Crystal structures of the Chromobacterium violaceumω-transaminase reveal major structural rearrangements upon binding of coenzyme PLP


D. T. Logan, Department of Biochemistry and Structural Biology, Lund University, SE-22 100 Lund, Sweden
Fax: +46 46 222 4116
Tel: +46 46 222 1443;
P. Berglund, KTH Royal Institute of Technology, Division of Biochemistry, School of Biotechnology, AlbaNova University Centre, SE-106 91 Stockholm, Sweden
Fax: +46 8 5537 8468
Tel: +46 8 5537 8366


The bacterial ω-transaminase from Chromobacterium violaceum (Cv-ωTA, EC2.6.1.18) catalyses industrially important transamination reactions by use of the coenzyme pyridoxal 5′-phosphate (PLP). Here, we present four crystal structures of Cv-ωTA: two in the apo form, one in the holo form and one in an intermediate state, at resolutions between 1.35 and 2.4 Å. The enzyme is a homodimer with a molecular mass of ∼ 100 kDa. Each monomer has an active site at the dimeric interface that involves amino acid residues from both subunits. The apo-Cv-ωTA structure reveals unique ‘relaxed’ conformations of three critical loops involved in structuring the active site that have not previously been seen in a transaminase. Analysis of the four crystal structures reveals major structural rearrangements involving elements of the large and small domains of both monomers that reorganize the active site in the presence of PLP. The conformational change appears to be triggered by binding of the phosphate group of PLP. Furthermore, one of the apo structures shows a disordered ‘roof ’ over the PLP-binding site, whereas in the other apo form and the holo form the ‘roof’ is ordered. Comparison with other known transaminase crystal structures suggests that ordering of the ‘roof’ structure may be associated with substrate binding in Cv-ωTA and some other transaminases.


The atomic coordinates and structure factors for the Chromobacterium violaceumω-transaminase crystal structures can be found in the RCSB Protein Data Bank ( under the accession codes 4A6U for the holoenzyme, 4A6R for the apo1 form, 4A6T for the apo2 form and 4A72 for the mixed form

Structured digital abstract


pyridoxal 5′-phosphate


Transaminases (EC2.6.1.X) are enzymes that normally act in amino acid metabolism by transferring amino groups. The transaminases, including the ω-transaminases, transfer amino groups to keto compounds by a ping-pong bi-bi reaction mechanism involving two half-reactions, as determined for aspartate transaminase [1]. According to this mechanism, pyridoxal-5′-phosphate (PLP) is coordinated in the active site as an internal aldimine formed by a covalent Schiff base linkage to the ε-amino group of a catalytic lysine. In the first half-reaction, an amino donor reacts with the internal aldimine to form an external aldimine consisting of amino donor and PLP. The catalytic lysine is then free to perform proton abstraction. This proton must be positioned perpendicular to the π-system of the external aldimine for abstraction to be favoured [2,3]. A planar quinonoid structure is then formed which will be activated for hydrolysis by abstracting a proton from the catalytic lysine. Nucleophilic attack by water will lead to the release of a keto product and the formation of pyridoxamine-5′-phosphate. In the second half-reaction, essentially the reverse of the first one, an amino acceptor (a carbonyl-containing compound) will react with pyridoxamine-5′-phosphate to release water and form a second external aldimine. After proton abstraction by the catalytic lysine, a second planar quinonoid is formed. The catalytic cycle is completed when the catalytic lysine releases the second product, a chiral amine or amino acid, by forming a new internal aldimine.

For biocatalytic purposes, the ω-transaminases (EC2.6.1.18) are a particularly interesting group. These enzymes can transfer virtually any primary amino group (i.e. not only amino acids) to a wide range of ketones. The products are chiral amines, which in many instances are of significant importance to the pharmaceutical industry [4]. In recent years, the ω-transaminases have received much attention due to their ability to accept inexpensive amines, such as isopropylamine, as amino donors, in combination with their ability to produce chiral amino compounds of high enantiomeric excess [5–13]. Enantiopure amino compounds are important both as pharmaceuticals and as key chiral intermediates for the agrochemical, chemical and pharmaceutical industries [12–14]. The chemical preparation of chiral amines is a challenging task [14], and an enzymatic transamination route using transaminases is an attractive alternative to traditional chemical methods in both environmental and economical terms [7,15]. Industrial reaction conditions differ from those in nature, and therefore both the reaction conditions and the ω-transaminases require optimization by engineering before application. One common issue with the transamination reaction is the unfavourable reaction equilibrium. Various systems to displace the reaction equilibrium have been developed [11,16,17]. The ω-transaminases often suffer from other obstacles, such as substrate and product inhibition, poor thermostability, low solvent tolerance or low activity for the desired substrates. Various methods to solve these issues by reaction engineering [8,16,18–22], protein engineering through directed evolution [4,23,24] or combined directed evolution and point mutations guided by homology modelling [15] have been reported. The company Cambrex Corporation applies transaminase technology for the commercial manufacture of various chiral amines, such as aminotetralins and benzylamines, by asymmetric synthesis on a multiton scale [4,7].

The (S)-selective (or l-amino acid) transaminases belong to the most common transaminase fold-type I [25], whereas the (R)-selective (or d-amino acid) transaminases belong to fold-type IV [25]. A number of microorganisms containing (S)-selective ω-transaminases have been reported: Alcaligenes denitrificans Y2k-2 [26], Arthrobacter citreus [23], Bacillus megaterium SC6394 [27], Bacillus thuringiensis JS64 [18], Chromobacterium violaceum [28], Klebsiella pneumoniae JS2F [29], Mesorhizobium [30] and Vibrio fluvialis JS17 [31]. Until recently, only two (R)-selective ω-transaminases were known: Arthobacter sp. KNK168 [32] and the commercial ATA117 ( According to bacterial strain collection screening, the (R)-selective ω-transaminases seem to be rare in nature [10]. Recently, 17 novel (R)-selective ω-transaminases of different origins were discovered using an in silico approach [33,34].

The ω-transaminase isolated from the pathogenic bacterium C. violaceum (abbreviated as Cv-ωTA here) was first reported in 2007 [28]. Kaulmann and coworkers identified the ω-transaminase gene from PSI-BLAST searches using the sequence of the V. fluvialis JS17 ω-transaminase [31] as a template. Since then, a number of studies showing the potential of Cv-ωTA as a biocatalyst have been published [28,35–37]. The enzyme is sensitive towards both reaction and substrate conditions [37]. Buffers containing borate [37], phosphate [28,37] or sulfate [37] ions have been shown to have a negative effect on the enzymatic activity. This may be due to a release of the coenzyme by competition with ions that can bind to the ‘phosphate group binding cup’ [38,39]. The optimum reaction conditions were found to be HEPES buffer (50 mm), pH 8.3 at 37 °C [28,35,37]. Under these reaction conditions, the enzyme shows a specific activity of 7.2 × 10−3 μmol·min−1·mg−1 using 1.2 mm acetophenone and 50 mm isopropylamine [37]. Higher substrate concentrations result in substrate inhibition, which is a common trait among ω-transaminases. The enzyme has a broad substrate spectrum, but prefers aromatic (S)-amino donors such as (S)-α-benzylethylamine or (S)-1-aminoindane, and keto compounds such as glyoxylate or pyruvate [28]. The wide substrate acceptance of the enzyme has led to successful biocatalytic applications for the synthesis of chiral amino alcohols on the preparative scale [28,36,40,41].

Because of low homology among ω-transaminases of different origins, homology modelling is difficult and experimentally determined crystal structures are preferred. To date, protein engineering work on the ω-transaminases has mostly focused on directed evolution, partly because of the lack of crystal structures as guides for point mutation. However, some successful protein engineering work using point mutations guided by homology modelling has been reported [15,42,43]. For successful engineering, knowledge of the target amino acid residues and their locations upon binding of a given ligand and structural movements need to be considered. Even though homology models have been successfully applied, understanding detailed hydrogen-bonding patterns and structural movement schemes of a specific enzyme will, to a greater extent, enable successful construction of new enzyme variants suitable for synthetic use. Recently, the crystal structure of the ω-transaminase from V. fluvialis JS17 [31], in the apo form with the ‘roof loop’ (see later) of one of the two active sites unfolded, was solved by Jang and coworkers (PDB ID: 3NUI). A crystal structure of one monomer of the ω-transaminase from Pseudomonas putida (PDB ID: 3A8U) in the holo form has been determined. However, in neither case have structural analyses been published. Aiming to improve fundamental structural knowledge on ω-transaminases, especially for Cv-ωTA, we present the crystal structures of Cv-ωTA in both the apo and holo forms. These crystal structures reveal unique structural rearrangements upon PLP binding.


Buffer screening

Prior to crystallization, a buffer screen was performed to detect buffer conditions promoting protein stabilization over longer times. The screen was performed using differential scanning fluorimetry by measuring the melting temperatures of purified protein solutions (0.1 mg·mL−1) dissolved in various buffers with pH values ranging from 5 to 10, according to Niesen et al. [44] (Table S1). A reference protein sample was dissolved in HEPES buffer (100 mm, pH 7.4) with 100 mm NaCl, and this displayed a Tm of 78 °C. Cv-ωTA showed similarly high melting temperatures in almost all buffers, indicating a thermostable protein. The largest destabilizing effect (ΔTm = −5 °C) was obtained in sodium acetate buffer at pH 5.0, indicating that low pH is not optimal. The highest stabilization (ΔTm = 2 °C) was measured in 100 mm HEPES buffer pH 7.4, with 5% v/v glycerol and 300 mm NaCl. However, we considered the latter stabilization to be marginal, and for protein crystallization the protein was prepared in Tris buffer (20 mm Tris, 100 mm NaCl, 0.1 mm PLP, pH 7.5) in order to follow a previously published crystallization protocol for Cv-ωTA [45].

Dynamic light scattering

Prior to crystallization experiments, samples of Cv-ωTA stored in different manners were explored by visual inspection and dynamic light scattering. The purified protein, 12 mg·mL−1 in Tris buffer (see above), was stored under three different conditions: (a) at 21 °C for 3 days and then on ice for 2 days, (b) on ice for 5 days, (c) immediately frozen in liquid nitrogen and stored for 5 days at −80 °C. All fractions were thawed on ice 24 h before visual inspection and light scattering measurement. All samples showed a yellow colour. The sample frozen in liquid nitrogen showed the most intense yellow colour, whereas the fraction stored at 21 °C showed least yellow colour. The protein sample stored at 21 °C for 3 days and then on ice for 2 days showed a more narrow size distribution with a major peak of radius of 8.5 ± 0.7 nm, compared with the other two protein samples, which had a more polydisperse, larger size distribution with a major peak of radius 9.3 ± 1.8 nm (Fig. S1 and Table S2). Both protein fractions showed a high amount of protein in dimeric form with at molecular mass close to 100 kDa. Despite the better homogeneity of the sample stored at 21°C, this sample produced only the apo crystal structures of Cv-ωTA, presumably because of greater dissociation of PLP during storage at ambient temperature.

Relationship to other ω-transaminases

Alignments of the protein sequence of Cv-ωTA to other well known (S)-selective ω-transaminases such as Al. denitrificans Y2k-2 [26], A. citreus [23,24], B. megaterium SC6394 [27], P. putida and V. fluvialis JS17 [19] revealed relatively low (30–40%) sequence identities (Table 1). The ω-transaminases from P. putida (PDB ID: 3A8U) and V. fluvialis JS17 (PDB ID: 3NUI) are the only enzymes of this kind with crystal structures deposited in the Protein Data Bank. By PSI-BLAST searches (, a transaminase from Silicibacter pomeroyi (PDB ID: 3HMU) showing 51% sequence identity to Cv-ωTA was found. This structure was used as a template to solve the crystal structures by molecular replacement.

Table 1.   Sequence identitya (%) of Cv-ωTA to other well known (S)-selective ω-transaminases.
  1. a Sequence identities calculated using PSI-BLAST [62] (

Alcaligenes denitrificans Y2k-233
Arthrobacter citreus28
Bacillus megaterium SC639430
Pseudomonas putida33
Vibrio fluvialis JS1738


Initial attempts to cocrystallize Cv-ωTA with PLP involved incubating the protein at room temperature for 72 h. This resulted in transparent crystals without PLP belonging to space group P1 and containing one dimer in the asymmetric unit, corresponding to crystal form A described by Sayer et al. [45]. A further apo structure was obtained, also in space group P1 with very similar cell dimensions, using a different precipitant, namely poly(ethylene glycol) instead of poly(acrylic acid). The active PLP-containing form, containing two dimers in the asymmetric unit, as well as a ‘mixed’ form containing both apo and holo conformations, were both obtained by storing the purified protein on ice in the dark for 1 week before freezing for long-term storage. Upon thawing, the protein was reincubated with 100 μm PLP before data collection. The Matthews volume is 2.06 Å3·Da−1 and the solvent content is 40.4%. The holo form thus obtained corresponds to crystal form B of Sayer et al. [45]. This crystal form is related to form A by an approximate doubling of the c-axis length, accompanied by the appearance of a peak at (0, 0, ½) in the native Patterson function at 7.5–10% of the height of the origin peak, indicating that the two dimers in the form B unit cell are related by weak translational pseudosymmetry. All the crystal forms have a Matthews volume of ∼ 2.05 Å3·Da−1 and a solvent content of ∼ 40%, except for the apo form obtained from poly(ethylene glycol), which is more tightly packed, with a Matthews volume of 1.88 Å3·Da−1 and a solvent content of 34.6%.

Overall structure

The crystal structures of Cv-ωTA have been solved in two apo forms, one PLP-bound form and a fourth form in which around half of the molecules are apo and around half holo (Table 2). The holo form was obtained from crystals cocrystallized with 1 mm pyruvate and 1 mm isopropylamine, but no trace of these substrates can be found in the electron density. Likewise the ‘mixed’ form was obtained from cocrystals with 1 mm cycloserine, but the substrate is not observed. The most complete structure is the holo form, in which ordered structure is observed between residues 5 and 458 (the C-terminus) in all chains. The apo form crystallized in the presence of poly(acrylic acid) can be modelled between residues 34 and 459. The apo form crystallized from poly(ethylene glycol) and NaSCN is very similar, but two segments on opposite sides of the active site (residues 152–178 from one monomer and 311–320 of the other) are disordered with respect to the first apo form.

Table 2.   Data collection and structure quality statistics. Figures in parentheses refer to the highest resolution shell.
Data collection
 Space groupP1P1P1P1
 Cell dimensionsa = 58.6 Å, b = 61.9 Å, c = 63.5 Å
α = 108.0°, β = 87.3°, γ = 105.2°
a = 57.2 Å, b = 60.2 Å, c = 61.1 Å
α = 109.1°, β = 85.6°, γ = 103.2°
a = 61.3 Å, b = 62.1 Å, c = 119.2 Å
α = 105.3°, β = 90.6, γ = 104.5°
a = 61.5 Å, b = 62.3 Å, c = 119.5 Å
α = 105.4°, β = 90.9°, γ = 104.5°
 Dimers in asymmetric unit1122
 Matthews volume (A3·Da−1)2.051.882.052.06
 Solvent content (%)39.934.640.140.4
 Resolution (Å)30–1.35 (1.38–1.35)30–1.7 (1.8–1.7)30–2.4 (2.45–2.4)30–1.8 (1.85–1.8)
 Rmerge (%)6.7 (79.9)6.6 (37.0)9.9 (49.8)6.4 (41.3)
 I/σ (I)12.6 (1.9)12.0 (3.9)8.5 (2.1)8.7 (1.6)
 Completeness (%)94.1 (89.4)93.9 (79.5)95.5 (82.6)93.9 (87.5)
 No. of observations690977211597145272286254
 No. of unique observations1697417893061839143722
 Redundancy4.0 (3.8)2.7 (2.6)2.3 (2.2)2.0 (1.9)
 Wilson B-factor (Å2)18.923.437.228.8
 Resolution (Å)30–1.35 (1.38–1.35)30–1.7 (1.75–1.7)30–2.4 (2.45–2.4)30–1.8 (1.82–1.8)
 Rmodel (%)13.3 (25.5)15.7 (27.2)17.3 (23.5)16.5 (32.8)
 Rfree (%)16.3 (31.1)18.6 (34.0)23.3 (33.3)20.2 (34.8)
 No. protein atoms755662251438114339
 Water molecules8654895441111
rmsd from ideal geometry
 Bond lengths (Å)0.0080.0120.0020.005
 Bond angles (°)
Ramachandran plot analysis (%)
 Preferred regions98.
 Additional allowed regions1.

The overall fold of Cv-ωTA is typical of the class I transaminase fold. Each monomeric subunit comprises two major domains (Fig. 1). The large domain (residues 93–313) folds into a typical three-layered α/β/α sandwich comprising a central seven-stranded β sheet with strand order 4-10-9-8-7-5-6. All the β strands except β10 are parallel. The catalytic Lys288 is located in the loop preceding the only antiparallel strand, β10. The large domain is flanked by a small domain consisting of residues from the N- and C-termini, which can be divided into two lobes joined by a short parallel β-interaction between strands β3 and β14. The N-terminal lobe is a three-stranded antiparallel β sheet capped at one end by the two most N-terminal helices α1 and α2, which are oriented perpendicular to the sheet. However, the fold of the N-terminal lobe would be unstable in the absence of extensive interactions with monomer 2 (Fig. 1B, described in more detail below) that are critical for structuring the active site. The C-terminal lobe of this domain is a four-stranded antiparallel β sheet consisting of strands 11, 12, 16 and 15 (Fig. 1B).

Figure 1.

 (A) Overall fold of one chain of the holo form of Cv-ωTA. The large domain is coloured pink and the two lobes of the small domain are coloured light and dark blue, respectively. The linker, between the N-terminal lobe of the small domain and the large domain, containing helix α4, is coloured grey. The PLP/Lys288 internal aldimine is shown in a stick representation covered by a semitransparent surface to pinpoint the active site. (B) Topology diagram for the holo form of Cv-ωTA. The central domain is highlighted by a grey background. The position of Lys288 in the sequence is indicated by a black triangle (after strand β9). The N- and C-terminal lobes of the outer domain are shown to the left and right of the central domain respectively. (C) The dimer of Cv-ωTA. The large domains are coloured dark pink and dark blue, respectively, and the small domains are coloured light pink and light blue, respectively. PLP is shown as spheres.

Because of the prior lack of ω-transaminase crystal structures, we had previously created homology structures of Cv-ωTA (M.S. Humble, K.E. Cassimjee, V. Abedi, H.-J. Federsel & P. Berglund, unpublished) and A. citreusω-transaminase [46] based on the class III aminotransferase from S. pomeroyi (PDB ID: 3HMU). A comparison between the homology model of Cv-ωTA and the template structure showed that 896 Cα positions of a total of 916 could be aligned with an rmsd of 1.01 Å. In the active site, the backbone structure is almost identical, with only a few variations in side-chain rotamers. The most significant variations are found in the outer loop regions. Interestingly, even though the template structure was in the apo form, the rearranging loops around the active site were modelled in the same positions as in the PLP-bound holo form presented here. The Cv-ωTA homology structure was successfully applied for molecular docking simulations and used to predict targets for site-directed mutagenesis in order to improve the enzyme properties (unpublished data). Nevertheless, the crystal structure is more informative, since the hydrogen-bonding network in the homology model is incomplete due to differences in side-chain rotamers and lack of crystal water molecules.

Active site

The active sites of the Cv-ωTA dimer are located at the dimeric subunit interface, and contain amino acid residues from both subunits. Each active site contains a PLP-binding pocket and a substrate-binding region. In the holo form, PLP is joined by a covalent Schiff base linkage to the ε-amino group of the catalytic lysine Lys288 (Fig. 2). The proton on the pyridine ring nitrogen is fixed by hydrogen bond coordination to Asp259, as is common in transaminases. The phosphate group of PLP is coordinated by eight hydrogen bonds from the three nonester phosphate oxygen atoms, directly or through water molecules, to the side chains of Ser121 and Tyr153 from the same monomer, as well as to the main-chain amide nitrogen atoms of Thr321′ and Tyr322′, the side-chain hydroxyl group of Thr321′ and the side-chain amide of Asn118′ from the other monomer. This network of hydrogen-bonding interactions has been described as ‘the phosphate group binding cup’ [47]. Finally, the hydroxyl group on PLP makes water-bridged hydrogen bonds to Asp226. The aromatic ring of PLP is further sandwiched between the side chains of Tyr153 (in an edge-on interaction) and Val261. The interplay of structural elements from both monomers is crucial for the correct structure of the PLP-binding site, as will become evident from analysis of the apo form.

Figure 2.

 Close-up representation of the active site in the holo form. The PLP internal aldimine complex with Lys288 is drawn in stick representation inside a grey semitransparent surface that shows an ‘omit’-style difference map calculated with phases from a model from which Lys288 and PLP were removed. This map is contoured at a level of 2.5 σ. Residues within 4 Å of PLP are shown in stick representation. Hydrogen bonds, direct or through water molecules, are shown as dotted lines. Red spheres indicate water molecules near PLP that participate in hydrogen bonds to the aldimine. Carbon atoms in residues from monomer 1 are coloured grey while those from monomer 2 are coloured yellow.

Conformational transitions between inactive and active forms

The transition from the inactive to the active conformation of Cv-ωTA involves extensive refolding of secondary structure elements from both monomers of the dimer, which are highly intertwined. However, the rearrangements are mostly limited to one side of the active site, whereas the other side is more or less rigid. In the following discussion we refer to the flexible side of the active site cavity as the ‘base’ and the other side as the ‘roof’. The monomer to which PLP is covalently bound will be referred to as monomer 1 and the other as monomer 2. The apo form crystallized from poly(acrylic acid) will be referred to as apo1 and the form crystallized from poly(ethylene glycol) 3350 as apo2.

The residues in monomer 1 in the active site base that refold (Fig. 3) are 5–37, constituting helices α1 and α2, as well as the following loop that joins these to a small three-stranded β sheet (β1–β3, residues 37–53). Residues 5–34 are completely disordered in the apo form. Residues 5–37 are intertwined with loop 83′–94′ of monomer 2, which refolds to pack more closely against the critical loop 316′–328′. In the apo1 structure, loop 316′–328′ contains a small helix between residues 317′ and 322′ (Fig. 3A). This helix packs against the edges of strands 4 and 10 of the central domain’s β sheet. A small hydrophobic core is formed by the packing of Phe316′, Phe320′ and Ser323′ from this helix against Phe115, Thr126 and Met130. Asp315 makes a salt bridge to Asp133 that locks the outer edge of the interaction area. In this conformation, Thr321, an important and conserved element of the phosphate-binding site, is oriented towards solvent with its closest atom (Cα) ∼ 17 Å from the phosphate-binding site in the holo structure. Tyr322, which makes important stabilizing interactions to the ‘PLP loop’ containing Lys288, is here involved in a hydrogen bond to the main-chain carbonyl group of Phe89.

Figure 3.

 Illustration of the conformational changes on active complex formation. (A) Conformation of the active site in the apo form. Monomer 1 is shown in light blue, monomer 2 in beige. The loops from monomer 2 that refold upon PLP binding are shown in dark red. Tyr322 and Thr321, which are critical in the active conformation, are shown in stick representation, as are residues forming the hydrophobic core that stabilizes the helix packing against strands 4 and 10 of the β sheet. (B) Active site conformation in the PLP complex. The colour scheme and representation are as in (A) except that the residues of the hydrophobic core are omitted, and the internal aldimine between PLP and Lys288 is shown as sticks. (C) Close-up of the ‘PLP loop’ containing Lys288 to illustrate the subtle conformational differences between apo and holo forms. The apo form is drawn in grey, the active form in light blue. The loop from monomer 2 that includes Tyr322 is shown in dark red.

In the active PLP-bound conformation the space occupied by loop 316′–328′ in the apo1 structure is invaded by the ‘capping helices’ from the N-terminal lobe and their intervening loop. Loop 316–328 loses all secondary structure and extends into the active site of the other monomer (Fig. 3B). The side-chain hydroxyl atoms of Thr321 and Tyr322 move by 17 and 19 Å, respectively. The main-chain atoms of residue 318–321 participate in several hydrogen bonds to the phosphate group of covalently bound PLP, as does the side chain of Thr321. The other elements of the PLP-binding site, particularly on the roof, are largely invariant between the active and inactive forms. Val261 moves slightly from its position in the apo form to accommodate PLP in the sandwiching interaction described above.

A critical interaction for the correct orientation of Lys288 in the active conformation comes from the projection of the side chain of Tyr322 of monomer 2 towards the ‘PLP loop’ containing Lys288 in monomer 1. This is completely dependent on the refolding of loop 316–328. In the apo form, residues 286–298 forming the ‘PLP loop’ contain a very short distorted helical segment between 288 and 292, in which the carbonyl group of Lys288 makes hydrogen bonds to both Ser292 (3.2 Å) and Gly293 (both 2.9 Å; Fig. 3C). In this conformation, the side chain of Lys288 is projected away from the active site. The insertion of the side chain of Tyr322 into this space has a subtle effect on the PLP loop, pushing away residues 287–289 and altering the main-chain torsion angles such that the side chain of the Lys288/PLP internal aldimine is now directed towards the active site. Interestingly, in the higher resolution apo1 form (1.35 Å) the loop can be modelled as a 50/50 mixture of the inactive and active conformations, suggesting that the loop exists in a dynamic equilibrium between these forms that can be readily altered by pyridoxamine-5′-phosphate binding.

In the apo2 form, the critical loop 316–328 is disordered between residues 311 and 320, despite the remainder of the structural elements being in the same conformation as in the other apo form, although electron density can be seen at lower contour levels (0.5 σ) for the helical portion of the loop in one of the monomers. The reasons for this disorder are not evident, but it might be an indirect effect of crystal packing, which is more asymmetric in the apo2 form. The N-terminus of loop 316–328 is positioned ∼ 10  Å from the C-terminus of neighbouring polypeptides in the crystal.

In form apo1, electron density is observed in the space occupied by the phosphate group and Phe320 in the active form, for what appears to be a fragment of a larger polymeric molecule. We have modelled this as a tetramer of poly(acrylic acid) from the crystallization solution (Fig. 4). We attempted to model other buffer components in the density, e.g. HEPES and ethylene glycol, but poly(acrylic acid) gave the best fit. However, the inactive conformation does not appear to be an artefact caused by the presence of this molecule, because essentially the same structure is obtained when Cv-ωTA is crystallized from poly(ethylene glycol) 3350 (apo2). In fact, the observed molecule may be partly responsible for stabilization of the active site roof, because a segment of the protein that contacts it (residues 152–178) is disordered when the protein is crystallized from poly(ethylene glycol) 3350. This area of the structure forms a ‘lid’ over the top of the aromatic moiety of PLP and is disordered in several other transaminase structures (see Discussion).

Figure 4.

 Ordering of the active site roof loop by a putative tetramer of poly(acrylic acid) (PAA). An ‘omit’ difference electron density map, unbiased by the poly(acrylic acid) model, is shown contoured at 2.5 σ. Side chains of interacting residues are shown as sticks. Bound water molecules in the vicinity of poly(acrylic acid) are shown as spheres. Monomer 1 is coloured grey and monomer 2 is yellow.

The fourth structure presented here is of a mixed form in which the asymmetric unit, like the holo form, contains two dimers. This was obtained by soaking the crystals in 1 mm cycloserine, although it is not observed in the electron density. One of these (A/B) is in the apo form and the other (C/D) in the holo form. This is further proof of the malleability of the protein structure in this enzyme. Intriguingly, the polypeptide chain in monomer A itself seems to exist in a mixture of apo and holo forms. Weak density is seen for helix α1 (Fig. 5A), which is only present in the holo form. Residues 86′–92′ and 315′–323′, with which the following loop would clash in the holo form, are poorly ordered compared with the pure apo structure, indicating that the electron density reports on an ensemble average of A/B dimers in the crystal, some of which have undergone the transition to the holo form. There is also density, visible at ∼ 0.5 σ, in the active site of monomer A (Fig. 5B), suggesting partial occupancy of PLP/Lys288.

Figure 5.

 Evidence for a mixture of holo and apo conformations in dimer A/B of the ‘mixed’ structure. (A) The N-terminal region of chain A. The ‘mixed’ structure is shown in a tube representation. The thickness of the tube is proportional to the B-factor of the residue and it is also coloured by B-factor, from blue for B-factors < 10 Å2 via yellow to red for B-factors over 130 Å2. The holo form is drawn with grey lines. A 2|Fo| − |Fc| map calculated from the apo structure (i.e. unbiased by helix α1) is shown, contoured at 0.6 σ. Loops 84–93 and 313–324 in the apo structure which have to refold to accommodate the N-terminal helices in the holo form (grey lines) have very high B-factors compared to the purely apo structure, indicating that they are not fully occupied in this conformation. (B) The active site of chain A is partially occupied by PLP (sticks). A 2|Fo| − |Fc| map calculated from the apo structure (unbiased by PLP) is shown, contoured at 0.6 σ.


We have determined the structures of Cv-ωTA in two different apo forms, in a holo form with PLP covalently bound as an internal aldimine to Lys288 and in an intermediate form in which more than half of the chains in the asymmetric unit have transitioned from apo to holo. To our knowledge, this is the first time a ‘relaxed’ conformation has been described for the apo form of a transaminase or that such extensive rearrangements between apo and holo forms have been seen. The apo structure of Cv-ωTA is highly unlikely to be an artefact, because of the well-ordered structure evidenced by the helix between residues 317 and 322 and its packing to strands 4 and 10 of the central β sheet with a well-defined hydrophobic core. The reasons why this region is less well ordered in the apo2 form, produced from a different precipitant, are not clear, but the structure is consistent with high reorganization propensity in this region that may be coupled to substrate binding.

Structural movements between different forms have already been observed for aspartate transaminase, the first PLP-binding enzyme whose structure was extensively studied [48]. However these differences were limited to rigid body movements changing the relative orientations of the large and small domains. Moreover, they occurred upon binding of substrate and not PLP. The N-terminal lobe of the small domain in aspartate transaminase and many related class I enzymes is limited to an arm that stretches over to the other monomer. Cv-ωTA has a more complex N-terminal lobe consisting of a three-stranded β sheet with two capping helices. The protrusion of these helices and the loop between them into the space occupied by the 317′–322′ loop in the apo form is a central part of the structural reorganization.

The refolding of loop 317′–322′ contributes in two critical ways to the correct orientation of covalently bound PLP: first, it contributes important hydrogen-bonding interactions orienting the phosphate group of the cofactor; and second, it ensures that Lys288 is projected into the active site by stabilizing the active conformation of residues 287–290, which appears to exist in a dynamic equilibrium between two conformations in the apo structure. One of these conformations is the active one, whereas in the other Lys288 is projected away from the active site. Covalent reaction of PLP with Lys288, or simply the binding of PLP may be the driving force for conformational rearrangement. Because even in the apo form, a certain proportion of Lys288 conformations is oriented towards the active site, formation of the internal aldimine presumably shifts the equilibrium towards the active form and drives more and more of the enzyme into this conformation. This is supported by the observation that no other elements of the protein occlude the active site in the inactive form, leaving free space for PLP to react with Lys288. Analysis of the most structurally homologous transaminases identified by the PDBeFold [49] and DALI [50] servers supports this hypothesis. All these homologues have active PLP-loop conformations. The most similar, the class III aminotransferase from S. pomeroyi (PDB ID: 3HMU; Toro et al., unpublished) is in an apo form with ‘active’ PLP loop. This might appear to contradict the idea that PLP binding drives the conformational rearrangement; however, this structure contains sulfate from the crystallization buffer in the PLP phosphate-binding pocket and is thus, on the contrary, strong evidence that binding of the PLP phosphate group could be the principal driver of the rearrangement. Two further putative aminotransferase structures have a mixture of unreacted PLP and internal aldimine in their active sites: MLL7127 from M. loti (PDB ID: 3GJU) and YP_614685.1 from Silicibacter sp. (PDB ID: 3FCR). Two further aminotransferase structures, PRK07036 from Rhodobacter sphaeroides (PDB ID: 3I5T) and acetylornithine aminotransferase from Thermotoga maritima (PDB ID: 2E54) contain entirely unreacted PLP in the active site. Thus, although none of these structures has been determined in an apo form exactly resembling the one we observed, all are consistent with the hypothesis that phosphate binding causes a structural rearrangement that organizes the PLP loop in Cv-ωTA, and that the conclusions may be extendable to other subgroup II transaminases.

Activation is clearly a slow process: protein stored at room temperature produced two different apo forms, apparently because breakdown of PLP at room temperature was faster than activation and drove the equilibrium towards the apo form. Neither the addition of 0.1 mm PLP to protein stored in this way 45 min before crystallization nor inclusion of the same in the cryo solutions was sufficient to induce the holo form. However, storage of the protein for 1 week at 4 °C before crystallization at room temperature resulted in the holo form and the ‘mixed’ form, suggesting that enough PLP stayed intact at this temperature to feed the slow activation process. Slow activation is consistent with biochemical experiments, in which Cv-ωTA was purified using IMAC chromatography and mixed with excess PLP before desalting. Enzyme treated in this way exhibited an 129% increase in activity after overnight incubation at room temperature, or at 37 °C for 4 h, without addition of further PLP [16]. Nevertheless, we are currently unable to determine why the ‘mixed’ crystal form contains a mixture of apo and holo forms, because the protein used was treated in the same way.

Two further questions are whether the roof of the active site that extends over the aromatic ring of PLP ordered or disordered in the apo form of Cv-ωTA, and what is the possible function of such an order–disorder transition? In the apo form crystallized from a solution containing poly(acrylic acid), the roof is ordered, but when crystallized from poly(ethylene glycol) it is not. It appears that a molecule we have identified as a poly(acrylic acid) tetramer may be responsible for structuring the roof, although it is insufficient to trigger the restructuring that leads to the active conformation of Lys288. One of its putative carboxylate groups lies at approximately the same place as the phosphate group of PLP in the holo structure, but presumably the greater number of hydrogen-bonding opportunities presented by the latter is required to initiate this reorganization of the PLP loop. The fewer interactions with the putative acrylic acid tetramer may, however, be enough to help structure the roof. An interesting parallel is seen in the enzyme 7,8-diaminopelargonic acid synthase from B. subtilis [51]. In the apo form (PDB ID: 3DRD), the exact same area of the active site roof is disordered as in our apo [poly(ethylene glycol)] structure. In the complex with PLP and substrate 7-keto-8-aminopelargonic acid (PDB ID: 3DU4), the roof becomes completely ordered. This was attributed to interactions with the amino group of the substrate [51], which lies very close to one of the carboxylate groups in our modelled poly(acrylic acid) tetramer. However, in this structurally homologous enzyme (rmsd to Cv-ωTA 1.74 Å for 433 Cα positions) the PLP loop is in the active conformation in the apo form, thus it is clear that it may not be possible to generalize the conformational rearrangements seen in Cv-ωTA to all structurally related transaminases.

Materials and methods

Protein expression and purification

The gene coding for C. violaceumω-transaminase DSM 30191 (GenBank accession no. NP_901695) [9] was inserted into the plasmid pET28a(+) with an N-terminal His6-tag after digestion by NheI and HindIII and transformed into Escherichia coli BL21(DE3) by electroporation [37]. The protein was expressed in E. coli BL21(DE3) for 24 h at 25 °C, 120 rpm agitation, by adding one overnight culture (20 mL) to Luria–Bertoni medium (180 mL) containing kanamycin (50 mg·L−1) and isopropyl thio-β-d-galactoside (0.4 mm), using baffled flasks. The cells were harvested by centrifugation, resuspended in IMAC binding buffer (20 mm HEPES, 500 mm NaCl, pH 8.0) and lyophilized by addition of BugBuster® 10X (Merck, Solna, Sweden). The cell debris was removed by centrifugation and filtration. The supernatant was applied to a column with Chelating Sepharose FastFlow resin (GE Healthcare, Uppsala, Sweden) pre-treated with a saturated water solution of cobalt(II) chloride (Sigma-Aldrich, Stockholm, Sweden). The resin was washed with IMAC binding buffer prior to elution by IMAC elution buffer (20 mm HEPES, 500 mm NaCl, 500 mm imidazole, pH 8.0). The protein fraction was collected and incubated for 20 min at 20 °C in an excess of PLP. The protein solution was then desalted using a PD10 column (GE Healthcare) and stored in Tris buffer (20 mm Tris, 100 mm NaCl, 0.1 mm PLP, pH 7.5). The amount of protein was measured spectrophotometrically at 280 nm and concentrated using centrifugal filter units (Millipore, Solna, Sweden) to 12 mg·mL−1. The protein was stored in three different fractions: at 21 °C for 3 days and then on ice for 2 days, on ice for 5 days or immediately frozen in liquid nitrogen and stored 5 days at −80 °C for 5 days.

Dynamic light scattering

Dynamic light scattering was measured at a 173° backscattering angle at 25 °C in a Zetasizer Nano instrument (Malvern Ltd, Malvern, UK). Measurements were made in triplicate at 20 °C using ZEN0040 cuvettes (Malvern), without previous centrifugation of the samples. Samples of 50 μL were incubated at 25 °C for 1 min before measurement.


Crystals were grown using the sitting drop vapour diffusion method. Initially, Cv-ωTA was crystallized using the protein fraction stored at 21 °C for 3 days before continued storage on ice. Crystals of apo-Cv-ωTA were produced at 18 °C in MRC plates (Molecular Dimensions Ltd, Newmarket, UK) at the MAX-lab crystallization facility. 200 + 200 nL drops were dispensed using a Mosquito robot (TTP Labtech Ltd, Royston, UK) and the JCSG+ screen (Molecular Dimensions). Crystals used for data collection were grown from 0.2 m MgCl2, 0.1 m HEPES pH 7.5 and 22% w/v poly(acrylic acid) 5100 sodium salt. The crystal was cryoprotected using 20% v/v ethylene glycol and 80% reservoir solution. A second apo-Cv-ωTA structure was determined from crystals grown from protein stored in the same way, but with a reservoir solution containing 20% w/v poly(ethylene glycol) 3350 and 0.2 m NaSCN, 0.1 m HEPES pH 7.5. This crystal was cryoprotected using 20% v/v ethylene glycol, 20% w/v poly(ethylene glycol) 3350, 0.2 m NaSCN and 0.1 mm PLP.

Initial attempts to crystallize Cv-ωTA using the protein stored at −80 °C yielded only very thin, needle-shaped crystals. Instead, a protein fraction stored in the dark on ice for a week and thereafter frozen and stored at −80 °C was used for further crystallization. To trap the holoenzyme, additional amounts of PLP (0.05 mm) and 1 mm of different additives (cycloserine, gabaculine, pyruvate and isopropylamine) were tested. Initial crystals were found at several conditions in the ProPlex screen (Molecular Dimensions). Crystals used for data collection were grown at 18 °C from drops containing 200 nL of protein/additive (0.1 mm PLP and 1 mm of respectively additive) and 200 nL of reservoir consisting of 0.1 m HEPES pH 7.5, 150–300 mm NaCl and 22.5–27.5% poly(ethylene glycol) 4000. Crystals grew to their maximum size in 2–3 days. The cryoprotectant used for holo-Cv-ωTA consisted of 20% glycerol, 80% reservoir and 0.1 mm PLP.

Data collection and structure determination

All crystals were soaked in cryoprotectant and flash-cooled in a stream of liquid nitrogen prior to data collection at beamline I911-2 of the MAX-II synchrotron in Lund, Sweden using a 165 mm marCCD detector (MarResearch, Norderstedt, Germany). The wavelength was 1.0397 Å and the temperature was 100 K. The data were processed and reduced using the xds program package [52]. The structure of apo-Cv-ωTA was determined by molecular replacement using the program phaser [53] through the pipeline mrbump [54] in the ccp4 program suite [55]. Five models were prepared using the Chainsaw method [56] based on known structures with highest sequence identity to the current structure. The most successful model was PDB entry 3HMU, the class III aminotransferase from S. pomeroyi.phaser found a solution with a rotation function Z-score of 28. This model had an initial Rmodel and Rfree of 51.4% and 51.8% that decreased to 37.4% and 40.1% after 30 cycles of refinement in refmac5 [57]. Automatic rebuilding of the model was carried out using buccaneer [58]. This resulted in the building of 854/916 residues in the asymmetric unit, of which 844 were sequenced. Rmodel and Rfree were 28.2 and 31.3%, respectively at this stage. Manual rebuilding was then carried out using coot [59] and refinement used refmac5 and phenix.refine [60]. Riding hydrogen atoms were added to all structures and TLS parameters [61] were refined in all cases. The final Rmodel and Rfree were 13.3% and 16.3%, respectively (Table 2). The other structures were solved by molecular replacement using the structure of the apo protein in as search model. These structures were all manually rebuilt using coot and refined using refmac5 in combination with phenix.refine [60]. In each case, the optimum geometry and B-factor weighting producing the lowest Rfree were found in the final refinement cycles.

Structural analysis and comparison

Homologous structures were found by searching the Protein Data Bank using the PDBeFold [49] and DALI [50] servers, as well as searching the sequences of PDB entries using protein BLAST [62]. These servers found predominantly the same homologues, with a few exceptions, but all the hits were combined into a single list. All figures were made using pymol (Schrödinger Inc., Cambridge, MA, USA;


The Chromobacterium violaceumω-transaminase gene was kindly supplied by Prof. Wolfgang Kroutil. We acknowledge access to the MAX-lab protein crystallization facility and thank staff at MAX-II beamline I911 for an excellent data collection facility and for assistance during data collection. The strong support, involvement in scientific discussions, and the experimental activities conducted by Dr Andrew Wells, our former colleague at AstraZeneca, Charnwood, UK, have been highly appreciated during the lifetime of this project. This work was supported by VINNOVA (The Swedish Governmental Agency for Innovation Systems).