Mechanistic Insights into the Isochorismate Pyruvate Lyase Activity of the Catalytically Promiscuous PchB from Combinatorial Mutagenesis and Selection*

PchB from Pseudomonas aeruginosa possesses isochorismate pyruvate lyase (IPL) and weak chorismate mutase (CM) activity. Homology modeling based on a structurally characterized CM, coupled with randomization of presumed key active site residues (Arg54, Glu90, Gln91) and in vivo selection for CM activity, was used to derive mechanistic insights into the IPL activity of PchB. Mutation of Arg54 was incompatible with viability, and the CM and IPL activities of an engineered R54K variant were reduced 1,000-fold each. The observation that position 90 was tolerant to substitution but position 91 was essentially confined to Gln or Glu in functional variants rules out involvement of Glu90 in general base catalysis. Counter to the generally accepted mechanistic hypothesis for pyruvate lyases, we propose for PchB a rare [1,5]-sigmatropic reaction mechanism that invokes electrostatic catalysis in analogy to the [3,3]-pericyclic rearrangement of chorismate in CMs. A common catalytic principle for both PchB functions is also supported by the covariance of the catalytic parameters for the CM and IPL activities and the shared functional requirement for a protonated Glu91 in Q91E variants. The experiments demonstrate that focusing directed evolution strategies on the readily accessible surrogate activity of an enzyme can provide valuable insights into the mechanism of the primary reaction.

Isochorismate pyruvate lyase (IPL) 2 catalyzes the conversion of isochorismate to salicylate and pyruvate (Fig. 1). This reaction is the committed step in the biosynthesis of salicylate-based siderophores in several pathogenic bacteria (1)(2)(3)(4)(5)(6). In plants, salicylate produced from isochorismate is important for the plant defense mechanisms known as local and systemic acquired resistance (7,8). The catalytic mechanism of the IPL transformation is unknown, but it is formally related to the elimination of pyruvate from other shikimic acid metabolites, including chorismate, 2-amino-2-deoxyisochorismate, and 4-amino-4-deoxychorismate, catalyzed by chorismate lyase, anthranilate synthase, and 4-amino-4-deoxychorismate lyase, respectively (9). It has been proposed that proton abstraction by an enzymic general base initiates pyruvate elimination and aromatization in all of these pyruvate lyases (9 -12).
The best characterized IPL is PchB from the opportunistic human pathogen Pseudomonas aeruginosa (13). Its gene pchB is part of an operon for the biosynthesis of the siderophore pyochelin (4,14,15). The sequence of PchB is unrelated to the pyruvate lyases mentioned above but instead resembles well characterized chorismate mutases (CMs) of the AroQ class (16,17), which catalyze the [3,3]-sigmatropic Claisen rearrangement of chorismate to prephenate (Fig. 1). PchB has the typical length of an AroQ domain with 20% overall sequence identity. Secondary structure prediction indicated that PchB has the same arrangement of ␣-helices as in EcCM (13), the CM domain (AroQ p ) of the bifunctional Escherichia coli chorismate mutase-prephenate dehydratase, for which a crystal structure is available (18). The quaternary structure of PchB was found to be dimeric as it is for EcCM, and most catalytic residues of EcCM are conserved in PchB. Moreover, it was shown that pchB complements the CM deficiency of an E. coli mutant strain and that PchB has low CM activity in vitro (13).
Based on the structural and functional similarity with EcCM, a homology model was constructed for the presumed single active site of the catalytically promiscuous PchB (13). However, this model does not provide for a general base to abstract the C2 proton of isochorismate to initiate pyruvate elimination. Because the sequence alignment between PchB and AroQ CM domains is ambiguous at the C terminus, we have considered an alternative alignment involving multiple sequences from the AroQ family. The new alignment would position Glu 90 in the vicinity of the C2 proton of the substrate, where it could serve as a general base, as required for the previously proposed general base mechanism (9). Here, we report combinatorial mutagenesis and selection experiments with the surrogate CM activity of PchB (19 -25) to test the predictions of the two models. Surprisingly, our results suggest that the IPL mechanism of PchB is distinct from that proposed for other pyruvate lyases.
ing using the BigDye terminator cycle sequencing kit was performed on an ABI PRISM 310 Genetic Analyzer (Applied Biosystems, Rotkreuz, Switzerland). All other chemicals were from Fluka/Sigma-Aldrich (Buchs, Switzerland).
For the library PCR, 10 ng of template DNA, 50 pmol of primers, 0.2 mM dNTPs, and 2.5 units of HotStarTaq DNA polymerase in 50 l of PCR buffer (Qiagen) were used. The reaction mixture was incubated at 95°C for 15 min, followed by 25 PCR cycles (1 min at 95°C, 1 min at 51°C, 1 min at 72°C) and a final extension step (7 min at 72°C). The purified PCR products were digested with NheI and SacII. Purified insert (164 bp, 0.2 pmol) and acceptor fragments (4,124 bp, 0.2 pmol) were ligated for 16 h at 15°C. The ligation products were desalted with the Jetquick PCR Purification Kit (Genomed) and eluted in 2 mM Tris-HCl, 0.2 mM EDTA (pH 8.0). One quarter of the ligated plasmids were electroporated into Phe and Tyr auxotrophic KA12/pKIMP-UAUC cells (20).
Single colonies growing on selective plates were restreaked onto LB Amp 150 Cam 30 agar plates (containing 150 g/ml sodium ampicillin and 30 g/ml chloramphenicol). Plasmid DNA from purified clones was isolated and sequenced with primer 04-T7TR (23). The complementation ability of representative selected library members was verified by retransforming the corresponding plasmids into KA12/pKIMP-UAUC and testing transformants on M9c/P ϩ F and M9c/P ϩ FY minimal media plates at 34°C. Transformants with pKIL10A-W and pKIL10A-0 were used as positive and negative controls, respectively.
Growth Rates in Liquid Minimal Media-Growth rates were determined in liquid M9c/P ϩ F and M9c/P ϩ FY media (modified by omitting FeSO 4 and increasing concentrations of sodium ampicillin to 150 g/ml and isopropyl ␤-D-thiogalactopyranoside to 100 M). For each mutant, a 5-ml M9c/P ϩ FY preculture was inoculated with a single colony and grown overnight at 30°C and 220 rpm. 40 l of this preculture were diluted 1:100 into selective M9c/P ϩ F medium (two parallel 4-ml cultures) and into non-selective M9c/P ϩ FY medium (positive control). The cultures were again incubated at 30°C and 220 rpm, and the OD 600 nm was directly monitored in the test tubes using a Spectronic Helios Beta UV-Vis spectrophotometer equipped with a test tube holder (Thermo Electron Corp., West Palm Beach, FL). The exponential growth phase was fitted to an exponential equation to deduce growth rates defined as: ϭ ⌬ln(OD 600 nm )/⌬t.
Gene Expression and Protein Purification-T7 RNA polymerasedriven pchB expression in E. coli strain KA13 (at 30°C) and protein purification by nickel-nitrilotriacetic acid affinity chromatography were performed following published protocols (23). All proteins examined in this work had the PchB H format. Protein purity was verified by SDS-PAGE using the Phast System (Amersham Biosciences, Otelfingen, Switzerland). Protein concentrations were determined by UV spectroscopy using molar extinction coefficients derived from the amino acid sequence using the Wisconsin Package Version 10.3 (Accelrys Inc., San Diego, CA), which gave results similar (within 10%) to those obtained with the Micro BCA protein assay reagent kit (Pierce) using bovine serum albumin as the standard.
Circular Dichroism Spectroscopy-CD spectra were measured with an Aviv CD spectrometer model 202 (Aviv Biomedical, Lakewood, NJ) at 25°C in 20 mM potassium phosphate buffer (pH 7.5) with 4 M protein in a cuvette with a path length of 0.2 cm. Spectra were recorded five times for each sample from 260 to 195 nm and averaged. The step size and the bandwidth were 1 nm, and the averaging time at each wavelength was 1 s.
Chorismate Mutase and Isochorismate Pyruvate Lyase Assays-Chorismate was produced using E. coli strain KA12 (36) and isochorismate was isolated from K. pneumoniae 62-1/pKS3-02 (28). Steady-state parameters (k cat and K m ) were derived from the initial rates of substrate disappearance. For the CM reaction, the absorbance change at 274 nm (⌬⑀ 274 ϭ 2,630 M Ϫ1 cm Ϫ1 ) or 310 nm (⌬⑀ 310 ϭ 370 M Ϫ1 cm Ϫ1 ) was followed using an established protocol (37). For the IPL reaction, an analogous on-line UV spectroscopic assay was established monitoring the absorbance decrease at 278 nm (⌬⑀ 278 ϭ 6,640 M Ϫ1 cm Ϫ1 ) as a function of time between isochorismate (⑀ 278 ϭ 8,300 M Ϫ1 cm Ϫ1 ; Ref. 38) and the products of the conversion to salicylate (⑀ 278 ϭ 1,630 M Ϫ1 cm Ϫ1 ) and pyruvate (⑀ 278 ϭ 27 M Ϫ1 cm Ϫ1 ). Appropriate corrections for background reaction without enzyme were made. Substrate purity was assessed by determining the difference in absorption before and after full enzymatic conversion of a substrate sample. Substrate concentrations were varied between 7 M and 2.3 mM (chorismate) or between 0.3 and 100 M (isochorismate). Standard kinetics were measured in 50 mM potassium phosphate buffer (pH 7.5) at 30°C. For pH variation studies, universal buffer (as in Ref. 17 but without NaCl) was used between pH 4.9 and 9.0.
Enzyme Inhibition Studies-Two inhibitors, the endo-oxabicyclic diacid 1 first described by Bartlett (39) as a potent transition state analog of the CM reaction and adamantane-1-phosphonate 2 known to inhibit (some) CMs (40), were probed for inhibition of both PchB activities (for the structures of 1 and 2, see Fig. 6A). Substrate concentrations were varied between 10 and 400 M (for chorismate) and between 0.2 and 20 M (for isochorismate) at fixed inhibitor concentrations, ranging from 0 to 12 M for 1 and from 0 to 10 mM for 2. K i values were derived from the replot of the slopes of Lineweaver-Burk reciprocal plots against inhibitor concentration (41).

Sequence Comparison and Homology
Modeling-The primary sequence of PchB has ϳ20% sequence identity with the CMs of the AroQ class (16,17) available in GenBank TM (42). While homology over most of the protein is weak, identities include specific conserved amino acid residues known from structural and mechanistic studies to be important for the function of AroQ CMs (18,25,(43)(44)(45). However, the assignment of certain active site residues is not completely unambiguous. Two slightly different sequence alignments, based on a comparison between PchB and EcCM alone (alignment A) or on a multiple sequence alignment between PchB and other members of the AroQ class (alignment B), are shown in Fig. 2.
Based on the x-ray structure of EcCM complexed with transition state analog inhibitor 1 (18), separate homology models for the active  (18) and from activity studies (17). A, best alignment considering only PchB and EcCM. B, optimized alignment resulting from inclusion of the 19 prokaryotic AroQ class sequences from (17). Note that an additional gap was introduced in front of the Tyr 87 of PchB. As a consequence, Glu 90 rather than Gln 91 corresponds to the catalytically important Gln 88 of EcCM.
site of PchB were built for each sequence alignment. For model building, the relevant active site residues in the EcCM structure were exchanged with the corresponding residues of PchB. The protein backbone was kept rigid, while side chain rotamer conformations were chosen that minimized steric overlap. Of the about 20 amino acids shared by EcCM and PchB (see alignment A or B in Fig. 2), 10 belong to the set of 19 EcCM residues with side chains within 6 Å of any carbon or oxygen atom of bound inhibitor 1, indicating strong conservation of the mechanistic machinery between AroQ CMs and PchB. Fig. 3 shows a simplified schematic representation of both models of the active sites for PchB and of EcCM with bound substrates isochorismate and chorismate, respectively. Because the arrangement of potentially critical active site residues (Tyr 87 , Ile 88 , Glu 90 , and Gln 91 ) differs in the two PchB models, two alternative reaction mechanisms for the IPL reaction can be postulated, depending on the structural model considered. In the past, pyruvate lyase reactions have been formulated as general base-dependent processes (9), involving abstraction of a proton from the substrate (see "Discussion"). In alignment B and model B, the glutamate at position 90 could serve the role as general base in PchB (Figs. 2B and 3B). Subsequent cleavage of the ether bond could be facilitated by polarization of the ether oxygen by Lys 43 lying on the opposite side of the active site. In contrast, the catalytic center in model A lacks such a base. The critical residues Lys 43 and Gln 91 , whose homologs are thought to facilitate the cleavage of the ether bond of chorismate in EcCM (18,44), are present, however. In analogy to the [3,3]-sigmatropic rearrangement of chorismate to prephenate, a reaction also catalyzed by PchB (13), a plausible, albeit uncommon mechanism for model A would be a pericyclic [1,5]sigmatropic rearrangement (Fig. 3). As was found for CM (46), C-O bond breaking may significantly precede formation of the new bond, leading in the extreme case to a two-step dissociative E1 mechanism.
Library Design-To establish which of the models better describes the actual mechanism, we applied directed molecular evolution technology to probe presumed active site residues. Model A predicts that residue 91 is crucial for transition state stabilization, whereas in model B the residue at position 90 would be oriented for proton abstraction. Therefore, two gene libraries (ME1 and ME2) were designed, each having the two codons for the residues at positions 90 and 91 randomized. The ME1 library additionally contained a randomized codon for position 54; the encoded WT arginine is absolutely conserved among all members of the AroQ class (16,17), and codon 54 was included here as a control.

Construction of the Gene Libraries and in Vivo Selection
Experiments-PCR fragments obtained using a pchB fragment as template and long, partially degenerate oligonucleotide primers were cloned into plasmid  (20). Transformants were plated on M9c/P ϩ F minimal medium plates and incubated for 3 days at 34°C to select for functional pchB variants. P sal and P T7 are the promoters used to express pchB for in vivo complementation and for subsequent protein isolation, respectively. The bla gene provides ampicillin resistance to the library plasmids; lacI codes for the Lac repressor regulating the T7 RNA polymerase-controlled gene expression system. pKIL10A-0 that encodes the N-and C-terminal ends of PchB H (Fig. 4). The primer design allowed codons for all 20 natural amino acids and only one stop codon (TAG) at the randomized positions. The resulting gene libraries ME1 and ME2 were each introduced into CM-deficient E. coli cells (KA12/pKIMP-UAUC) and subjected to in vivo selection for CM activity on M9c/P ϩ F minimal plates.
The final ME1 and ME2 libraries contained 6,400 and 17,000 members, respectively, encompassing 18% and 100% of all possible pchB gene variants, or 55% and 100% of all possible PchB protein variants in the respective library. The proportion of complementing clones was 0.3% for the ME1 and 3.2% for the ME2 library. Fourteen mutants from the ME1 library and 47 mutants from the ME2 library were sequenced. The fact that all of the sequenced active ME1 library members coded for Arg 54 experimentally confirms the importance of this residue for CM catalysis.
Furthermore, 55 of the 61 active genes had either glutamine or glutamic acid at position 91 (Fig. 5). Included in this set of 55 are three clones with the amber stop codon (TAG) at position 91, which is partially suppressed by Gln in our selection strain KA12 (supE44) (20,47). The six exceptions coded for Phe 90 -Ser 91 (found twice), Lys 90 -Ser 91 , Asn 90 -Ser 91 , His 90 -Ala 91 , and Lys 90 -Met 91 . However, all of the mutants without Gln 91 or Glu 91 grew only poorly on minimal medium (Fig. 5). In sharp contrast to the conservation patterns found at positions 54 and 91, base triplets encountered at position 90 in active library clones coded for 18 of the 20 possible amino acids.
The set of 61 plasmids encoding the PchB mutants shown in Fig. 5 were retransformed into the selection strain KA12/pKIMP-UAUC and their ability to complement the CM deficiency was confirmed. In addition, growth rates of a subset of interesting mutants were determined in liquid minimal media (TABLE ONE). All 10 mutants tested plus the negative control (pKIL10A-0) were able to grow in supplemented M9c/P ϩ FY medium as fast as the positive controls carrying the gene for WT PchB H (pKIL10A-W) or for EcCM (pKSECM-W), with a growth rate of around 0.36 h Ϫ1 . In contrast, significant differences were FIGURE 5. Encoded residues and growth phenotype of active clones selected from the ME1 and ME2 libraries. Since all active ME1 clones that were sequenced retained an Arg at position 54, the amino acid distribution is only shown for positions 90 and 91. Residues are ordered according to increasing side chain volume (58). Colony size on selective agar plates (phenotype) is rated from 0 (no growth) to 4 (WT complementation). The number at the top of the columns indicates the occurrence of a particular mutant among the 61 sequenced variants. The clones with Glu or Gln (including amber stops) at position 91 are shaded in light gray, whereas the other 6 are shaded in dark gray.  (13). c K m values of mutants 7-17, 8-11, and R54K were too high for accurate determination by the assay used; thus, the corresponding k cat values are estimates. d EcCM was purified as described (17). The catalytic parameters are in good agreement with those previously reported (k cat ϭ 69 s Ϫ1 ; k cat /K m ϭ 230,000 M Ϫ1 s Ϫ1 ) (37). apparent in selective medium M9c/P ϩ F for individual variants (TABLE ONE). Again, fast growth correlated with the identity of the residue at position 91 with mutant Q91N (constructed by site-directed mutagenesis to assess the impact of a single, conservative replacement) being the worst in this series. R54K, another engineered variant, did not grow at all.
In Vitro Characterization of WT PchB H and PchB Variants-The WT enzyme, eight library variants, and the two mutants R54K and Q91N were overproduced and purified to homogeneity (as judged by SDS-PAGE). The secondary structure of WT PchB H and of all purified mutants was assessed by CD spectroscopy. All purified proteins showed minima at 208 and 222 nm typical for ␣-helical proteins (48) (data not shown), in agreement with the secondary structure prediction for WT PchB (13).
Using on-line spectroscopic assays rather than the previously employed stop assays (13), we found that WT PchB H catalyzes the IPL reaction with k cat ϭ 1.06 s Ϫ1 and K m ϭ 0.95 M and the CM reaction with k cat ϭ 0.085 s Ϫ1 and K m ϭ 38 M (TABLE ONE). The effect on WT PchB H activity of two standard CM transition state analog inhibitors (49), the endo-oxabicyclic diacid 1 and adamantane-1-phosphonate 2, was also assessed (Fig. 6). While both the CM and the IPL activity of WT PchB H were competitively inhibited by 1 with K i values of 2.3 and 2.2 M, respectively, 2 was found to be a very weak competitive inhibitor with a K i for each activity above 5 mM (data not shown). It should be noted that the K i values obtained here for transition state analog 1 are 10-fold lower than those reported previously (13) but well within the range (1-4 M) of the corresponding values for other CMs (29, 44, 50 -52).
CM and IPL activities were detectable for all of the mutants, including those with Ser 91  and Ala 91 (SA-55) but excluding Lys 90 -Met 91 (very poor protein yield). Even the R54K variant was active, although this mutation decreased k cat /K m by 3 orders of magnitude (TABLE ONE). Regarding CM activity, all variants with either Gln or Glu at position 91 (excluding R54K) had a k cat /K m of at least 20% of the WT value. In contrast, all tested variants lacking Gln 91 or Glu 91 were below that level. Only variants with the WT Gln 91 retained high IPL activity (k cat /K m within a factor of four of the WT), while the other variants had an IPL activity at least 20-fold below that of the WT enzyme. These findings underscore the importance of the residue at position 91 (as opposed to 90) for both the CM and the IPL reactions catalyzed by PchB.
The differential impact of mutations on IPL and CM activities is seen from the ratios of the respective k cat [(IPL/CM) V ] and k cat /K m [(IPL/ CM) V/K ] values (TABLE ONE). For each variant, the IPL reaction remained dominant over the corresponding CM activity. While (IPL/ CM) V for all variants fluctuated by less than a factor of six around the WT ratio, the two PchB mutants containing the Q91E substitution showed a drop in (IPL/CM) V/K of about 30-fold.
pH Dependence-Natural CMs with a glutamate residue at the position corresponding to the active site residue Gln 91 of PchB in model A (Fig. 3A) have their optimum catalytic efficiency at low pH (29,45). If model A is correct, as suggested from the conservation patterns in the directed evolution experiment, it can be predicted that the ionizability of the residue at position 91 should determine the pH rate profile of PchB. In accord with this prediction, WT PchB H shows a broad pH optimum with little pH dependence of k cat and K m in the range of pH 5-9 for both the CM and IPL activities (Fig. 7). However, a Q91E exchange, as occurs in the mutants 7-17 (Glu 90 -Gln 91 ) and 8-11 (Tyr 90 -Glu 91 ), causes catalytic efficiency (k cat /K m ) to become strongly pH-dependent. In fact, both the CM and the IPL reactions are accelerated by about 3 orders of magnitude as pH is decreased from 9 to 5 (Fig. 7). While k cat is unaffected by the change in pH, a pronounced shift in the K m values for chorismate and isochorismate is observed for the two mutants. Upon decreasing the pH from about 8 to 5, the K m values drop by roughly two orders of magnitude. Analogous results were reported for the corresponding Q88E variant of EcCM (43,44).

DISCUSSION
The IPL reaction catalyzed by PchB results in the elimination of pyruvate from isochorismate to yield the aromatic compound salicylate (Fig.  1). Analogous reactions involving isochorismate or chorismate, the alternative PchB substrate, or their amino derivatives are promoted by at least four different enzymes that have no homology to the AroQ fold (9,53). These enzymes include salicylate synthase (5,53) and the structurally diverse but well characterized anthranilate synthase (TrpE; 2amino-2-deoxyisochorismate lyase step) (11), chorismate lyase (UbiC) (10), and 4-amino-4-deoxychorismate lyase (PabC) (12). A mechanism involving initial general base-catalyzed deprotonation of the substrate followed by cis-elimination of pyruvate was suggested 15 years ago for the aromatization step for the latter three enzymes (9). Moreover, in light of the structural details now available, His 398 has been proposed to initiate C2 proton abstraction in TrpE (11), whereas Glu 155 effects de- protonation of C4 in UbiC (10). In PabC, which contains a tightly bound pyridoxal phosphate cofactor, Lys 159 is believed to deprotonate the Schiff base formed between the substrate 4-amino-4-deoxychorismate and pyridoxal phosphate at C4 (12).
From sequence alignment B and homology modeling with EcCM ( Figs. 2 and 3), Glu 90 is the best candidate for a general base in PchB. However, the fact that active variants with 18 of the 20 natural amino acids were found at position 90 (Fig. 5) and that the WT enzyme is pH insensitive allows us to rule out this possibility. Furthermore, the strict requirement for glutamine or (protonated) glutamic acid at position 91 for high CM activity strongly supports model A. In the latter, there is no enzymic residue in proximity to C2 of isochorismate which could serve as the general base for initiating the IPL reaction.
Instead of general base catalysis, we propose that PchB residues Lys 43 and Gln 91 facilitate pyruvate elimination and aromatization by stabilizing developing negative charge on the ether oxygen of the substrate in a polarized transition state. Such a charge would develop if the C-O bond is fully cleaved to give an ion pair, followed by loss of proton and aromatization (E1 mechanism). Alternatively, and in analogy to the [3,3]sigmatropic rearrangement carried out by CMs, PchB may employ a [1,5]-pericyclic process (Fig. 3, A and C). As in CM catalysis (20,21,44,46,54), the IPL reaction may proceed as a concerted but asynchronous rearrangement with initial C-O bond breakage preceding C-H bond formation to release pyruvate from isochorismate. There is precedent for an analogous [1,5]-pericyclic elimination of acetaldehyde from 3-(methoxycarbonyl)-5-ethenoxy-1,3-cyclohexadiene (55) and of pyruvate from chorismate to give 4-hydroxybenzoate (56).
Such a rare [1,5]-pericyclic mechanism draws support from mechanistic parallels to the CM reaction. First, PchB uses the same active site for its IPL and CM functions (13), as shown, for instance, by identical competitive inhibition of both activities by two standard CM inhibitors (Fig. 6), indicating binding of both substrates in similarly folded conformations (Fig. 3). In sequence alignment A, Lys 43 and Gln 91 correspond to Lys 39 and Gln 88 of EcCM, which are believed to stabilize developing negative charge in the polar transition state of the CM reaction (44). It is plausible that this catalytic feature in the active site of PchB is exploited for transforming both chorismate and the structurally similar isochorismate. Second, mutations of the catalytic site residues Gln 91 or Arg 54 presumed to be important for CM activity have roughly a parallel impact on both PchB functions (TABLE ONE), arguing for a common catalytic machinery. Third, the pH profiles are the same for both PchB activities in the WT on the one hand and the two tested Q91E variants on the other hand (Fig. 7), showing that the protonation state of the residue at position 91 is similarly critical for both reactions.
While our results clearly demonstrate that Arg 54 cannot even be conservatively replaced (by Lys) without losing a factor of 10 3 in k cat /K m for each of the PchB reactions, several alternatives exist for Gln 91 . Interestingly, mutants with Glu 91 (found with about half the frequency of Gln 91 clones on selection plates) showed higher k cat values than the WT with its isosteric Gln 91 . Similar observations were made in comparisons of corresponding Glu and Gln variant pairs of AroQ class CMs from E. coli (43,44) and Saccharomyces cerevisiae (45), and it was speculated that transient protonation of the ether oxygen of chorismate, possible only with the protonated Glu, might bring about improved stabilization of the CM transition state (45). The fact that k cat for both the CM and IPL activities benefits from the Q91E exchange argues in favor of a similar stabilizing effect and thus catalytic mechanism. Another consequence of the Q91E replacement is a pronounced pH dependence of K m . To the extent that K m reflects substrate binding, substrate affinity is highest at acidic pH (Fig. 7). This mirrors the behavior of other AroQ CMs having a Glu at the homologous position (29,43,44) but now also includes the IPL activity. An elevated K m at higher pH values in Q91E variants can be explained by postulating that productive substrate binding can only occur in active sites where Glu 91 is protonated.
Since differential enzyme concentrations arising from variations in gene expression, protein folding and stability will affect the growth rate, only an imperfect correlation between in vivo phenotype and in vitro activities can be expected. Nevertheless, the six slowly growing variants among the 61 initially picked clones from the selection plates, which had either Ser, Ala, or Met at position 91, also exhibited poor activity in vitro. In the absence of detailed structural information, explanations for the residual activity observed for these exceptions remain speculative. Conceivably, a water molecule might replace the missing interaction between Gln/Glu 91 and the ether oxygen of the substrate in the transition state. Alternatively, the choice of residue at the adjacent position may compensate for the loss of an H-bond donor at position 91 (note that position 90 has a positive charge in 3 of the 6 clones).
While the mutations generally affected both IPL and CM activities in similar ways, there are several notable trends and exceptions (TABLE  ONE). The observed differences can be explained in part by the distinct structural demands of isochorismate and chorismate binding to the same active site and the geometric and electronic differences in the transition states for the two PchB reactions. For instance, a Q91E exchange results in a much more dramatic loss in catalytic efficiency (k cat /K m ) for the IPL than for the CM reaction, which could result predominantly from more severely perturbed binding of isochorismate due to the proximity of its 2-OH group to position 91 (Fig. 3A). The trend to reduced (IPL/CM) V/K ratios upon replacing Gln 91 may also reflect the fact that there is more to lose by interfering with the evolutionarily optimized alignment of the charge stabilizing group at position 91 in the case of the IPL activity compared with the CM activity, which is probably a sluggish evolutionary relict (13).
The application of directed evolution for investigation and design of enzymes is often limited by the unavailability of a suitable genetic selection system (25). The evolutionary approach chosen here to unravel the reaction mechanism of IPL circumvented this problem by employing selection for CM activity, a minor promiscuous activity of the enzyme (57). More generally, if an enzyme to be studied shows (sequence) similarity to a catalyst that is amenable to genetic selection, it may be possible to set up a directed evolution scheme by exploiting substrate or catalytic promiscuity.