Thiamin diphosphate in biological chemistry: exploitation of diverse thiamin diphosphate-dependent enzymes for asymmetric chemoenzymatic synthesis


M. Müller, Institut für Pharmazeutische Wissenschaften, Albert-Ludwigs-Universität, Albertstraße 25, 79104 Freiburg im Breisgau, Germany
Fax: +49 761 203 6351
Tel: +49 761 203 6320


Thiamin diphosphate-dependent enzymes participate in numerous biosynthetic pathways and catalyse a broad range of reactions, mainly involving the cleavage and formation of C–C bonds. For example, they catalyse the nonoxidative and oxidative decarboxylation of 2-keto acids, produce 2-hydroxy ketones and transfer activated aldehydes to a variety of acceptors. Moreover, they can also catalyse C–N, C–O and C–S bond formation. Because of their substrate spectra and different stereospecificity, these enzymes extend the synthetic potential for asymmetric carboligations appreciably. Different strategies have been developed to identify new members of this promiscuous enzyme class and the reactions they catalyse. This enabled us to introduce solutions for longstanding synthetic problems, such as asymmetric cross-benzoin condensation. Moreover, through a combination of protein structure analysis, enzyme and substrate engineering, and screening methods we explored additional stereochemical routes that have not been described previously for any of these interesting enzymes.


benzaldehyde lyase


benzoylformate decarboxylase


enantiomeric excess




2-succinyl-5-enolpyruvyl-6-hydroxy-3-cyclohexene-1-carboxylate synthase


phenylacetyl carbinol


pyruvate decarboxylase


thiamin diphosphate



In 1832, Liebig & Wöhler [1] published a report on benzoin condensation, the homocoupling of two benzaldehyde molecules in the presence of cyanide, in the journal Annalen der Pharmacie. This is now acknowledged as the first named reaction in synthetic chemistry. It was 50 years before Emil Fischer [2] used this transformation for the cross-coupling of two different aldehydes, i.e. the condensation of benzaldehyde with furfural, resulting in a ‘mixed benzoin’. Afterwards, many applications of this useful ‘Umpolung’ reaction were introduced. During the first half of the 20th century, Lapworth, Buck and Ide, and many others studied the detailed mechanisms and synthetic scope of these reactions.

Ugai et al. [3] published a report on thiazolium-catalysed benzoin condensation in a Japanese journal in 1943. Based on Lapworth’s work [4,5], Breslow resolved the mechanism behind this landmark transformation [6]. The demonstrated usefulness of the ‘classical’ cyanide-catalysed benzoin condensation and the versatility of the thiazolium organo-catalyst resulted in an unprecedented increase in new organo-catalytic transformations. In 1966, Shehan & Hunneman [7] introduced the first asymmetric benzoin condensation. The Stetter reaction, the conjugate Umpolung condensation of an aldehyde and an α,β-unsaturated carbonyl species, was introduced in the 1970s [8]. Triazolium-catalysed asymmetric transformations were established by Enders et al. [9], and Knight & Leeper [10,11] introduced novel bicyclic thiazolium salts for asymmetric benzoin condensation. During the last decade, numerous new heterazolium-catalysed transformations were set up. In a recent review of more than 200 publications, most of which were published between 1990 and 2007, Enders et al. [12] give a comprehensive in-depth overview of the progress of organo-catalysis by N-heterocyclic precatalysts.

Independent of the development of organo-catalytic transformations, thiamin diphosphate-dependent enzymatic transformations have also been explored. As early as 1921, the first modern biotechnological process on an industrial scale was invented, based on a bakers’ yeast whole-cell biotransformation [13,14]. The original process is still in use in the production of (R)-phenylacetyl carbinol [(R)-PAC], the precursor of (−)-ephedrine [15]. However, the structure of thiamin diphosphate (ThDP) was not published until 1937 [16]. A recent review by Kluger & Tittmann [17] summarizes the state-of-the-art with respect to enzymic and nonenzymic covalent intermediates.

inline image

Pyruvate decarboxylase (PDC), the enzyme responsible for enantioselective C–C bond formation, catalyses the nonoxidative decarboxylation of pyruvate to acetaldehyde as a main reaction. In a side reaction, an activated acetaldehyde is ligated with benzaldehyde in a benzoin condensation-like manner (Scheme 1).

Figure Scheme 1..

 Enzymatic formation of (R)-PAC used for the production of (−)-ephedrine [15].

The potential to catalyse C–C bond formation and/or cleavage is typical of ThDP-dependent enzymes, such as 2-keto acid decarboxylases, transketolases (TK) and benzaldehyde lyase [18–21]. With the rapid progress in molecular biology, many of the genes encoding for ThDP-dependent enzymes have been cloned and overexpressed, mostly in recombinant Escherichia coli strains. Hence, considerable amounts of purified enzymes have become available. Based on the known nonenzymatic Umpolung reactions, we successfully introduced, among others, enzymatic benzoin condensation [22,23], asymmetric cross-benzoin condensation [24], the racemic resolution of 2-hydroxy ketones via C–C bond cleavage [25], the synthesis of bis(2-hydroxy ketones) [26] and the enzymatic homocoupling of aliphatic aldehydes [27]. These transformations were predominantly performed with wild-type enzymes, for the most part with benzoylformate decarboxylase (BFD) from Pseudomonas putida [28,29], benzaldehyde lyase (BAL) from Pseudomonas fluorescens [30] and different 2-keto acid decarboxylases like PDC [15] and KdcA [31] (Scheme 2) [32]. Moreover, other ThDP-dependent enzymes like TK [33] and 2-succinyl-5-enolpyruvyl-6-hydroxy-3-cyclohexene-1-carboxylate synthase (MenD) [34] have been applied to similar C–C bond formations.

Figure Scheme 2..

 Examples of ThDP-dependent enzyme-catalysed transformations.

Directed evolution using random and site-directed mutagenesis have been successfully applied to solve limitations with respect to substrate range [35–37], steric hinderance [15,24,38], solvent tolerance [39] and stereochemical problems [36,40,41].

Thus, enzymatic ThDP-dependent transformations complement the toolbox of synthetic organic chemistry pioneered by Justus Liebig and Emil Fischer. However, the question remains, how to discover as yet unexplored transformations and how to gain access to new, uncharacterized ThDP-dependent enzymes. We exploited different strategies to identify new members and variants of this highly versatile enzyme class: (a) enzyme engineering and in vitro high-throughput screening; (b) in vivo growth selection; (c) structure analysis and molecular modelling; (d) biosynthesis as a model; and (e) substrate and reaction engineering.

These five methods complement each other and enabled us to develop solutions for as yet unexplored trajectories. For example, synthesis of (S)-PAC and its derivatives has never been observed for any ThDP-dependent transformation. Selective organo-catalytic synthesis of each of the four regioisomeric and enantiomeric products of aromatic–aliphatic cross-benzoin condensation is not feasible under thermodynamic control, because benzoin condensations are known to be reversible and hence result in the most thermodynamically stable product. Moreover, and most importantly, this diversity-oriented approach helped us to deal with the dilemma that ‘we will find only what we are looking for’ (according to Frances Arnold). For most ThDP-dependent enzymes, catalytic polyreactvity (commonly called catalytic promiscuity) has been shown or can be assumed; hence, restrictions are typically because of the constraints of our imagination, rather than limitations of the cofactor’s reaction space.

Enzyme engineering and in vitro high-throughput screening

In order to identify new enzyme variants and access a broad range of 2-hydroxy ketones, a high-throughput screening system based on triphenyltetrazolium chloride was established. Detection relies upon oxidation of the 2-hydroxy ketone to the diketone, thereby generating a red formazane dye (Scheme 3). This assay was originally established to identify improved enzyme variants for the production of PAC (Scheme 3) [42]. Afterwards it was expanded to a broader range of 2-hydroxy ketones [43] and the procedure was shortened [44].

Figure Scheme 3..

 Triphenyltetrazolium chloride assay to detect 2-hydroxy ketones.

Although there is clearly a lower detection limit for aromatic and mixed aliphatic/aromatic 2-hydroxy ketones of ∼ 0.5 mm, aliphatic acyloins such as acyloin, propioin and 5-hydroxy-2,7-dimethyl-octan-4-one can be also detected with lower detection limits of 0.5, 2 and 5 mm, respectively. However, the assay is not sensitive enough to detect 4-hydroxy-2,5-dimethyl-hexan-3-one, the ligation product of two isobutyraldehyde molecules (lower detection limit 40 mm). In order to shorten the procedure described originally and allow the detection of base-sensitive 2-hydroxy ketones derived from dimethoxyacetaldehyde, both the tetrazolium salt and 1 m NaOH are preferably added simultaneously.

Thus, the triphenyltetrazolium chloride assay is a versatile tool to fingerprint the substrate ranges of enzymes. It is highly reliable for purified enzymes and works well with crude cell extracts, if cell debris are removed. The assay is routinely used prior to detailed instrumental analyses.

To identify mechanistically important amino acids in the active site, the increase in the substrate range and altered stereoselectivity, several variants of pyruvate decarboxylase from Zymomonas mobilis (ZmPDC) and benzoylformate decarboxylase from P. putida (PpBFD) were generated using site-directed mutagenesis and directed evolution.

The PAC-forming activity of ZmPDC using acetaldehyde and benzaldehyde as substrates was significantly improved by site-directed mutagenesis of a bulky tryptophane residue in the substrate channel. The highly potent variant ZmPDCW392M was tested in a continuous enzyme–membrane reactor, giving space–time yields of 81 g·L−1·d−1 under nonoptimized conditions with an enzyme concentration of 1 mg·mL−1 and a residence time of 1 h [15,45]. Furthermore, position 472 (isoleucine) was determined to be a hot spot influencing enantio- and chemoselectivity [35,36].

PpBFD is an exceptional ThDP-dependent enzyme, because it catalyses the formation of (S)-2-hydroxy-1-phenylpropan-1-one (HPP) from benzaldehyde and acetaldehyde. This is one of the very few known S-selective reactions (92% enantiomeric excess [ee]) catalysed by this otherwise R-selective class of enzymes [46]. BFD further allows access to various (S)-HPP analogues based on the carboligation of acetaldehyde with different aromatic, heteroaromatic, cyclic aliphatic as well as olefinic aldehydes [28,29]. Hence, the selectivity for aromatic aldehydes as donor aldehydes is very high, which is also reflected in the relative activities of PpBFD concerning the formation of (S)-2-HPP (7 U·mg−1; 92%ee), (R)-benzoin (0.25 U·mg−1; 99%ee) and (R)-acetoin (0.01 U·mg−1, 35%ee) [27–29,36]. Depending on the substitution pattern of the aromatic ring, diverse HPP analogues are accessible in high yields and with good to high optical purity. The selectivity, activity and stability of PpBFD have been optimized using reaction engineering. The best results have been obtained by adjusting very low benzaldehyde concentrations in a continuous reactor [28,29]. Directed evolution yielded variants with an amino acid exchange in position of leucine 476 with enhanced ligase activity and stereoselectivity for the formation of (S)-HPP, improved stability toward organic solvents and an enlarged donor substrate range [38,39]. Site-directed mutagenesis studies yielded variants with increased carboligase activity toward benzoin derivatives (PpBFDH281A) [24], as well as an altered substrate range for the decarboxylation reaction (PpBFDA460I), which also showed improved enantioselectivity during carboligation of benzaldehyde and acetaldehyde [36].

Recently, a further hot spot in PpBFD was identified with leucine 461, a residue defining an S-pocket in the acceptor binding site during carboligation [40,53]. Shaping the S-pocket by site-directed mutagenesis of leucine 461 toward alanine and glycine resulted in variants which opened access to various (S)-hydroxy ketones with high enantioselectivity. The biochemical data show that both the perfect stabilization of the acceptor aldehyde, and also the interplay of donor and acceptor aldehydes fitting into the active site influence the chemo- and enantioselectivity of the biocatalysts (see below).

In vivo growth selection

A growth selection system was established using P. putida, which can grow on benzaldehyde as the sole carbon source. In contrast to P. putida strain ATCC12633, the origin of the well-characterized BFD [22,28,29], strain KT2440 is unable to grow on benzoylformate-containing selective media, but growth could be restored by expression in trans of genes encoding BFD. The selection system was used to identify three novel BFDs, two originating from a chromosomal library of P. putida ATCC12633 and the third from an environmental DNA library. The novel P. putida enzymes BfdB and BfdC exhibited 83% identity with BFD from P. aeruginosa and 63% with the enzyme MdlC from P. putida ATCC12633, whereas the metagenomic BfdM showed 72% identity with a putative BFD from Polaromonas naphthalenivorans. BfdC was overexpressed in E. coli and the enzymatic activity was found to be 22 U·mL−1 using benzoylformate as the substrate. These results clearly demonstrate that P. putida KT2440 is an appropriate selection host strain suitable for finding novel BFDs. In addition, this system should be applicable to the identification of a broad range of different biocatalysts, including industrially important enzymes such as BAL and hydroxynitrile lyases, which all catalyse the formation of benzaldehyde from appropriate substrates. This selection system constitutes a powerful tool to identify new biocatalysts from very large libraries and may also serve to provide experimental proof of the enzymatic activities of putative enzyme-encoding ORFs identified in numerous genome sequencing and metagenome projects [47].

P. fluorescens is able to grow on (R)-benzoin as the sole carbon and energy source because it harbours the enzyme BAL which cleaves the acyloin linkage, using ThDP as a cofactor. In the reverse reaction, the enzyme catalyses the carboligation of two aldehydes with high stereo- and substrate specificity. BAL is unique among the well-characterized ThDP-dependent enzymes because of its broad substrate range, which extends the synthetic potential for carboligations appreciably (see below). Furthermore, its potential to cleave acyloin linkages has not yet been observed with other enzymes and allows the kinetic resolution of racemic benzoins [25]. By contrast to BFD, BAL is strictly R-selective. The enzyme structure was determined by X-ray diffraction at 2.6 Å resolution [48], which was fundamental for the identification of the so-called S-pocket of ThDP-dependent decarboxylases (see below).

Structure analysis and molecular modelling

Superimposition of the crystal structures of BFDwt from P. putida [49], BAL from P. fluorescens [48], PDCs from Zymomonas mobilis (ZmPDC) [50] and Saccharomyces cerevisiae (ScPDC) [51], as well as the recently solved structures of PDC from Acetobacter pasteurianus (ApPDC) (pdb code 2vbi; D. Gocke, C. L. Berthold, G. Schneider and M. Pohl, unpublished results) and KdcA from Lactococcus lactis (LlKdcA) [52] gave profound insights. Although there is no structural element called the S-pocket visible in BAL, the S-pockets of the other enzymes increased in the order PpBFD < LlKdcA < ZmPDC/ScPDC < ApPDC. By shaping the active site of BFD through the use of rational protein design, structural analysis and molecular modelling, optimal steric stabilization of the acceptor aldehyde in the S-pocket was identified as the predominant interaction for adjusting stereoselectivity. Our studies revealed leucine 461 as a hot spot for stereoselectivity in BFD. Replacing leucine with alanine or glycine resulted in variants that catalyse the S-enantioselective addition of larger acceptor aldehydes, such as propanal with benzaldehyde and its derivatives – a reaction not catalysed by the wild-type enzyme (Scheme 4). Crystal structure analysis of the variant BFDL461A supports the modelling studies [53].

Figure Scheme 4..

 BFDL461A catalysed S-selective synthesis of 2-hydroxy ketones.

Biosynthesis as a model

From biosynthesis to biocatalysis

To date, BAL has been found only in P. fluorescens Biovar I. This strain is able to utilize 2-hydroxy ketones such as anisoin as a sole carbon source. The enzyme responsible for this catabolic activity is BAL which catalyses cleavage of the aromatic acyloins to aldehydes; the aldehydes are further catabolized in the β-ketoadipate pathway (see above). Although BAL was first identified by its lyase activity, we have previously reported that it is also able to catalyse the reverse reaction [25]. Subsequently, the enzyme was used to catalyse the asymmetric synthesis of a multitude of aromatic and heteroaromatic 2-hydroxy ketones [23–25,27,30].

In order to enlarge the range of accessible 2-hydroxy ketones, we studied the carboligase properties of a recently described branched-chain 2-keto acid decarboxylase (KdcA). Two highly homologous enzymes have been found in different L. lactis strains. These enzymes are involved (in form of the Lactococcus cells) in the process of cheese ripening because of their decarboxylase activity toward 2-keto acids, which are formed through transamination of the corresponding branched-chain amino acids [54]. We investigated the nonphysiological carboligase properties of KdcA from L. lactis sup. cremoris B1157. Compared with other decarboxylases, its substrate range for the enzyme-catalysed acyloin condensation is broader, encompassing several aliphatic aldehydes such as acetaldehyde, propanal, butanal, isobutyraldehyde and cyclopropanecarbaldehyde as donor and/or acceptor, respectively. Moreover, LlKdcA catalyses the carboligation of enolizable CH-acidic aldehydes such as indole-3-acetaldehyde and phenylacetaldehyde (see below) [31].

The ThDP-dependent enzyme MenD from E. coli K12, formerly known as 2-succinyl-6-hydroxy-2,4-cyclohexadiene-1-carboxylate synthase, catalyses the decarboxylation of 2-ketoglutarate and the subsequent addition of the resulting succinyl-ThDP to isochorismate (Scheme 5A). This is the second step in the biosynthesis of menaquinones (vitamin K2 derivatives), which play an important role as electron shuttles in anaerobic respiration in bacteria. We tested the enzyme for nonphysiologial C–C bond-forming reactions. Condensation of 2-ketoglutarate after decarboxylation to a broad range of aldehydes gave 2-hydroxy ketones with isolated yields from 26% to 87% and 94% to 97.5%ee for addition to aromatic aldehydes (Scheme 5B,C). MenD accepts a wide range of aldehydes as acceptor substrates to produce chiral 2-hydroxy ketones with conserved regioselectivity, where the activated succinylsemialdehyde serves selectively as the donor [34]. Regioselectivity is inverted only for the condensation of 2-ketoglutarate with pyruvate (resulting in activated acetaldehyde) as a donor. In addition to 2-ketoglutarate, pyruvate and oxalacetate are accepted as donors in combination with benzaldehyde and 2-fluorobenzaldehyde as acceptors, however, with decreased carboligase activity (see below). The physiological 1,4-addition of 2-ketoglutarate to isochorismate was enlarged to 2,3-dihydroxy-2,3-dihydrobenzoate (2,3-CHD) [55] as a substrate, which lacks the pyruvyl found in isochorismate (Scheme 5D). Hence, a wide variety of new chiral building blocks are available through effective asymmetric enzymatic synthesis with MenD [34].

Figure Scheme 5..

 MenD-catalysed physiological (A) and nonphysiological (B–D) transformations.

Thus, biosynthesis and catabolic/metabolic pathways have been shown to be a prosperous source of new and useful ThDP-dependent enzymes. Application of the identified enzymes in nonphysiological asymmetric reactions broadens the scope of this enzyme class further. Others, like Balskus & Walsh [56], also adopted this strategy for the identification of ThDP-dependent enzymes, underlining the potential of this approach.

Substrate and reaction engineering

The intrinsic catalytic polyreactvity (promiscuity) of ThDP-dependent enzymes calls for diversity-oriented approaches to identify new enzymatic transformations. This necessitates variations with respect to putative substrate combinations, the supposed test reactions and the adaptation of reaction conditions (pH value, substrate concentration, choice of substrates).

As mentioned above, BAL was first identified by its lyase activity. In addition to recognition of its ability to catalyse the reverse reaction, the application of dimethylsulfoxide as a cosolvent was crucial in order to increase the solubility of benzoin [57]. This simple yet essential modification enabled Ayhan Demir, working at that time as a visiting scientist in our group, to introduce several useful asymmetric transformations [25]. Nowadays, BAL is recognized as one of the most useful and versatile ThDP-dependent enzymes in asymmetric synthesis.

In principle, the products of mixed aromatic–aliphatic benzoin-type condensation can be two symmetrical and two unsymmetrical 2-hydroxy ketones, each as a pair of enantiomers, thus eight possible products are possible in total. This is even more complex in the case of α,β-unsaturated substrates. Here, in addition to the above-mentioned eight 2-hydroxy ketones, two enantiomeric γ-hydroxy enones (Scheme 6B) and up to four stereoisomeric Stetter products (1,4-diketones; Scheme 6D) might be formed. Through a combination of substrate and enzyme screening, we identified suitable conditions for the selective synthesis of products according to Scheme 6A,C [58]. Thus, we extended the donor–acceptor concept developed for cross-benzoin condensation [24] and broadened the scope of reactivity by using aromatic and aliphatic α,β-unsaturated aldehydes as substrates in ThDP-dependent enzymatic transformations.

Figure Scheme 6..

 ThDP-dependent catalytic reaction possibilities of α,β-unsaturated aldehydes.

Substrate variation is typically used for the formation of different products. Sometimes, however, substrate variation will enable different routes to the same product. The physiological donor 2-ketoglutarate of MenD was replaced by oxalacetate and pyruvate. Both, when incubated with the acceptor 2-fluorobenzaldehyde, react to 2-fluoro-PAC (Scheme 7). Whereas the conversion with pyruvate was in the range of 5%, a significantly higher yield was obtained with oxalacetate as a donor. Because 4-(2-fluorophenyl)-4-hydroxy-3-oxobutanoic acid formation was not detected by GC-MS, 3-oxopropanoate must be decarboxylated before condensation or afterwards. Higher conversion rates with oxalacetate suggest that addition of 3-oxo-propanoate occurs prior to decarboxylation. The R-configuration was determined for both products by chiral phase HPLC. Thus, oxalacetate might be used as a mimic for pyruvate or acetaldehyde, depending on the enzyme’s preference.

Figure Scheme 7..

 Oxalacetate and pyruvate as donors resulting both in 2-fluoro-(R)-PAC.

An interesting induction of regioselectivity by applying different substrate combinations was observed when different 2-keto acids were tested for MenD-catalysed carboligation (Scheme 8). If pyruvate and 2-ketoglutarate are used in one reaction, the normal donor role of 2-ketoglutarate is changed and it functions as the acceptor substrate for decarboxylated pyruvate (acetaldehyde), resulting in 4-hydroxy-5-oxohexanoate. However, with acetaldehyde as the substrate, 5-hydroxy-4-oxohexanoate is isolated as the product. Determination of stereoselectivity by chiral phase GC, optical rotation and CD shows both condensation products (pyruvate or acetaldehyde as substrates) to be racemic.

Figure Scheme 8..

 Regioisomeric products dependent on the choice of substrates.

We have successfully tested numerous aldehydes as substrates for enzymatic ThDP-dependent asymmetric C–C bond formation. However, highly reactive aldehydes like CH-acidic aldehydes or unstable derivatives like indole-3-acetaldehyde have to be masked. Here, the ability of many ThDP-dependent enzymes to catalyse both decarboxylation of the respective 2-keto acid and activation of the corresponding aldehyde enables the application of the more appropriate derivative. For example, use of indole-3-pyruvate allows the in situ generation and enzyme-catalysed carboligation of the otherwise unstable indole-3-acetaldehyde [31].

Finally, it should be emphasized that the above-mentioned asymmetric transformations can be applied under appropriate conditions on a technical scale [59], either in a batch or continuous mode [60,61]. This, of course, is also true for other ThDP-dependent enzymes like TK [62].


The intrinsic catalytic polyreactvity (promiscuity) of ThDP-dependent enzymes has been shown to result in a broad range of versatile synthetic methods [63]. We have developed strategies to identify and introduce new enzyme-catalysed transformations, and, at the same time, cope with the intrinsic promiscuity of ThDP-dependent enzymes. However, we assume that this is just the tip of the iceberg. Awareness of many known organo-catalysed Umpolung reactions suggests screening enzymes and variants thereof, with high-throughput methods for ‘non-natural’ transformations. The metagenome approach in combination with in vivo growth selection will give access to new enzymes, irrespective of the identified characteristic ‘ThDP enzyme sequence motifs’. Yet, the richest source of ‘new’ transformations remains hidden in the, mostly unexplored, biosynthetic machinery evolved by nature [64]. This, in combination with catabolic and metabolic transformations, will provide us with innovative new strategies for asymmetric synthesis. Here, solutions for C–C bond formations with ketones and imines, as well as alternative electrophiles, for example epoxides, are in the focus of ongoing work [65].

Moreover, a thorough descriptive investigation into how to use this versatile cofactor will hopefully give us some hints as to why nature has evolved ThDP as the cofactor for Umpolung reactions. This might even help us to distinguish between successful and less probable reaction trajectories, and even to predict putative new transformations.


We thank all our talented and enthusiastic coworkers, whose names are listed in the references, for their indispensable contributions. The projects were supported by the Deutsche Forschungsgemeinschaft in frame of SFB-380, DFG Mu 1322/6-1 and DFG Po 558/4-1.