Structures of a γ-aminobutyrate (GABA) transaminase from the s-triazine-degrading organism Arthrobacter aurescens TC1 in complex with PLP and with its external aldimine PLP–GABA adduct

The structure of the γ-aminobutyrate transaminase A1R958 from A. aurescens TC1 in complex with PLP–GABA reveals the active-site determinants of GABA binding in this enzyme.

Abstract

Two complex structures of the γ-aminobutyrate (GABA) transaminase A1R958 from Arthrobacter aurescens TC1 are presented. The first, determined to a resolution of 2.80 Å, features the internal aldimine formed by reaction between the ∊-amino group of Lys295 and the cofactor pyridoxal phosphate (PLP); the second, determined to a resolution of 2.75 Å, features the external aldimine adduct formed between PLP and GABA in the first half-reaction. This is the first structure of a microbial GABA transaminase in complex with its natural external aldimine and reveals the molecular determinants of GABA binding in this enzyme.

1. Introduction  

Transaminases (TAs), or aminotransferases, are pyridoxal 5′-phosphate (PLP) dependent enzymes that fulfil a central role in metabolism, catalyzing the interconversion of amino acids and keto acids through the transfer of ammonia (Mehta et al., 1993 ▶; Hwang et al., 2005 ▶). There are several types of TAs; they were originally divided into four classes (Mehta et al., 1993 ▶), but have recently been refined into five subgroups by Kim and coworkers (Hwang et al., 2005 ▶). The γ-aminobutyrate TAs (GABA-TAs) from subgroup III (Hwang et al., 2005 ▶), previously class II (Mehta et al., 1993 ▶), are a type of TA enzyme that has been the subject of particular study, as in mammals they catalyse the breakdown of γ-aminobutyric acid in the synapse (Olsen, 2002 ▶). In the resting state of the enzymes, the cofactor PLP is bound in the active site of the enzyme through a Schiff-base linkage between the aldehydic carbon of PLP and the side chain of an active-site lysine residue, forming an ‘internal’ aldimine. In the first of two half reactions catalyzed by GABA-TA (Fig. 1 ▶), γ-aminobutyric acid (1), acting as an ammonia donor, is transformed into succinic acid semialdehyde (2), with transfer of ammonia from this first substrate to the cofactor PLP to form the modified cofactor pyridoxamine 5′-­monophosphate (PMP) via a PLP–GABA adduct known as the ‘external’ aldimine. In the second half-reaction, the cosubstrate α-­ketoglutarate (3) is transformed into the amino-acid product l-­glutamic acid (4) using PMP as the ammonia donor, again via an external aldimine adduct. The GABA-TAs from pig (Storici et al., 1999 ▶, 2004 ▶) and the bacterium Escherichia coli (Liu et al., 2004 ▶, 2005 ▶) have been structurally and biochemically characterized. The bacterial GABA-TA active site, which is found at the interface between two closely associated monomers, each of which contributes interactions to PLP binding, is surprisingly well conserved in the mammalian enzymes, particularly in terms of PLP binding, and serves as a useful research model in terms of ease of expression and crystallization. Using the E. coli enzyme, Liu and coworkers determined the structures of both the internal aldimine complex, in which an active-site lysine (268) forms a covalent Schiff base with the PLP cofactor, and an external aldimine complex with the inhibitor aminooxyacetate (AOA), which although one atom shorter is a structural analogue of GABA itself (Liu et al., 2004 ▶). They determined that the carboxylate of AOA was fixed in the active site by a salt bridge with Arg141, and also proposed roles for other residues: Glu211 was thought to neutralize the side chain of Arg398 in the first half-reaction of GABA-TA activity, Arg398 was thought to bind the α-carboxylate of l-glutamate in the second half-reaction, Val241 was determined to facilitate the formation of an external aldimine with α-amino-acid substrates, and Ile50 was thought to form a hydrophobic lid at the top of the active site.

Figure 1.

Reaction catalysed by GABA-TA. In the first half reaction, the ammonia group of GABA (1) is transferred to the pyridoxal 5′-phosphate (PLP) cofactor to form pyridoxamine 5′-phosphate (PMP) and succinic semialdehyde (2); in the second half reaction, PMP transfers its ammonia group to the keto-acid acceptor α-ketoglutarate (3) to form l-glutamate (4).

The organism Arthrobacter aurescens TC1 has been studied intensively from the perspectives of bioremediation science and also microbial genomics, as the organism is able to degrade s-triazine-like pesticide compounds such as atrazine, initially through dechlorination and deamination of the triazine ring (Strong et al., 2002 ▶). The genome sequence of the organism was published in 2006 (Mongodin et al., 2006 ▶) and suggested that this strain of Arthrobacter was well adapted to the metabolic transformation of amines, with at least five genes encoding amine oxidases, several genes encoding amine flavoprotein dehydrogenases (Shapir et al., 2007 ▶) and, by our analysis, at least 26 genes encoding putative TA enzymes. A phylogenetic tree of those 26 TA targets is shown in Fig. 2 ▶, in which the branches are assigned to subgroups of TAs as assigned by Kim and coworkers (Hwang et al., 2005 ▶). Of the five subgroups of TAs, each appears to be represented in A. aurescens TC1, with the group III enzymes that encompass the GABA-TAs and the ω-TAs being particularly well presented. In the interests of both describing the amine-metabolizing enzymes of Arthrobacter, as well as discovering possible new bio­catalysts for transamination reactions, we determined the structure of two complexes of the GABA-transaminase A1R958 from this organism. In the first, the internal aldimine between the active-site lysine and the cofactor PLP is observed, while the second, in which the protein was cocrystallized with GABA, reveals the external aldimine form of the enzyme in which the active-site lysine is displaced by the primary amine of GABA. The structures confirm the extent of conservation of active-site architecture amongst bacterial GABA-TAs and also the role of Arg164 as the primary determinant of GABA carboxylate binding in the first half of the transamination reaction.

Figure 2.

Phylogenetic tree showing the evolutionary relationships between enzymes in the transaminase complement of A. aurescens TC1. Subgroups of transaminases have been allocated according to the designations described by Kim and coworkers (Hwang et al., 2005 ▶). Subgroups I and II are aromatic and aspartic acid transaminases, subgroup III includes the ω-transaminases and the GABA transaminases, subgroup IV includes the branched-chain transaminases and subgroup V includes the serine and histidinol phosphate transaminases.

2. Experimental  

2.1. Gene cloning, expression and protein purification  

Genomic DNA of A. aurescens was purchased from the ATCC. The gene encoding A1R958 (AAur_3069) was amplified by PCR from the genomic DNA using the following primers: CCAGGGACCAGCAATGACCACCACCGCGAACGAACTCTC (forward) and GAGGAGAAGGCGCGTTACTGCGCAAGCAACCGGGTTGC (reverse). Following gel analysis of the PCR product, a band of the appropriate size was eluted using a PCR Cleanup kit (Qiagen) and the gene was cloned into the pET-YSBL-LIC-3C vector using a previously described ligation-independent cloning (LIC) procedure (Atkin et al., 2008 ▶). The recombinant plasmid was then used to transform E. coli XL1-Blue cells (Novagen). Small cultures of transformants yielded plasmids using standard miniprep techniques that were submitted for sequencing to confirm the integrity of the gene. The recombinant vector containing the A1R958 gene was then used to transform E. coli BL21 (DE3) cells using 30 µg ml⁻¹ kanamycin as an antibiotic marker on Luria–Bertani (LB) agar. A single colony from an agar plate grown overnight was used to inoculate 5 ml LB broth, which was then grown overnight at 310 K with shaking at 180 rev min⁻¹. The starter culture served as an inoculum for a 500 ml culture of LB broth in which cells were grown until the optical density (OD₆₀₀) reached a value of 0.6. The expression of A1R958 was then induced by addition of 1 mM isopropyl β-d-1-thiogalactopyranoside (IPTG). The culture was then incubated at 291 K in an orbital shaker overnight at 180 rev min⁻¹ for approximately 18 h. The cells were harvested by centrifugation for 15 min at 4225g in a Sorvall GS3 rotor in a Sorvall RC5B Plus centrifuge and resuspended in 50 ml 50 mM Tris–HCl buffer pH 7.5 (‘buffer’) containing 300 mM NaCl. The cells were sonicated for 3 × 30 s bursts at 277 K with 1 min intervals and the soluble and insoluble fractions were separated by centrifugation for 30 min at 26 892g in a Sorvall SS34 rotor. The clear supernatant was loaded onto a 5 ml His-Trap Chelating HP nickel column. After washing with ten column volumes of buffer containing 30 mM imidazole, the A1R958 enzyme was eluted with a gradient of 30–500 mM imidazole over 20 column volumes. Column fractions containing the yellow A1R958 protein (as determined by SDS–PAGE analysis) were pooled and concentrated using a 10 kDa cutoff Centricon filter membrane. The concentrated enzyme was then loaded onto a pre-equilibrated S75 Superdex 16/60 gel-filtration column and eluted with 120 ml of the same buffer at a flow rate of 1 ml min⁻¹. Fractions containing pure A1R958, as determined by SDS–PAGE analysis, were pooled and stored at 277 K.

2.2. Protein crystallization  

Crystallization conditions for A1R958 were determined using a range of commercially available trial screens in 96-well sitting-drop format plates with 300 nl drops. The best crystals were obtained using conditions consisting of 28%(w/v) PEG 3350, 0.2 M ammonium acetate, 3%(v/v) dioxane and 1 mM PLP in bis-tris propane buffer pH 7.5 with protein at a concentration of 10 mg ml⁻¹ (1:1 ratio of protein and precipitant solution). Larger crystals for diffraction analysis were prepared using optimized conditions by the hanging-drop vapour-diffusion method in 24-well Linbro plates. The conditions consisted of 28%(w/v) PEG 3350, 0.2 M ammonium acetate, 3%(v/v) dioxane and 1 mM PLP in bis-tris propane buffer pH 7.5. Crystallization drops of 2 µl volume with protein at a concentration of 10 mg ml⁻¹ (1:1 ratio of protein and precipitant solution) were used. Despite the inclusion of PLP in the crystallization solution, initial data sets obtained from crystals grown under these conditions failed to yield significant occupancy for the PLP cofactor in the active sites of the enzyme. In order to improve the occupancy, protein after purification was dialysed against buffer containing 1 mM PLP, concentrated to 10 mg ml⁻¹ and crystallized under the conditions detailed above. Crystals of protein prepared in this way featured PLP density in the OMIT maps after structure solution and refinement (below). In order to obtain the GABA–PLP adduct complex, the protein after dialysis against PLP and concentration was crystallized using equivalent conditions but in the additional presence of 10 mM GABA in the crystallization mixture. Crystals of each complex were flash-cooled in liquid nitrogen in a cryogenic solution consisting of the mother liquor with 10%(w/v) glycerol and tested for diffraction using a Rigaku MicroMax-007 HF fitted with Osmic multilayer optics and a MAR Research MAR345 imaging-plate detector. Those crystals that diffracted to greater than 3 Å resolution were retained for data collection at the synchrotron.

2.3. Data collection, structure solution, model building and refinement of A1R958 structures  

Complete data sets for A1R958 complexed with PLP and with PLP–GABA were collected on beamline I04 at the Diamond Light Source, Didcot, Oxfordshire, England. The data were processed and integrated using XDS (Kabsch, 2010 ▶) and scaled using SCALA (Evans, 2006 ▶) as used within the xia2 processing system (Winter, 2010 ▶). The data-collection statistics are given in Table 1 ▶. The crystals of the complexes were isomorphous and belonged to space group P2₁2₁2. The structures were solved using MOLREP (Vagin & Teplyakov, 2010 ▶) with a monomer of the transaminase from Mycobacterium smegmatis (PDB entry 3oks; 63% amino-acid sequence identity to A1R958; Seattle Structural Genomics Center for Infectious Disease, unpublished work) as a search model. Each solution contained 12 molecules in the asymmetric unit, representing three homotetramers, with a solvent content of 49.2%. The structures were built using Autobuild in the PHENIX suite of programs (Adams et al., 2010 ▶) and Coot (Emsley & Cowtan, 2004 ▶) and were refined using REFMAC (Murshudov et al., 2011 ▶) with local NCS restraints. For the PLP complex structure, following building and refinement of the protein and water molecules, the OMIT maps clearly contained residual density in the region extending continuously from the side chain of Lys295, which was modelled and refined as the internal aldimine adduct of PLP. For data resulting from A1R958 crystals grown in the presence of 10 mM GABA, the OMIT maps in the active site showed discontinuous residual density between the ∊-amino group of Lys295 and PLP, with additional density extending from the region of the PLP aldehydic carbon. This was successfully modelled and built as the GABA–PLP external aldimine adduct. The final structures exhibited R _cryst and R _free values of 19.5% and 22.9%, respectively, for PLP and 20.2% and 24.8%, respectively, for PLP–GABA. Each structure was finally validated using PROCHECK (Laskowski et al., 1993 ▶). Refinement statistics are presented in Table 1 ▶. The Ramachandran plot for the PLP complex showed 94.5% of the residues to be situated in the most favoured regions, 4.8% in additional allowed regions and 0.7% outlier residues. These corresponded to the cofactor-binding residue Lys295, the adjacent Ala294 and also Cys100 in each subunit. For the PLP–GABA complex, the corresponding values were 94.0, 5.3 and 0.7%, respectively. The coordinates and structure factors for the A1R958–PLP complex and the A1R958–GABA complex have been deposited in the Protein Data Bank (Berman et al., 2012 ▶) with accession codes 4atp and 4atq, respectively.

Table 1. Data-collection and refinement statistics for transaminase A1R958 in complex with PLP (internal aldimine) or PLP–GABA (external aldimine).

Values in parentheses are for the highest resolution shell.

	Internal aldimine complex with PLP (PDB entry 4atp)	External aldimine complex with PLP–GABA adduct (PDB entry 4atq)
Beamline	Diamond I04	Diamond I04
Wavelength (Å)	0.9795	0.9795
Resolution (Å)	89.84–2.80 (2.87–2.80)	97.61–2.75 (2.82–2.75)
Space group	P21212	P21212
Unit-cell parameters (Å)	a = 175.7, b = 291.0, c = 104.6	a = 176.7, b = 292.8, c = 106.0
No. of molecules in asymmetric unit	12	12
Unique reflections	125730	136067
Completeness (%)	100 (100)	100 (100)
R merge	0.18 (0.72)	0.12 (0.74)
R p.i.m.	0.075 (0.30)	0.050 (0.31)
Multiplicity	7.4 (7.6)	7.4 (7.5)
〈I/σ(I)〉	10.5 (2.6)	14.3 (2.6)
CC1/2	0.99 (0.78)	1.00 (0.78)
Overall B factor from Wilson plot (Å2)	49	60
R cryst/R free (%)	19.5/22.9	20.2/24.8
R.m.s.d. 1–2 bonds (Å)	0.01	0.01
R.m.s.d. 1–3 angles (°)	1.48	1.45
Average main-chain B (Å2)	28	44.7
Average side-chain B (Å2)	29	45
Average water B (Å2)	16	32
Average ligand B (Å2)	26	43

3. Results  

3.1. Cloning and expression of the A1R958 gene from A. aurescens TC1 and purification of the enzyme  

The gene encoding A1R958 was amplified from the genomic DNA of A. aurescens, cloned into the pET-YSBL-LIC-3C vector using a previously described procedure (Atkin et al., 2008 ▶) and expressed in E. coli BL21 (DE3). Small-scale expression tests revealed the gene to be expressed in the soluble fraction of the same cells. The growth of the cells expressing the A1R958 gene was scaled-up and the enzyme was purified using nickel-affinity chromatography followed by gel filtration. SDS–PAGE analysis of the protein indicated that the protein (calculated monomeric molecular weight of 47.6 kDa) was pure.

3.2. Structures of the GABA-TA A1R958  

The A1R958 transaminase is most similar in sequence overall to GABA transaminases of known structure from M. tuberculosis (PDB entry 3oks; 63% amino-acid sequence identity; Seattle Structural Genomics Center for Infectious Disease, unpublished work), M. marinum (PDB entry 3r4t; 62% amino-acid sequence identity; Seattle Structural Genomics Center for Infectious Disease, unpublished work) and E. coli (PDB entries 1sf2 and 1sff; 44% amino-acid sequence identity; Liu et al., 2004 ▶, 2005 ▶). The structures of A1R958 were initially solved by molecular replacement using a monomer of 3oks as the search model. Although many GABA-TA structures, including 3oks, have been determined with four monomers in the asymmetric unit constituting one homotetramer, which itself is constituted of two homodimers with two PLP molecules bound at the A/B subunit interface, the solutions of both the PLP complex of A1R958 and the GABA–PLP adduct complex were obtained with 12 monomers constituting three tetramers (1, 2 and 3 in the discussions below) in the asymmetric unit. The structure of the A/B/C/D tetramer is shown in Fig. 3 ▶. The monomeric structure overlapped well with the structures of 3oks (r.m.s.d. of 0.8 Å), 3r4t (r.m.s.d. of 0.8 Å) and the E. coli enzyme (r.m.s.d. of 1.4 Å) and gave DALI (Holm & Sander, 1996 ▶) Z-scores of 65.4, 64.4 and 58.7, respectively. The major differences between the published E. coli structure and the A1R958 structure include an extended N-terminal region of approximately 20 amino acids in A1R958 (Fig. 4 ▶). There is also an expanded loop region between residues Glu200A and Glu209A (Gly177A–Asp181A in 1sff), in which Glu201A forms a stabilizing salt-bridge interaction with Arg213D in the equivalent loop of a monomer in the neighbouring dimer. In each subunit of both the PLP and GABA complex structures the peripheral loop regions between Leu369 and Ser375 were poorly defined and were not built.

Figure 3.

Structure of the A1R958–PLP complex tetramer, with subunits A, B, C and D coloured red, blue, coral and green, respectively. PLP molecules, which are bound at the A/B or C/D dimer interfaces, are shown in cylinder format with C atoms in grey.

Figure 4.

Superimposition of the A/B dimer of E. coli GABA-TA (PDB entry 1sff; subunit A, green; subunit B, coral) with the A/B dimer of A1R958 in ribbon format (subunit A, blue; subunit B, red), highlighting the extended N-terminus and the loop between residues 200 and 209 in the A1R958 structure. PLP molecules, which are bound at the A/B dimer interfaces, are shown in cylinder format with C atoms in white (1sff) or grey (A1R958).

3.3. A1R958 structure complexed with PLP  

After model building and refinement, the data sets resulting from initial crystals grown in the presence of 1 mM PLP did not yield substantial electron density in the active-site region(s) between the monomeric subunits of the closest associated dimers that could be interpreted as PLP. In further experiments, pure A1R958 was dialyzed against a solution of 1 mM PLP prior to concentration for crystallization. Further crystals were obtained that diffracted to approximately 2.8 Å resolution. After structure solution using the apo A1R958 structure obtained in the initial experiments, significant density was clearly visible in the OMIT map extending from the active-site Lys295 and into the active site. PLP was built into each of these 12 OMIT-map peaks and the structures were refined after building the bond from the ∊-amino group of Lys295 to the aldehydic carbon of PLP to form a Schiff-base linkage. The structure of the active site with the PLP adduct is shown in Fig. 5 ▶. In the descriptions of interactions that follow in §4, the dimer formed by subunits A and B is used as the model dimer to illustrate the interactions observed between all dimer pairs (A/B, C/D, E/F, G/H, I/J and K/L) in the structure, where A represents the first subunit and B the second. The pyridine N atom of PLP is held by a salt-bridge interaction with the carboxylate side chain of Asp266A and the phenolic group with the side chain of Gln269A. The planar aromatic ring is stacked between the side chains of Tyr161A and Val268A; the phosphate is secured by interactions with the backbone NH groups of Gly134A and Ala135A, the peptidic NH and the side chain of Thr324B and two water molecules. Residues Ile73A, Glu238A and Arg429A correspond to Ile50A, Glu211A and Arg398A, respectively, in the E. coli GABA-TA structure and presumably play equivalent roles in substrate recognition and catalysis, as described in §1. Notable differences in the active-site region between the two enzymes include the substitution of Gln279B (Ec-GABA-TA) by Met102B (A1R958) on the opposite face of the substrate channel facing Arg164A and the replacement of Ser112A in the E. coli enzyme, which binds the PLP phosphate, by an alanine residue, Ala135A.

Figure 5.

Structure of the active site of the A1R958 internal aldimine complex with the side chain of Lys295A covalently bound to the aldehydic C atom of PLP (shown in ball-and-stick format). Electron density in blue corresponds to the OMIT (F _o − F _c) map at a level of 3σ obtained from refinement performed in the absence of both Lys295A and the cofactor.

3.4. A1R958 structure complexed with GABA–PLP adduct  

In order to obtain an external aldimine complex of A1R958 with an amino-acid substrate, the pure protein that had been dialysed against 1 mM PLP was crystallized using equivalent conditions to those used for the PLP complex but in the presence of 10 mM GABA. Solution of the structure using a monomer of apo A1R958 again resulted in a structure that featured 12 monomers in the asymmetric unit making up three tetramers. In the OMIT maps that resulted from the first rounds of refinement, it was clear that the covalent bond between the aldehydic carbon of PLP and the side chain of Lys295 had been broken and that linear residual electron density extended from the PLP aldehydic carbon to a bifurcated terminus into a substrate-binding channel between the monomers of each dimer (Fig. 6 ▶). A GABA molecule complexed with PLP through the aldehyde C atom to form an external aldimine adduct was successfully modelled into this density in each subunit. In addition to the bonding interactions observed for PLP in the active site of the A1R958–PLP complex and the movement of the Lys295 side chain, the GABA adduct was secured by ionic interactions between its carboxylate terminus and both the NE atom and the amino-terminus of the side chain of Arg164A. Apart from the movement of the lysine side chain away from PLP and a slight rotation of PLP itself along its longitudinal axis, very little movement of either secondary elements, tertiary structure or side chains was noticed between the PLP complex and the GABA–PLP complex. The r.m.s.d. between the A/B/C/D tetramer in each complex, for example, was 0.15 Å.

Figure 6.

Structure of the active site of the A1R958 external aldimine complex with the aldehydic C atom of PLP now bound to the amine N atom of GABA (shown in ball-and-stick format). Electron density in blue corresponds to the OMIT (F _o − F _c) map at level of 3σ obtained from refinement performed in the absence of Lys295A and the PLP–GABA adduct.

4. Discussion  

The structures of subgroup III transaminases from A. aurescens TC1 are interesting both with respect to the characterization of the amine-metabolizing complement of this organism and also to investigation of the potential of these enzymes for industrial biotransformation reactions (Hwang et al., 2005 ▶). Whilst the structure of the A1R958 enzyme overlaps closely with that of the E. coli enzyme with 44% sequence identity, the complex structure of the latter enzyme was obtained in the presence of the inhibitor AOA rather than the native substrate for the first half-reaction, GABA itself. The A1R958 structure in complex with the external aldimine PLP–GABA adduct therefore provides the first structure of a GABA-TA in complex with its natural external aldimine. Superimposition of the PLP–GABA complex of A1R958 and the E. coli GABA-TA complexed with AOA (PDB entry 1sff; Fig. 7 ▶) reveals that despite the longer chain of GABA compared with the inhibitor, the carboxylates of the two molecules superimpose almost exactly, with approximately the same distance between the ligand carboxylate(s) and the terminal amino group of Arg164A (Arg141 in E. coli GABA-TA). Substitution of Gln79B in the E. coli enzyme by Met102B in A1R958 results in the Met side chain being at a greater distance (4 Å) from the GABA carboxylate than in the 1sff structure (3.4 Å) and of course removes the potential for hydrogen-bonding interaction between the ligand and the side chain. However, one additional interaction is made in the A1R958 GABA complex between the carboxylate of GABA and the backbone NH of Gly323B which does not feature in the E. coli enzyme in complex with AOA, perhaps offsetting the loss of the Gln79B interaction in the latter structure.

Figure 7.

Superimposition of the active site of the A1R958 GABA–PLP external aldimine complex (C atoms in grey) with E. coli GABA-TA complexed with aminooxy­acetate (AOA; C atoms shown in white). The first residue number corresponds to the A1R958 sequence. Residue(s) Tyr161(138)A have been removed for clarity. The carboxylate of the GABA–PLP complex is seen to make interactions (shown in black dashed lines) with both the side chain of Arg164A and the backbone NH of Gly323B.

The structures of transaminase enzymes from bacteria serve as useful models of mammalian TA action, but also provide information on enzymes that will continue to serve as a useful biotechnological resource in the future. The structure of a natural GABA–PLP complex provides new information on GABA binding within this subgroup of TAs and also allows comparison with ω-transaminases of the same subgroup (Humble et al., 2012 ▶), thus assisting in the identification of active-site residues that help to discriminate individual substrate-binding profiles in each of these enzymes.

Supplementary Material

PDB reference: A1R958, PLP complex, 4atp

PDB reference: PLP–GABA adduct complex, 4atq