Crystal Structures of Lysine-Preferred Racemases, the Non-Antibiotic Selectable Markers for Transgenic Plants

Lysine racemase, a pyridoxal 5′-phosphate (PLP)-dependent amino acid racemase that catalyzes the interconversion of lysine enantiomers, is valuable to serve as a novel non-antibiotic selectable marker in the generation of transgenic plants. Here, we have determined the first crystal structure of a lysine racemase (Lyr) from Proteus mirabilis BCRC10725, which shows the highest activity toward lysine and weaker activity towards arginine. In addition, we establish the first broad-specificity amino acid racemase (Bar) structure from Pseudomonas putida DSM84, which presents not only the highest activity toward lysine but also remarkably broad substrate specificity. A complex structure of Bar-lysine is also established here. These structures demonstrate the similar fold of alanine racemase, which is a head-to-tail homodimer with each protomer containing an N-terminal (α/β)8 barrel and a C-terminal β-stranded domain. The active-site residues are located at the protomer interface that is a funnel-like cavity with two catalytic bases, one from each protomer, and the PLP binding site is at the bottom of this cavity. Structural comparisons, site-directed mutagenesis, kinetic, and modeling studies identify a conserved arginine and an adjacent conserved asparagine that fix the orientation of the PLP O3 atom in both structures and assist in the enzyme activity. Furthermore, side chains of two residues in α-helix 10 have been discovered to point toward the cavity and define the substrate specificity. Our results provide a structural foundation for the design of racemases with pre-determined substrate specificity and for the development of the non-antibiotic selection system in transgenic plants.

Two types of amino acid racemases are known: pyridoxal 59phosphate (PLP)-dependent [6] and PLP-independent enzymes [8]. Alanine racemase (Alr, E.C. 5.1.1.1), which catalyzes the interconversion of D-and L-Ala, is the best characterized of the prokaryotic and eukaryotic PLP-dependent enzymes. Several Alr crystal structures have been determined, and the structures are compact homodimers with each protomer containing an (a/b) 8 barrel and a C-terminal domain having three b-sheets [9,10,11,12,13,14]. The dimeric interface contains the active-site cleft, which is formed by residues from the (a/b) 8 barrel of one protomer and residues from the C-terminal domain of the other protomer. Structural and site-directed mutagenesis studies of Bacillus stearothermophilus Alr (BsAlr, PDB code: 1L6G) have identified two catalytic residues: Lys 39 from one subunit and Tyr 265 from the other subunit [13,15,16,17]. When alanine is absent, the C4A atom of PLP forms an aldimine bond with the Lys39 Nf of BsAlr, whereas PLP forms an aldimine bond with the amino nitrogen of the alanine substrate during catalysis [9,13].
A constricted entryway has been characterized in Mycobacterium tuberculosis Alr (MtAlr; PDB code: 1XFC) [18] that is believed to be the active site. The entryway is formed by three layers of residues (outer layer: Asp 357 , Lys 178 , and Ala 241 ; middle layer: Arg 3169 , Ile 362 , Arg 2969 , and Asp 177 ; inner layer: Tyr 2719 , Tyr 364 , Tyr 2909 , and Ala 176 ) (prime symbol represents the residue from the other subunit of a dimer). After the entryway, PLP is found at the base of the cavity, and importantly, the strict Alr specificity for alanine is defined by the relatively small separation (,2.7 Å ) between two innermost layer tyrosines: MtAlr, Tyr 2719 and Tyr 364 ; Streptococcus pneumoniae Alr (SpAlr, PDB code: 3S46 [19]), Tyr 2639 and Tyr 352 ; BsAlr, Tyr 2659 and Tyr 354 [18].
In addition to Alrs, a PLP-dependent racemase is found to possess a broad amino-acid specificity (broad specificity amino acid racemase, Bar, E.C. 5.1.1.10 [20,21]). Moreover, several microorganisms, including Proteus vulgaris [22], P. putida [23], Proteus spp. and Escherichia spp. [24], have lysine racemization activities, although the genes encoding lysine racemases and the lysine racemases themselves have not been isolated. Recently, lyr encoding a lysine racemase was obtained from a library con-structed from the genetic material of garden soil microbes [25]. By screening a P. mirabilis BCRC10725 genomic library, a lyr was found that complemented a lysine-auxotrophic P. mirabilis mutant cultured on M9-containing agar supplemented with D-lysine [26].
Of note, the sequences of Bar and Lyr contain conserved lysine and tyrosine residues at positions equivalent to the catalytic residues of BsAlr (Bar: Lys 75 , Tyr 3019 ; Lyr: Lys 74 , Tyr 2999 ), despite their otherwise limited sequence identity (28% between Bar and BsAlr; 24% between Lyr and BsAlr). Furthermore, the P. mirabilis BCRC10725 lyr sequence is identical with P. mirabilis ATCC29906 gene (NCBI accession number: ZP 03840526), which is annotated as a putative alanine racemase. According to a phylogenetic study that used Molecular Evolutionary Genetics Analysis, 24 prokaryotic amino acid racemases from 18 species have been classified into different groups: Group I contains Alrs from Gram-negative bacteria, Group II contains Alrs from Gram-positive bacteria, and Group III are non-Alr racemases, which have evolutionarily diverged from group I and II Alrs [26]. Lyr and Bar are therefore Group-III racemases. Expression and biochemical characterization of recombinant Lyr has shown that its preferred substrate is lysine [26]. Notably, Lyr is also active against arginine, whereas Bar racemizes arginine, lysine, methionine, serine, cysteine, leucine, and histidine [27], a useful feature for biotechnological applications.
It is interesting that lyr has been successfully used as a nonantibiotic selectable marker gene for plant transformation lately [28] since L-lysine is toxic to tobacco and Arabidopsis while Dlysine supports their growth [28]. Such a non-antibiotic selection system for transgenic plants is much desirable due to the ecological concerns and food safety issues in this modern world. To further develop lyror bar-based marker genes with desired properties, we determined the three-dimensional structures of P. putida DSM84 Bar and P. mirabilis Lyr, which are the first structures for lysinepreferred racemases. In addition, the first liganded Bar-lysine structure was also established. All structures contain PLP, which confirms their PLP-dependent nature, and the positions of their catalytic lysines and tyrosines are similar to those in Alr structures. Site-directed mutagenic, kinetic, and modeling studies characterized the Lyr/Bar PLP-binding residues and the residues responsible for their substrate specificities.

Bar and Lyr Structures
The final Lyr and Bar structures had R values of 22.0% (R free = 26.0%) and 15.8% (R free = 21.5%), respectively. Refinement statistics are given in Table 1. The asymmetric unit in the Bar crystal contains two molecules. With the exception of the positions for the first 25 residues, those of all other main-chain and side-chain atoms are well defined. Non-peptide electron density near Lys75 was fit with an aldimine bond between the Lys75 Nf and the PLP C4A. The average B-factor is 37.4 Å 2 . The Bar molecule contains two domains: an N-terminal (a/b) 8 barrel, which contains PLP, and a b-stranded C-terminal domain ( Figure 1A). Two Bar molecules associate to form an AB-type dimer, and their positions are related by a noncrystallographic two-fold axis ( Figure 1B). The subunits have essentially the same fold (root mean square deviation (RMSD) = 0.219 Å for equivalent Ca positions).
We next solved the Lyr structure by molecular replacement with the Bar structure as the template. One Lyr molecule is found in the asymmetric unit. With the exception of those for the first 36 residues, the positions of all main-chain and sidechain atoms are well defined. The average B-factor is 29.2 Å 2 .
Molecular Lyr and Bar have the same domain structure, and PLP is found in an equivalent position in the two molecules ( Figure 1C). Two Lyr molecules associate to form a dimer related by a two-fold crystallographic axis. In the Bar and Lyr dimers, the two protomers are arranged in a head-to-tail manner, stabilized by multiple interactions between N-terminaldomain residues in one monomer and C-terminal-domain residues in the other protomer ( Figure 1B, D). The tertiary structures of the Lyr and Bar protomers are similar (RMSD = 0.89 Å for Ca positions).

Structural Comparisons
When the Lyr or Bar structure was used as the query, the topscoring proteins (Z scores .38) returned by DALI were Alrs, even though the sequence identities for the Alrs and Lyr, and the Alrs and Bar are low (20227% identity for Lyr; 23231% identity for Bar). The tertiary folds of Lyr, Bar, and the Alrs are similar [29], and the protomers in each racemase are oriented head-to-tail.
Superposition of Lyr, Bar, SpAlr, BsAlr, and SlAlr confirmed their structural similarity (Figure 2A). The C-terminal domains superimpose relatively well (Q-scores within 0.68-0.95 for Ca positions) [30]. However, the Ca positions of a few regions in the N-terminal domains (corresponding to residues 1082121, 1492155, and 1912195 in Lyr) have lower structural similarities between our structures and Alrs (Q-scores within 0.45-0.60). Moreover, the Ca positions in the g3-b10 loop between the Nand C-terminal domains have the greatest deviation (Q-scores within 0.38-0.45) and show that the hinge angles between the monomer domains are slightly different (Figure 2A), which are also discovered in Alr structures [14,19]. Lyr Lys 74 and Bar Lys 75 reside in the b2-a2 loop and bind PLP via an aldimine bond as has been found for the corresponding lysines and PLP in the Alrs ( Figure 2B). The combined sequence and secondary structure alignments of Lyr, Bar, and the three Alrs demonstrate that this loop along with b2, a2, b11, b11-b12, and b12 share similar secondary structural elements ( Figure 3). Additionally, the Ca atoms in the sequences of these regions in Lyr and Bar superimpose very well with those in the Alrs (Q-scores within 0.85-0.95). Therefore, Lyr, Bar, and the Alrs probably originated from a common ancestor involved in racemization of amino acids for the synthesis of cell building blocks.

Racemase Active Sites
The Alr active-site cleft is located at the dimeric interface and contains the two conserved catalytic residues, a lysine from one subunit and a tyrosine from the other subunit (SpAlr: Lys 40 and Tyr 2639 ; BsAlr: Lys 39 and Tyr 2659 ; SlAlr: Lys 38 and Tyr 2709 [9,11,31]). Sequence alignment of Lyr, Bar, and the Alrs showed that the two catalytic residues are conserved in Lyr and Bar ( Figure 2B). Furthermore, superpositioning the Lyr, Bar, SpAlr, BsAlr, and SlAlr structures showed that the two catalytic residues are situated at nearly identical positions within all five structures ( Figure 1, 2, 3).
Most of the Alr PLP-binding residues are also conserved in Lyr and Bar (Table 2). Two conserved arginines (Lyr: Arg 173 and Arg 260 ; Bar: Arg 174 and Arg 261 ; SpAlr: Arg 136 and Arg 218 ; BsAlr: Arg 136 and Arg 219 ; SlAlr: Arg 136 and Arg 224 ) in the (a/b) 8 barrel interact with the pyridine N1 and the phenolic O (O3) of PLP, respectively, and they orientate the PLP phenolic ring. A conserved lysine in SpAlr/BsAlr (Lys 129 ) can be carbamylated at its Nf (KCX) and interacts electrostatically with the guanidinium group of Arg 136 , thereby further stabilizing the position of PLP [12,14]. However, an equivalent lysine is not found in Lyr or Bar ( Figure 4). Instead, the side chain of the asparagine that is adjacent to the corresponding arginine (Lyr: Asn 174 ; Bar: Asn 175 ) makes contacts with the arginine guanidinium moiety in Lyr and Bar, suggesting that these Asn/Arg interactions stabilize the orientation of PLP.

Lyr and Bar Substrate Binding Cavities
A funnel-like cavity ( Figure S1) consisting of inner and outward layers is seen for Lyr or Bar ( Figure 5), an arrangement that is also found in MtAlr [18]. An inner-layer tyrosine located in a-helix 10 (Tyr 352 in SpAlr; Tyr 354 in BsAlr) interacts with the PLP O3P-an interaction believed to be necessary for substrate specificity [32]. Together with the catalytic tyrosine, these two tyrosine residues (SpAlr: Tyr 352 and Tyr 2639 ; BsAlr: Tyr 354 and Tyr 2659 ), which are separated by ,2.7 Å , are strictly conserved and orient the alanine substrate and PLP properly. The a-helix 10 Tyr is not, however, present in Lyr and Bar, but is replaced by Thr 391 in Lyr and Ala 393 in Bar ( Figure 425). Two water molecules contact the PLP O3P in Lyr and Bar, filling the space occupied by the tyrosine and PLP in SpAlr and BsAlr ( Figure 4). Given the relatively larger space between Thr 391 and PLP in Lyr, and Ala 393 and PLP in Bar as compared with those in the Alrs, a substrate larger than an alanine may better fit the space.
For E. coli Alr, negatively and positively charged residues are found on the opposite sides of the surface of its cavity surface (Asp 164 , Glu 165 and Arg 2809 , Arg 3009 ) [33]. This array may help properly orientate the substrate for catalysis [33]. Superpositioning the corresponding residues in Lyr (Glu 208 , Asp 209 , Arg 3249 , and Lys 3449 ), Bar (Glu 209 , Asp 210 , Arg 3269 , and Lys 3469 ), SpAlr (Asp 170 , Glu 171 , Arg 2889 , and Arg 3079 ), and BsAlr (Asp 171 , Glu 172 , Arg 2909 , and Arg 3099 ) revealed that the charge distribution is similar in the four racemases ( Figure 5, see also Figure   S1), which supports the suggestion that these residues orient the substrate for catalysis.

Effects of Mutations on Enzymatic Activity
To evaluate the roles of Arg 173 , Asn 174 , and Thr 391 in Lyr and Arg 174 , Asn 175 , and Ala 393 in Bar (corresponding to Arg 136 , Leu 137 , and Tyr 354 in BsAlr, respectively), we performed sitedirected mutagenesis and characterized the effects of the mutations had on racemase activity. The conserved Arg 173 in Lyr (Arg 174 in Bar) was replaced with an alanine or a lysine. Asn 174 in Lyr (Asn 175 in Bar) was replaced with a leucine to mimic Leu 137 in BsAlr. Thr 391 in Lyr (Ala 393 in Bar) was replaced with a tyrosine to mimic the conserved Alr tyrosines (Figure 4; Tyr 354 in BsAlr and Tyr 352 in SpAlr). Each mutant was expressed as a (His) 6tagged protein in E. coli and purified by Ni-affinity chromatography. CD analysis of each mutant revealed nearly identical profile to that of the wild-type enzyme ( Figure S2). The racemase activity of each mutant toward L-Lys, L-Arg, and L-Ala was measured.
R173A and R173K were inactive (Table 3) as were the corresponding Bar mutants R174A and R174K, supporting the importance of a conserved arginine at this position. Possibly, the Lyr Arg 173 guanidinium group orients the PLP pyridine ring, to position the PLP-Tyr 2999 interaction for catalysis. The positive electrostatic field of the guanidinium group may also help lower the pKa value of the Tyr 2999 phenolic hydroxyl (pKa = 8.0-9.0) for effective catalysis. A replacement of the neighboring asparagine with a leucine (N174L in Lyr and N175L in Bar) abolished racemization activity, which supports our hypothesis that this asparagine helps orient the PLP pyridine ring, revealing a comparable role as KCX found in Alr in conjunction with Arg 173 in Lyr and Arg 174 in BAR.
Mutation of the inner-layer Thr 391 of Lyr to tyrosine, which is positioned similarly to the conserved tyrosines in Alrs (Tyr 354 in BsAlr), reduced the activity of Lyr towards L-Lys to 53% ( Table 3). The L-Lys K m value for this mutant was greater than that for Lyr, whereas the k cat value for the mutant with L-Lys as the substrate was 42% of Lyr (Table 4). Consequently, racemase activity for the mutant was significantly reduced (,50% residual activity, Table 3). A larger K m value towards L-Arg was also measured ( Table 4), suggesting that the T391Y mutation might reduce the binding affinity for substrates with bulky side chains. In contrast, the k cat values were comparable with L-Arg as the substrate, indicating that the transaldimination rate of PLP and L-Arg was unaffected by the mutation. No detectable activity was found when L-Ala was the substrate, implying that alanine still could not access the Lyr-T391Y active site.
The Bar A393Y mutant exhibited detectable, although reduced activity towards L-Lys (55% residual activity, Table 3). The k cat value for the mutant with L-Lys as the substrate was reduced (44% residual activity, Table 4), and the K m value increased 1.6-fold resulting in an overall reduction in activity. A393Y also was less activity toward L-Ala (25% residual activity) as compared with the wild-type enzyme ( Table 3) and was inactive toward L-Arg. The tyrosine replacement at residue 393 in Bar therefore significantly reduced racemization activity, particularly when L-Arg was the substrate. Interestingly, Enterococcus gallinarum VanT, the C-terminal domain of which shows 31% sequence identity with BsAlr, is a serine racemase that has an asparagine at the site equivalent to position 393 in Bar [32]. Together, these results support the possibility that the residue at this site is critical in defining substrate specificity.
We also generated the Lyr triple mutant A165K/N174L/ T391Y (Bar, A166K/N175L/A393Y) to mimic the corresponding residues in Alrs. However, no activity was detected towards L-Ala, L-Lys, or L-Arg. Superpositioning of Lyr and the Alrs revealed a subtle difference in the active-site frameworks ( Figure 5, see also Figure S1), which might explain the inability of the triple mutant to racemize L-Ala.
Of note, the nearby Bar residue Tyr 396 in the same helix (a10), for which the phenolic ring also points toward the cavity, has been identified as a valuable engineering site by Kino and colleagues [27]. Substitution of Bar Tyr 396 with a cysteine or histidine greatly increased racemization activity toward tryptophan [27]. When the corresponding Lyr site was mutated (S394RY), the activity towards L-Lys was reduced to 31% that of Lyr, whereas activity towards L-Arg increased 2.1-fold [26]. Thus, this site may be also important for catalytic activity in Bar and Lyr. We generated the Lyr double mutant T391Y/S394Y, which exhibited significantly reduced specific activity towards L-Lys (6% residual activity) and toward L-Arg (0.9% residual activity) ( Table 3). The k cat value when L-Lys was the substrate decreased by 22-fold, whereas the K m value decreased by only 1.6-fold (Table 4). Given these results, it is likely that the Thr 391 and Ser 394 side chains, which point toward the entryway, might help accommodate or orient the substrate to effect transaldimination with PLP. An even greater change in k cat (28-fold) and in K m (5-fold) was found when L-Arg was the substrate, suggesting that Thr 391 and Ser 394 in Lyr (Ala 393 and Tyr 396 in Bar) contribute to the binding and orientation of the substrate and therefore to catalytic activity. By considering specificity constants ((k cat /K m )Lys/(k cat /K m )Arg) in comparing mutations of Lyr and Bar, T391Y has no significant change in substrate specificity (,2.9 fold) but T391Y/S394Y does enhance the Lys specificity over Arg (,15.0 fold). This implies that Ser 394 Figure 3. Sequence Alignment of Bar, Lyr, SpAlr, BsAlr, and SlAlr. The catalytic residues, PLP-binding residues, and the mutation sites are identified with a red star, blue circle, and m (green), respectively. The relative conservation of each residue was drawn by the web-server WEBLOGO (http://weblogo.berkeley.edu/). doi:10.1371/journal.pone.0048301.g003

Docked Lyr and Bar Models
We  Table S22S5. In Lyr-PLY ( Figure 6), a few residues closely interact with cross-linked D-Lys: (i) the catalytic residues Lys 74 and Tyr 2999 contact the D-Lys amino nitrogen; (ii) Tyr 3189 and Met 3479 interact with the D-Lys carboxyl oxygens; and (iii) Thr 391 interacts with the D-Lys side-chain Nf. Similar structural features are also found in Bar-PLY, except for a polar interaction between Ala 393 and D-Lys ( Figure 6).
Conversely, Lyr-PDD and Bar-PDD ( Figure 6) contain many fewer polar interactions between D-Ala and PLP. In Lyr-PDD, Lys 74 does not hydrogen bond with the amino nitrogen of D-Ala. Instead, D-Ala is positioned closer to Arg 3249 , which is why L-Ala is not racemized by Lyr (Table 3). In Bar-PDD, D-Ala is closer to the PLP phosphate, which might also reduce its accessibility to the catalytic residue, Tyr 3019 , and therefore decrease racemase activity.

The Bar-Lys Complex Structure and Proposed Catalytic Mechanism
The complex-Bar structure was established ( Figure 7A, B) by cocrystallizing Bar with L-lysine. The final Bar-L-lysine structure had R values of 17.5% (R free = 21.9%). The apo and liganded Bar structures share the same two-domain architecture and are both composed of the AB-type dimer. Superposition of the apo-and complex-form reveals a RMSD of 0.34 Å for the Ca atoms, demonstrating an overall identical conformation. Moreover, structural comparison of complex-Bar structure and Bar-PLY docking model shows that the reactive intermediates (PLP-lysine) are situated at the comparable position and surrounded by identical residues to form resembling hydrogen-bonding networks ( Figure 7C), suggesting the feasibility of the docking approach utilized here. Two different mechanisms have been previously proposed for the reversible racemization of Alr enzymes: (1) the classical twobase mechanism [9] that describes the involvement of the quinonoid intermediate in the proton transfer between the two catalytic bases; and (2) a revised mechanism that proposes the substrate carboxylate group directly involved in the proton transfer based on the observation of a positively charged arginine (Arg 219 ) situated by the pyridine nitrogen atom of PLP in structures of Alr-L-ala and Alr-D-ala [13]. Given the intimate contact between Arg 219 and the pyridine nitrogen atom of PLP, it is not likely that a quinonoid intermediate proposed in the classical two-base mechanism could be formed. Instead, the substrate carboxylate group that is located in proximity to the two catalytic bases (Lys 39 and Tyr 2659 ), proposed in the revised mechanism, is most plausible to serve as the critical moiety to mediate the proton transfer between the two catalytic bases.
The determined Bar-Lys complex structure also shows that the positive guanidium group of Arg 261 (corresponding to Arg 260 of Lyr) is close to the pyridine nitrogen atom of PLP ( Figure 7C), hence may not favor the formation of the quinonoid intermediate proposed in the classical two-base mechanism. Moreover, the carboxylate group of L-lysine is situated at a position, allowing hydrogen-bonding interactions with both catalytic bases, respectively. Our docking models ( Figure 6) are also in good agreement with the liganded Bar-Lys structure. These results together suggest the importance of the substrate carboxylate involved in the catalytic racemization for Lyr and Bar (FigureS3).
In conclusion, the determined structures of apo and L-Lysliganded Bar present the internal and external aldimine linkages, respectively, as well as the anchored substrate in Bar. The apo Lyr structure has also been established within the internal aldimine linkage. Based on these structures, site-directed mutagenesis, kinetic, and modeling studies, a coordinated response between residues from a-helix 10 and those from the most peripheral region of the binding pocket (corresponding to Glu 208 and Asp 209 , and Arg 3249 and Lys 3449 of Lyr) is essential for lysine-preferred racemases to ensure the substrate orientation, entry and specific binding, hence PLP-dependent catalysis. Our results also shed light on the evolution of PLP-dependent amino acid racemases that retain a converged funnel-like pocket having an outward layer that help anchor the substrate while a base platform for an efficient catalytic racemization. Structural and biochemical investigations in this study provide the key basis to figure out the enzymatic mechanism of Lyr and Bar, which helps to develop and improve the non-antibiotic selection system for transgenic plants.

Materials and Methods
Cloning, Expression, and Purification P. mirabilis BCRC10725 lyr was PCR amplified from chromosomal DNA using gene specific primers (Table S1). Polymerase chain reaction (PCR) was performed with pfu DNA polymerase using a RoboCyclerH GRADIENT 96 (Stratagene, La Jolla, CA, USA). Initial denaturation was performed at 95uC for 6 min followed by 30 cycles of denaturation at 95uC for 1 min, annealing at 55uC for 45 s, and primer extension at 72uC for 90 s. The amplified product was inserted into pET21a (Novagen, Inc., USA) to generate pET21a-lyr. Escherichia coli BL21(DE3) cells were transformed with pET21a-lyr, and expression of lyr was induced by addition of 0.5 mM isopropyl-b-D-thiogalactopyranoside (IPTG) (final concentration). An overnight culture of E. coli BL21(DE3) carrying pET-lyr or mutants was inoculated into 500 ml of LB Figure 5. Crucial residues of binding cavities (See also Figure S1). (A) Crucial residues of Lyr, Bar, SpAlr and BsAlr in the binding cavities. (B) Superposition of binding cavities in Lyr (green), Bar (magenta), SpAlr (yellow), and BsAlr (cyan), respectively. Crucial residues are displayed as stick models. The ligands presented in binding pockets of all structures are drawn as ball-and-stick models. Oxygen, nitrogen, and phosphate atoms are colored red, blue, and orange, respectively. doi:10.1371/journal.pone.0048301.g005 Table 3. Racemization Activities toward L-Lys, L-Arg, or L-Ala by Wild-type Lyr and Bar, and Their Mutants.  medium containing ampicillin (100 mg/ml) and incubated at 37uC until an OD 600 of 0.7 was reached. Expression of the gene was induced by the addition of 0.5 mM IPTG and incubated at 17uC for 16 h. The cells were harvested by centrifugation at 10,0006g for 10 min and resuspended in 50 ml of 100 mM potassium phosphate buffer (pH 8.0) containing 150 mM NaCl and disrupted by sonication. The cell debris was pelleted at 10,0006g at 4uC for 10 min and the resulting supernatant was loaded onto a precharged nickel affinity column (HisBind Quick Column, Novagen, Madison, WI, USA). The His 6 -tagged proteins were eluted from the column using a buffer containing 150 mM NaCl, 100 mM imidazole and 100 mM potassium phosphate buffer (pH 8.0). The purified Lyr was verified by SDS-PAGE analysis. Protein concentration was assayed by the Bradford method using bovine serum albumin as the standard [34]. Pseudomonas putida DSM84 bar was PCR amplified from chromosomal DNA using gene specific primers (Table S1) and inserted into pET21a to generate pET21a-bar. The expression and purification of Bar were as described above.

Site-Directed Mutagenesis
pET21a-lyr and pET21a-bar served as the templates for overlap extension PCR amplification [35] of mutated genes using the primers listed in Table S1. Each amplified mutated gene was cloned into an XbaI-XhoI site in pET21a. The recombinant plasmids were individually introduced into E. coli NovaBlue or E. coli BL21(DE3). After isolating each plasmid, the presence of the desired mutation was confirmed by DNA sequencing.

Circular Dichroism Analysis
All circular dichroism (CD) measurements described here were made by using 1.0 mm cells. Spectra were recorded at 25uC on AVIV 62A DS spectropolarimeter (Aviv Associates, Lakewood, NJ). The concentration of proteins used for wavelength scanning of CD measurement was 0.08 mg/ml. All spectra were recorded from 195 to 260 nm. Three spectra were recorded and averaged for each protein.

Enzyme Assay
Lyr and Bar racemase activities were measured using a 0.5 ml mixture of 0.5-1.0 U of purified enzyme, 100 mM CHES-NaOH (pH 9.0), 100 mM of an L-amino acid enantiomer, and 30 mM PLP. Each mixture was incubated at 50uC for 10 min, and the reactions were terminated by heating at 100uC for 10 min. The chiralities of the amino acids were characterized by highperformance liquid chromatography using a Chirobiotic T column (4.06150 mm, Advanced Separation Technologies, USA) at a flow rate of 0.5 ml/min. The mobile phase was 50% ethanol/50% NaH 2 PO 4 (pH 4.5). The detection wavelength was 210 nm. One unit of racemase activity was defined as the amount of enzyme that produced 1 mmole of the amino acid enantiomeric product from the enantiomeric reagent per min.
A steady-state kinetics study of the purified enzymes was performed at 50uC with a reaction mixture that contained 100 mM CHES-NaOH (pH 9.0), 30 mM PLP, and an amino acid enantiomer (concentration, 10-150 mM). The K m and k cat values were determined by plotting the initial velocity as a function of substrate concentration and fitting the plots to the Michaelis-Menten equation.   Figure 6. Protein-ligand interactions in crystal structures and docked models (See also Figure S4 and Table S2

X-ray Data Collection and Processing
Crystals were flash frozen in a stream of liquid nitrogen and then screened and characterized using an RU-300 rotating-anode X-ray generator (Rigaku/MSC Inc., USA) at the Macromolecular X-ray Crystallographic Laboratory of National Tsing Hua University, Taiwan. The Lyr dataset was collected at the SPring-8 BL44XU beamline, Japan, with a Bruker-AXS DIP-6040 detector. Bar and SeMet-Bar datasets were collected at the SPXF beamline BL13B1 at the National Synchrotron Radiation Research Center, Taiwan, with an ADSC Quantum-315r CCD area detector. The multi-wavelength anomalous dispersion (MAD) dataset for a single SeMet-Bar crystal was collected at three wavelengths: peak (0.9791 Å ), selenium K-edge inflection (0.9792 Å ), and high remote (0.9638 Å ). The complex-Bar dataset was collected at the SPring-8 BL12B2 beamline, Japan, with the ADSC Quantum-210r CCD area detector. All datasets were indexed, integrated, and scaled using HKL-2000 [37]. Data collection statistics are shown in Table 1.

Structure Determination and Refinement
Thirteen selenium atoms were located in the asymmetric crystal unit of SeMet-Bar by the program PHENIX [38]. PHENIX was also used for phasing, density modification, and automated model building. The final model included 93.9% of the residues. The first 25 residues of Bar contained the signal peptide and were removed by auto-cleavage. Crystallographic refinement used the maximumlikelihood target function module in REFMAC5 [39,40]. The apoand complex-form Bar structures were constructed by MOLREP [39,40,41] and were refined using REFMAC5 coupled with ARP/ wARP [42], which automatically added water molecules.
The Lyr structure was constructed by the molecular replacement program in PHENIX with the Bar protomer (50% identity) as the template. Crystallographic refinement used the maximum-likelihood target function as implemented in RE-FMAC5 and coupled to ARP/wARP.
For the Lyr, Bar, complex-Bar and SeMet-Bar structures, 2Fo-Fc and Fo-Fc maps were produced and inspected after each automated refinement cycle to manually refine the structures with the aid of COOT [43]. The validities of the Lyr, Bar and complex-Bar structures were assessed by PROCHECK [44].
The atomic coordinates and structure factors have been deposited in the Protein Data Bank, Research Collaboratory for Structural Bioinformatics, Rutgers University, New Brunswick, NJ (http://www.rcsb.org/pdb/) under the accession codes 4DZA (Lyr), 4DYJ (Bar), and 4FS9 (Bar-L-lysine).

Structural and Sequence Comparisons
The Lyr and Bar structures were compared with all protein structures in the DALI server (http://ekhidna.biocenter.helsinki. fi/dali_server/). The structures of Lyr, Bar, complex-Bar, BsAlr [13], SpAlr [19], Streptomyces lavendulae Alr (SlAlr, PDB code: 1VFS; [12], and E. coli diaminopimelate decarboxylase (PDB code: 1KO0) (no reference), were superimposed by LSQMAN in O [45] to determine the pair-wise RMSD of the Ca atom positions. Qscore, a common measure to present the three-dimensional similarity, was applied to structural comparisons and ranged from 0, where no similarity exists, to 1 where structures are identical [30,46,47]. ESPript was used for the combined sequence and secondary structure alignments and figure preparation [48]. PyMol (http://www.pymol.org) was used to prepare the figures containing structures.

Structural Modeling
Discovery Studio v3.0 (Accelrys Inc., USA) was used to prepare, energy minimize, and refine the Lyr and Bar structures prior to docking D-Ala and D-Lys into Bar and Lyr, respectively, and for molecular dynamics. The default parameters of ChiRotor were used [49] for optimizing both protein side-chain conformations. Energies of the protein models were further minimized using CHARMm [50].
To prepare PLP-D-alanine (PDD) and PLP-D-Lys (PLY) models, D-Ala and D-Lys coordinates were extracted from the BsAlr (PDB code: 1L6G) [13] and E. coli D-lysine-diaminopimelate decarboxylase-ligand complexes (PDB code: 1KO0), respectively. LibDock [51] was used in conjunction with its default settings to generate conformations for ligand-containing Lyr and Bar. Hot spots were then aligned to select docked poses, which were then in situ minimized by Discovery Studio 3.0 ''Conjugate Gradient'' [52] to improve the quality of the conformations. All poses were scored, and the top 100-scoring poses were retained for subsequent analyses. Approximately 28% of the poses were retained from the generated 255 poses for each model (Bar: Bar-PDD, Bar-PLY; Lyr: Lyr-PDD, Lyr-PLY). Two additional scoring functions, Potential of Mean Force [53] and LigScore2 scoring [54,55], were used to compensate for an unfair penalty imposed on the binding-affinity measurement, which resulted because hydrogen atoms were eliminated during LibDock docking. The consensus score for the docked ligands in each protein-ligand model was calculated. Then, CDOCKER [56] with CHARMm forcefield was used to further refine the docked models.
The phosphate moiety in each cognate ligand (PDD or PLY) was fixed in place because its position was superimposable with those in the Alr crystal structures that were complexed with PLP (PDB codes: 1SFT [9], 1BD0 [10], 1L6G [13], 1VFS [12]) as a consequence of an extensive hydrogen-bonding network. When the RMSD of the phosphate in a docked model deviated by #1 Å with that of the reference pose (Table S2, S3, S4, S5), the docked model was selected for further study. According to the established mechanism of BsAlr (PDB code: 1L6G [13]), both catalysts (Lys 39 and Tyr 2659 ) were close to the Ca atom of the alanine substrate within hydrogen bonding distances and responsible for the racemization reaction. Hence, the distances between the atom Ca of docking ligands and reactive atoms of two conserved catalyst (Lys:N and Tyr':O) of Lyr and Bar will be constrained to the extensive hydrogen bonding distance (,4 Å ) (Table S22S5).
To estimate the practicability of the proposed docking procedure, we attempted to self-dock the co-crystal structures complexed with PLP-D-alanine (PDB code: 1L6G [13]) and PLP-D-lysine (PDB code: 1KO0). Top five docked poses with minimum RMSD #1 Å were derived. On average, the RMSD of self-docked poses for PLP-D-alanine and PLP-D-lysine are 0.65 Å and 0.84 Å respectively ( Figure S4). This suggests that the proposed docking approach is viable.

Supporting Information
Figure S1 Related to Figure 5: Top-view and side-view of the binding cavities of Lyr, Bar, and Alrs. The views are those obtained by superpositioning the structures of Lyr (green), Bar (magenta), SpAlr (yellow) and BsAlr (cyan). Crucial residues at the substrate entryway are displayed as stick models. PLP-lysine of Lyr, Bar, and SpAlr as well as PLP-D-alanine plus K39 of BsAlr, respectively, are drawn as ball-and-stick models. Spheres of the conserved tyrosine catalysts are colored as magenta. Spheres of positively charged, negatively charged, polar, and non-polar residues are colored as blue, red, pink, and yellow, respectively. Oxygen, nitrogen, and phosphate atoms are colored red, blue, and orange, respectively. (TIF)