006). The corresponding positions of ExaA in PedE were identified through a 3DM database consisting of about 12,000 sequences showing structural similarity to PedE (Kuipers et al., 2010). Of the residues with the highest B-values that were included in the 3DM database, residues D307, N310, E408, K352, N410 and R91 were chosen for mutagenesis (Table 1). In a second step, the most frequently occurring amino acids at these positions were selected to create small focused libraries from two subsets of the 3DM database that contained only sequences encoding for PQQ-dependent ADHs (1315 sequences) or sequences from extremophile organisms (111 sequences; see Table 1). Finally, a minimal degenerated codon, which allowed the substitution for all variants of interest, was used to generate five mutant libraries containing between five and ten variants.

Screening of mutant libraries and identification of beneficial combinatorial mutations
A total of about 200 clones were screened for increased thermal stability using the heterologous whole-cell assay Table 1. Composition of the small focused mutant libraries used in the study. Black dots mark the most abundant amino acids at the corresponding position in a subset of the 3DM database that only contained pyrroloquinoline quinone-dependent alcohol dehydrogenases (PQQ-ADHs). Red dots mark the amino acids frequently present in a subset of the 3DM database that contained only sequences of proteins derived from extremophile organisms. to guarantee 95% coverage of each mutant library. Upon three consecutive rounds of screening (see Material and Methods for details), seven variants from three different libraries were identified, which exhibited significantly increased thermal stability (two-tailed t-test, N = 5, a = 0.05; GRAPHPAD PRISM, version 7.03) upon 1 h incubation at 45°C (relative residual activity; Fig. 2). After sequence analysis, different amino acid substitutions at three positions were found. Three variants were identified at positions R91 (D, Q, E) and N410 (K, S, T) while only one variant was found at position E408 (P). The best mutants from each library were R91D, N410K and E408P with 41 AE 1.7%, 44 AE 1.3% and 69 AE 10% relative residual activities compared with 23 AE 2.3% for the wild-type allele. Notably, two of the three variants, namely R91D and N410K, were derived from the 3DM database subset containing only sequences from extremophile organisms. Furthermore, no variants with increased thermal stability at positions 307, 310 and 352 were identified.
To test whether a combination of the mutations would result in additional stability, the most thermostable variations were combined using mutant E408P as starting point. Data obtained with the wholecell activity assay showed that each consecutive mutation had an additive effect. While the best single mutant E408P exhibited a 2.3 AE 0.4-fold increased thermal stability, the R91D/E408P double mutant showed a 3.2 AE 0.5 and the R91D/E408P/N410K triple mutant a 4.0 AE 0.6-fold increased stability upon incubation at 45°C for 1 h compared with the wild-type allele (Fig. 3A).  Characterization of thermostable alleles as purified enzymes To verify the increased thermal stability of the PedE alleles, the C-terminally His-tagged single, double and triple mutants as well as the wild-type enzyme were purified by immobilized-metal affinity chromatography to visual homogeneity (Fig. S1). As the apo forms of the enzymes did not show any residual activity upon incubation at 45°C for 1 h (Fig. S2), a reconstitution step with PQQ and CaCl 2 was performed prior to the experiments. Interestingly, the reconstituted wild-type enzyme showed much higher residual activities upon incubation at 45°C for 1 h (87 AE 6%, Fig. S2) compared with the values determined for this enzyme in the whole-cell assay (23 AE 2%, Fig. 2). However, when the incubation temperature for the purified enzyme was increased to 65°C, a similar trend as in the whole-cell assay was observed. Under these conditions, steadily increasing thermal stabilities starting from the E408P single mutant (4.9 AE 1.3-fold), over the R91D/E408P double mutant (7.2 AE 1.8-fold), up to the R91D/E408P/N410K triple mutant (9.3 AE 2.6-fold) were observed (Fig. 3B). By determining the T 50 60 values, representing the temperature required to reduce the initial activity of the enzyme to 50% upon 1 h of incubation, we revealed that for the R91D/E408P/ N410K variant this value was increased by 7°C from 58°C to 65°C compared with the wild-type allele (Fig. 4A). In addition to the increased thermal stability, we further found that the R91D/E408P/N410K variant exhibited an increased stability towards the organic solvent dimethyl sulfoxide (DMSO), reflected by an about twofold increased relative residual activity after incubation with 20-50% DMSO for 1 h compared with the wild-type enzyme (Fig. 4B). However, no enzyme activity was detectable with PedE or the R91D/E408P/ N410K variant upon incubation with more than 60% DMSO. Finally, kinetic parameters of the wild-type enzyme and the R91D/E408P/N410K variant were determined with the model substrate ethanol. We found that although the maximum velocity V max of the wild-type enzyme was significantly (two-tailed t-test; a = 0.05; N = 3; GRAPHPAD PRISM, version 7.03) higher than  Table 2. Kinetic constants of the PedE wildtype and the R91D/ E408P/N410K triple mutant with ethanol as substrate. The maximum velocities (V max ), substrate affinities (K M ) and catalytic efficiencies (k cat /K M , were k cat is the turnover frequency per cofactor molecule) were derived from non-linear regression of specific activities to a Michaelis-Menten model. Kinetic constants are presented as best-fit values AE the standard errors. Significant differences between the enzymes are marked (*, two-tailed t-test, a = 0.05, N = 3; GRAPHPAD PRISM, version 7.03). that of the triple mutant (5.0 AE 0.2 U mg À1 vs 6.1 AE 0.2 U mg À1 ), no significant differences in the substrate affinities (K M ) or enzymatic efficiencies (k cat /K M ) were detected (Table 2).

Discussion
Screening mutant libraries remains one of the major bottlenecks in most protein engineering approaches. Thus, a robust assay for rapid activity screenings in whole cells based on the protein expression strain E. coli BL21 (DE3) was developed. After optimization of the assay conditions, the relative activities determined for the PedE enzyme with four model substrates were found to be very similar to those determined with the purified enzyme. It was further demonstrated that the whole-cell assay allows robust and comparable quantification of enzyme activities of PQQ-ADH enzymes from various organisms and with different cofactor dependencies. Together with the advantages of working with the laboratory workhorse E. coli BL21(DE3) as a screening host, permitting the use of standard protocols for efficient and effortless production and purification of plasmids and enzymes, these results show that the developed assay represents a suitable tool for rapid quantification of enzyme activities without the prior need for labour intensive and time-consuming cell disruption and protein purification procedures. For subsequent stability engineering, a combination of the B-FITTER and the consensus approach was used to generate small focused libraries consisting of the most frequently occurring amino acids at six positions of high flexibility (Lehmann et al., 2000;Reetz et al., 2006). From screening those libraries with the developed assay, we found that the introduction of conserved residues at the positions 91, 408 and 410 resulted in improved thermal stability of the enzyme. Notably, these residues are all within 20 A of the PQQ cofactor-binding site and on the same side of the b-barrel structure of the enzyme as the active site (Fig. 5). In contrast, positions 307, 310 and 352, for which no variants with improved thermal stability could be identified, are further distant from the active site and on the other side of the b-barrel structure. This suggests that rigidifying flexible residues close to the active site is more beneficial compared with more distant residues. A similar effect has also been reported for the lipases CalB of Candida antarctica and Lip1 of Candida rugosa (Xie et al., 2014;Zhang et al., 2016). The two studies suggest that for large proteins, mutating flexible residues located far from the active site, does not protect the active centre from heat-induced detrimental conformational changes. In this context, it is further worthwhile noting that we found that the purified apo-form of PedE is much less thermostable compared with the PQQ reconstituted form of PedE. This strongly indicates that binding of the cofactor to the active site is not only essential for the catalytic mechanism but also plays a crucial role in stabilization of the enzyme.
As described for many other enzymes, the combination of beneficial mutations had an additive effect and finally led to the PedE variant R91D/E408P/N410K, which exhibited an about 10-fold improved thermal stability upon incubation at 65°C and a 7°C increased T 50 60 of 65°C compared to 58°C for the wild-type allele. In addition, the R91D/E408P/N410K variant showed an improved stability in the presence of the hostile organic solvent DMSO. A similar correlation between temperature stability and solvent tolerance has also been reported for several other enzymes (Cowan, 1997;Hao and Berry, 2004;Reetz et al., 2010). Despite the fact that the underlying molecular mechanism of this correlation is not well understood, it is of particular interest for many biotechnological applications, where solvent tolerance is a key factor (Schmid et al., 2001;Bornscheuer et al., 2012). Comparison of kinetic parameters revealed that while a slight decrease in reaction velocity (V max ) from 6.1 to 5.0 U mg À1 was observed for the R91D/E408P/N410K variant when compared with the wild-type allele, no significant differences were found with respect to the substrate affinity (K M ) or enzymatic efficiencies (k cat /K M ). Hence, a dramatically more thermostable and solvent-tolerant mutant with only minor impairment on the enzyme kinetics at 30°C has been generated. However, as no screening for increased activity at 30°C was performed, it is very likely that by performing additional rounds of directed evolution, mutations that compensate for the decreased V max can be identified. Overall, in this study, we describe a heterologous whole-cell screening assay for enzyme activity measurements of soluble quinoproteins, and its subsequent use to identify different variants of the PQQ-ADH PedE from P. putida KT2440 with increased thermal stability. We further provide new insights into the impact of the quinone cofactor on the stability of PQQ-dependent enzymes and identify the combinatorial variant R91D/ E408P/N410K, which shows significantly improved thermal stability and tolerance towards the organic solvent DMSO, and can now serve as a starting point for future applications including the engineering towards novel enzymatic functions.

Strains and growth conditions
Escherichia coli BL21(DE3) was used as host for cloning and expression. If not stated otherwise, cells were routinely cultured at 37°C in liquid LB medium (Maniatis et al., 1982) on a rotary shaker at 180 rpm (HT Minitron, Infors AG, Einsbach, Germany). For plasmid maintenance and selection, 40 lg ml À1 kanamycin or 100 lg ml À1 ampicillin was supplemented to the medium.

Plasmid construction, library generation and evaluation
A detailed description of the construction of the broadhost-range expression vectors pJoe4036.1::exaA(His)6, pMW09 and pMW10 containing the exaA gene (GI: 15597178) from P. aeruginosa, pedE gene (GI: 26989393) and pedH gene (GI: 26989398) both from P. putida KT2440 with a C-terminal 6xHis-Tag that were used in this study can be found elsewhere (Chattopadhyay et al., 2010;Wehrmann et al., 2017). Small focused mutant libraries were generated using multichange isothermal mutagenesis (Mitchell et al., 2013). For this, the minimal degenerate codon combinations for the desired variance at each position was determined using MDC analyzer (Tang et al., 2014). For mutagenesis, selfcomplementary primers containing the desired degenerate codon in its centre, as well as primers that contained homologies to the plasmid pMWH09, were designed (see Table S1). Upon PCR, the two fragments and the HindIII and NdeI digested vector were assembled in a one-step isothermal assembly reaction, diluted threefold in MilliQ H 2 O and transformed into electrocompetent E. coli BL21 (DE3) cells (Gibson, 2011). To verify the desired diversity of the constructed small focused libraries, colonies of one plate from each transformed assembly reaction were resuspended in 2 ml MilliQ H 2 O with a Drigalski spatula, used for a plasmid extraction and submitted for sequence analysis. If the peak height distribution for the corresponding degenerate codon did not match the expectation (Acevedo-Rocha et al., 2015), potentially missing variants were either generated individually or the oversampling number was increased accordingly.

Whole-cell activity assay
For enzyme activity measurements of the different PQQ-ADHs, and variants thereof, a heterologous whole-cell assay based on E. coli BL21(DE3) was developed. For this, a previously described assay in Gluconobacter oxydans 621H was used as a starting point (Peters et al., 2013) and various parameters including amine source, buffer composition, PQQ, CaCl 2 or PrCl 3 supplementation, pH and cell density were optimized (see Table S2 and Fig. S3). In the final protocol, E. coli BL21(DE3) cells expressing the desired PQQ-ADHs (see below) were washed twice, resuspended in 100 mM Tris buffer (pH 8), and adjusted to an optical density between 1 and 4. The reaction buffer contained in a final volume of 250 ll in microtiter plates (96-well, PS, F-bottom; Greiner Bio-one, Frickenhausen, Germany): cell suspension 25 ll, Tris buffer (pH 8) 100 mM, pyrroloquinoline quinone (PQQ) 1 lM, 2,6-dichlorophenolindophenol (DCPIP) 150 lM, phenazine methosulfate (PMS) 150 lM, imidazole 25 mM, substrate 10 mM, and depending on whether a calcium-or lanthanide-dependent PQQ-ADH was tested either 1 mM CaCl 2 or 1 lM PrCl 3 . Due to substrate-independent background bleaching, the reaction mixture containing the cell suspension was incubated at 30°C for 30 min prior to substrate addition to allow the adjustment of a stable baseline. Enzyme activity was determined by following the reduction in the colorimetric dye 2,6-dichlorophenolindophenol (DCPIP, extinction coefficient e = 24.1 cm À1 M À1 at pH 8) at 600 nm within the first minute upon substrate addition in a microplate reader (POLARstar Omega, BMG Labtech, Ortenberg, Germany). Activities are displayed as units per gram cell dry weight (U gCDW À1 ), given that 1 l of cells at an OD 600 of 4 corresponds to 1 g cell dry weight, for better comparability.

Library screening
For the initial screen of the mutant libraries, one colony was used to inoculate 1 ml LB medium containing 10 mM Tris-HCl pH 7.5; 1 mM CaCl 2 ; 40 lg ml À1 Kanamycin; 0.6 lM pyrroloquinoline quinone (PQQ); 0.2% rhamnose in 2 ml 96-deep-well plates (Carl Roth, Karlsruhe, Germany) and incubated at 16°C and 180 rpm (HT Minitron, Infors AG, Einsbach, German). After 48 h, cells were harvested by centrifugation (4°C, 10 min, 3000 9 g), washed twice with 1 ml Tris buffer (100 mM, pH 8), and finally resuspended in 250 ll Tris buffer (100 mM, pH 8). 25 ll of the cell suspension were subsequently used to determine the initial enzyme activity (EA I ) using the whole-cell DCPIP assay with 10 mM ethanol substrate as described above. Another 20 ll cell suspension was diluted 1/10 in Tris buffer (100 mM, pH 8) and used to determine the optical density at 600 nm (OD 600 ). In addition, 100 ll of each cell suspension was incubated at 45°C in a thermal cycler (Mastercycler ep, Eppendorf AG, Hamburg, Germany) for 1 h and 25 ll were subsequently used for the quantification of the residual activity (EA R ) using the whole-cell DCPIP assay with 10 mM ethanol. From samples with an OD 600 > 1 which showed a higher relative residual activity at 45°C (EA R /EA I ) than samples of the wild-type variant, six independent clones were used for a second round of screening and analysed for statistically significant improvements compared with the wildtype allele (two-tailed t-test, a = 0.05, N = 6, GRAPHPAD PRISM (version 7.03), GraphPad Software, Inc., La Jolla, CA, USA. To exclude host effects, plasmids from variants with significantly increased stability were retransformed into E. coli BL21(DE3) and re-evaluated in a third round of screening for improved stability (two-tailed t-test, a = 0.05, N = 5, GRAPHPAD PRISM (version 7.03), GraphPad Software, Inc., La Jolla, CA, USA).
Expression, purification and characterization of wild-type enzymes and stability mutants C-terminally His-tagged PQQ-ADHs were expressed in E. coli BL21(DE3), purified using immobilized-metal affinity chromatography, and characterized for enzymatic activity with different substrates and substrate concentrations as described elsewhere (Wehrmann et al., 2017). As the purified apo-enzyme form of PedE did not show any residual activity after incubation at 45°C for 1 h, the minor changes during protein purification and enzyme stability measurements were made as follows: the buffer of the dialyzed enzyme solution was changed from 50 mM Tris buffer (pH 7.5) to 50 mM Tris buffer (pH 7.5) containing 10 lM PQQ and 10 mM CaCl 2 to allow reconstitution of the enzyme holo-form. To determine the thermal stability of the enzymes, 100 ll of a 100 lg ml À1 enzyme solution were incubated at various temperatures for 1 h and the residual activities were determined using an enzyme concentration of 10 lg ml À1 in the reaction buffer with 10 mM ethanol as substrate. The final concentrations of components within the assay solution and the procedure for quantifying enzymatic activities, including the preincubation, were identical for purified enzymes and whole cells. From plotting the relative residual activities against the incubation temperature, the T 50 60 values (temperature at which after 60 min incubation 50% residual activity is present) were determined. The stability in organic solvents was tested by incubating the reconstituted enzymes in varying concentrations of DMSO at 30°C for 1 h and subsequently measuring the residual enzyme activity with 10 mM ethanol as substrate as described above.

Enzyme kinetics measurements
The kinetic constants of the enzymes were determined as described previously (Wehrmann et al., 2017).

Homology model
The homology model of PedE was built using Swiss-Model (Biasini et al., 2014). As a template for model construction, the crystal structure of the PQQ-ADH ExaA of P. aeruginosa (PDB entry 1FLG; 88% sequence identity) was used. Model visualization was performed using PYMOL (Schrodinger LLC, 2015).

Biotransformations and metabolite quantification
Biotransformations were carried out in a similar reaction buffer as the whole-cell activity assay. The experiments were performed in a final volume of 500 ll and contained 100 mM Tris buffer (pH 8), 1 lM PQQ, 150 lM DCPIP, 150 lM PMS, 25 mM imidazole, 8 mM 2-phenylethanol (substrate) and either 50 ll (OD 600 of 3) of E. coli BL21(DE3) cells producing PedE or 10 lg ml À1 purified PedE enzyme. At several time points, samples were taken, filtered (0.2 lm) and acidified by adding two equivalents of a 50 mM H 2 SO 4 solution. Substrates and products were extracted with one equivalent of ethyl acetate, and the organic phase was subsequently analysed by gas chromatography using a GC-2010 system (Shimadzu, Kyoto, Japan) equipped with a HP-1 ms UI column (30 m 9 0.25 mm 9 0.25 lm; Agilent Technologies, Santa Clara, CA, USA) and flame ionization detection. Chromatography was performed with H 2 as carrier gas at a linear velocity of 30 cm s À1 . The temperature programme started from 70°C and increased at a rate of 20°C min À1 to 140°C, followed by an increase at a rate of 40°C min À1 until 320°C, which was finally held for 2 min. The injected volume was 1 ll with a split ratio of 1:10. The measured retention times for the 2-phenylethanol, phenylacetaldehyde and phenylacetic acid standards were 3.7, 3.2 and 4.5 min respectively. grant from the German Science Foundation (DFG; KL 2340/2-1).

Fig. S1
. SDS-PAGE analysis of production of PedE WT and stability mutants. Fig. S2. Relative residual activity of purified PedE wildtype protein upon incubation for 1 h at different temperatures with and without reconstitution with CaCl 2 and PQQ prior to incubation. Fig. S3. Specific activity of E. coli BL21(DE3) cells expressing PedE depending on cell density (OD 600 ) used. Table S1. Primers used in this study. Table S2. Overview of parameter optimization for the whole-cell activity assay using of E. coli BL21(DE3) cells expressing PedE. Table S3. Biotransformations (500 µl) with 8 mM 2-phenylethanol using E. coli BL21(DE3) cells producing PedE (A) or 10 µg ml À1 of purified PedE enzyme (B) in DCPIP assay solution at 30°C.