A modified magnesium shock procedure was used to separate periplasmic and cytoplasmic fractions of P. aeruginosa cultures grown in minimal medium. The established periplasmic enzyme *PheC [13,16] was used as a periplasmic protein marker and shikimate dehydrogenase was used as a cytoplasmic protein marker. It was preferable to use prephenate rather than L-arogenate as a substrate in the assay of *PheC activity, not only because the assay for L-arogenate dehydratase is tedious [17] but also because prephenate is much less expensive. However, *PheC cannot be monitored as prephenate dehydratase in the presence of cytoplasmic P-protein because the PheA domain of that protein is prephenate dehydratase. To avoid this problem, the aroQ•pheA-deficient mutant PAT1051 was used. All the prephenate dehydratase activity in PAT1051 is derived from *PheC, which thus can be used as a convenient marker for the quality of the separation of the cellular compartments.

*PheC can also be selectively assayed without interference from PheA in wild-type PAO1. In the wild-type, the prephenate dehydratase activity of the P-protein is completely inhibited by phenylalanine, whereas *PheC is completely insensitive. Thus, one can calculate the fraction of total activity contributed by each of the two enzymes by comparing activities measured in the presence and absence of L-phenylalanine. From Table 3, a calculated 91% of *PheC was found in the periplasmic fraction of PAT1051, compared to 93% of *PheC from PAO1.

*AroQ was found in the periplasm in P. aeruginosa PAO1. None of the periplasmic chorismate mutase activity was attributed to AroQ•PheA because no PheA-domain activity was detected (Table 3). Hence, the chorismate mutase activity in the periplasm must be exclusively *AroQ. Identification of this activity as *AroQ, and not AroQ•PheA, was verified by elution of a single peak of chorismate mutase in the wash fraction recovered after Mono Q chromatography of the periplasmic fraction. This is the established elution behavior of *AroQ [18]. Of the total aromatic aminotransferase activity, 10–20% was found in the periplasm.

*AroQ was purified from the periplasmic fraction of PAO1. Ninety-seven percent of shikimate dehydrogenase was in the cytoplasm of PAO1, compared to 93% from PAT1051. These data affirm a technically good separation of the periplasmic and cytoplasmic fractions. The purification steps are summarized in Table 4. After the last purification step, one distinctive band with a relative molecular mass (M_r) of around 22,000 was revealed by SDS-PAGE after staining with Coomassie blue. The M_r estimated by gel filtration was around 38,000, suggesting a native homodimer structure. Amino-terminal amino-acid sequencing gave a sequence QPLSPALQQL (in the single-letter amino-acid notation), which corresponds to a sequence beginning at residue 24 of gene PA5184 of the completed P. aeruginosa genome [19].

Aromatic aminotransferase (*Aat_Iβ) was purified from the periplasmic fraction of PAO1. The protein fractions with enzymatic activity that eluted from the gradient of a hydroxylapatite column were chromatographed on a mono Q column and recovered as single peaks. The leading peak fractions yielded one major band and several minor bands after SDS-PAGE. Amino-terminal amino-acid sequencing of the major band gave GSGATSAAGK, which corresponds to the sequence beginning with residue 29 of the deduced protein encoded by PA2531 in the completed P. aeruginosa genome [19]. In addition, one of the minor bands exhibited the sequence AGKPQETFLD, which corresponds to the sequence of the protein encoded by gene PA2531 beginning at residue 36. Thus, a favored and a minor cleavage site seem to exist for the signal peptide. The M_r estimated by SDS-PAGE staining with Coomassie blue was about 37,000. The M_r estimated by gel filtration was 73,000, indicating the presence of a homodimer.

E. coli BL21(pJB-1) containing *aroQ from S. typhimurium was grown to late-exponential growth in 1 liter of LB medium containing kanamycin. After cell harvest by centrifugation, 1.5 g of wet cells were subjected to osmotic shock [20]. After centrifugation, the supernatant was collected and designated as the shocked periplasmic fraction. An equal portion of wet cells, treated identically but not subjected to osmotic shock, was designated as the non-shock periplasmic fraction. Crude extracts were also prepared from each cell-pellet fraction.

The shocked periplasmic fraction contained 26,100 units (1 unit = 1 nmol of prephenate produced per minute) of chorismate mutase activity. As the prephenate dehydratase and prephenate dehydrogenase activities associated with the cytoplasmic AroQ•PheA and AroQ•TyrA proteins were not detected in this fraction, contamination with cytoplasmic species of AroQ was less than 1%. The additional assay for shikimate dehydrogenase as a marker for cytoplasmic contamination also showed less than 1% of total activity to be in the shocked periplasmic fraction. On the other hand, 97% of the total alkaline phosphatase (periplasmic marker) was recovered in the shocked periplasmic fraction. Although cross-reactivity was relatively poor, the presence of S. typhimurium *AroQ in periplasmic fractions and its absence in cytoplasmic fractions was demonstrated by western blotting using antibody obtained against E. herbicola AroQ [12].

The S. typhimurium *AroQ was purified using the protocol of Xia et al. [12] to give a final preparation having a specific activity of 59.1 units/mg, and an estimated 184-fold purification with 19% yield. SDS-PAGE showed a major band and one minor band. The sequence of the ten amino acids at the amino terminus of the major band (DSVTSVSLGA) matched perfectly with the deduced amino-acid sequence, beginning at the 21st residue, consistent with the predicted cleavage indicated in Figure 2.

Purified enzymes from P. aeruginosa, S. typhimurium and E. herbicola were compared with respect to kinetic parameters under identical assay conditions (Table 5). Each enzyme followed Michaelis-Menten kinetics. Parameters of K_m and k_cat were similar to one another, as well as to other chorismate mutases characterized in the literature. Table 5 also shows a comparison of isoelectric points and molecular masses for all of the periplasmic chorismate mutases.

When the periplasmic chorismate mutase from E. herbicola was initially cloned and sequenced, it was a unique case of an AroQ protein having a cleavable signal peptide and a carboxy-terminal extension. The current repertoire of additional AroQ proteins that can be gleaned from the contemporary databases on the criterion of these two features numbers eight. One of them has been located in the esophageal glands of the phytoparasitic nematode Meloidogyne javanica. Presumably, this AroQ protein is secreted into plant cells as part of the parasitic process [21]. The erratic phylogenetic distribution of *AroQ may have a variety of explanations that include massive gene loss, horizontal transfer or independent evolution of a signal peptide and carboxy-terminal extension.

Figure 2 shows a multiple sequence alignment which highlights conserved amino-acid residues in the *AroQ subfamily in relationship to the homologous residues that have been shown by X-ray crystallography [6] to be located in the active site of the AroQ domain of E. coli AroQ•PheA. *AroQ proteins are twice the size of the basic AroQ catalytic domain. The carboxy-terminal half of each *AroQ protein is of similar length but not highly conserved, and it has no known catalytic or regulatory function. When used as a query sequence, it returns no significant hits. One reasonable speculation is that as *AroQ maybe localized at the polar cap(s) of the cell, this region might accomplish some sort of protein-protein or protein-membrane interaction that might facilitate its spatial location at the polar regions of the periplasmic milieu.

Chorismate mutase catalyzes a facile rearrangement reaction that can progress to a significant extent under nonenzymatic conditions. The aroQ genes are easily missed during genome annotation because they are small and do not show strong sequence conservation. We have identified many initially dubious AroQ hits to BLAST queries by alignment of amino-acid residues known to be critical for catalysis.

Figure 3 illustrates several chemical rearrangements that closely parallel the chorismate mutase reaction. Zamir et al. [4] have demonstrated the isochorismate mutase reaction in higher plants, and Xie et al. [22] presented a rationale in support of the possibility that PchB from P. aeruginosa catalyzes the isochorismate mutase reaction as a step in salicylate biosynthesis [23]. 4-Amino 4-deoxy-chorismate mutase catalyzes a step in antibiotic (pristinamycin) biosynthesis [5].

There is precedent for the finding that within a homolog grouping whose members have differing substrate specificities, proteins sharing a substrate specificity do not necessarily cluster together; for example, B. subtilis AroA and close relatives are clustered more specifically with KdsA sequences than with E. coli AroA and its close relatives [24]. The phylogram tree at the left in Figure 3 illustrates this for the AroQ protein family. The current limitations of homology detection are such that the monofunctional AroQ proteins from Aeropyrum pernix, Enterococcus faecalis, Helicobacter pylori and Methanococcus jannaschii cannot be shown to be closer to one another or to any other AroQ proteins/domains than they are to PapB or PchB. This is because bootstrap values only support the branching nodes to the level shown in bold. Thus, even though the computational program used placed A. pernix AroQ with the valid grouping of Streptomyces coelicolor AroQ, C. difficile AroQ, and Mycobacterium tuberculosis AroQ, the A. pernix AroQ sequence could in fact be closer to any other group. Thus, one can recognize distinct and valid groupings (for example, the *AroQ periplasmic chorismate mutases), but not necessarily the exact relationship of these groupings with one another.

Figure 4 shows a phylogram tree of the entire set of AroQ protein or AroQ domain sequences available from Bacteria and Archaea. A number of cohesive clusters are evident. One of these is the periplasmic group of monofunctional AroQ proteins. The cytoplasmic monofunctional AroQ proteins include five-member, three-member, and two-member groups in addition to four other sequences isolated so far. AroQ domains that are fused with TyrA form a cohesive homology grouping within enteric bacteria and their closest relatives. These same organisms also possess AroQ•PheA bifunctional proteins whose AroQ domains also form a cohesive homology grouping. Other AroQ•PheA fusions are widespread, and the sequences of their AroQ domains are highly divergent. Only the AroQ•PheA proteins from P. aeruginosa/ P. stutzeri, from Neisseria gonorrhoeae/N. meningitidis and from Archaeoglobus fulgidus/Thermotoga maritima possess AroQ domains that cluster with bootstrap values greater than 50%. Analysis of interdomain linker regions and codon usage suggests that the aroQ and pheA genes have fused independently on many occasions (G.X. and R.A.J., manuscript in preparation). The three cases of AroQ•AroA fusions fall into a single cluster supported by a bootstrap value of 61%.

In the five-member group labeled PchB in Figure 4, the P. aeruginosa protein has been shown to participate in salicylate synthesis and the proposed reaction (catalyzed by isochorismate mutase) has been discussed in the previous section. The P. fluorescens protein undoubtedly has the same role. This substrate specificity does not necessarily apply to the remaining members. In P. aeruginosa, pchA (isochorismate synthase) and pchB are translationally coupled, and cis-expression is obligatory for activity [23]. In Vibrio vulnificus, pchB is closely linked to venB (isochorismatase) [25]. V. cholerae does not exhibit a pchB homolog upstream of the gene encoding isochorismatase (vibF) [26]. Perhaps the clustering of PchB proteins together reflects their interaction with proteins able to bind isochorismate.

It has been known for some time that phenylalanine-pathway redundancy exists in P. aeruginosa. The enzymological basis for this became apparent with the recognition of the presence of an 'extra' species of chorismate mutase (CM-F) and cyclohexadienyl dehydratase in addition to the E. coli-like enzymes [27]. This redundancy explained the "reluctant auxotrophy" of P. aeruginosa [27], as well as the "hidden overflow pathway" to phenylalanine [28]. The latter phenomenon is one whereby early-pathway regulatory mutations that increase intracellular chorismate concentrations result in exclusive metabolite flow toward L-phenylalanine because cyclohexadienyl dehydratase is insensitive to allosteric control. In the periplasmic system a mixture of flow routes (chorismate → prephenate → phenylpyruvate → phenylalanine or chorismate → prephenate → arogenate → phenylalanine) exist because of the ability of cyclohexadienyl dehydratase to accept either prephenate or arogenate and because of the ability of periplasmic aminotransferase (whose detailed characterization is not within the scope of this paper) to transaminate either phenylpyruvate or prephenate.

The spatial location of the overflow pathway in the periplasmic compartment was first indicated by the demonstration that cyclohexadienyl dehydratase had a cleavable amino terminus and was translocated to the periplasm. Indirect evidence favoring a periplasmic location for chorismate mutase as well in P. aeruginosa was the demonstration of a periplasmic chorismate mutase in E. herbicola [12]. The possible periplasmic location of an aromatic aminotransferase has been anticipated, but not directly examined until now. This paper demonstrates the identity and presence of all three enzymes constituting the overflow pathway in a single organism.

On the basis of broad comparative enzymological surveys [18,29], a similar intact overflow pathway can be anticipated to be present in Serratia and Erwinia. The application of queries to the finished and unfinished genomes at the National Center for Biotechnological Information [30] returned hits for periplasmic cyclohexadienyl dehydratase and periplasmic chorismate mutase in Yersinia pestis. Although no aminotransferase gene has yet been found in Y. pestis that has a predicted signal peptide, one is present that is predicted by P-Sort to be an inner-membrane protein.

Before the recognition of their periplasmic location, it was known [31] that some organisms might have *AroQ but not *PheC (Salmonella) or *PheC but not *AroQ (Klebsiella). It was pointed out that appropriate mutants are better producers of prephenate in Salmonella than in Escherichia because of the presence of *AroQ in the former but not in the latter. Detailed comparative enzymological work in enteric bacteria [18,29,32] indicates that many additional *aroQ genes will be identified. Within Enterocluster-1, Enterobacter cloacae resembles S. typhimurium in lacking *pheC but having *aroQ, whereas Enterobacter aerogenes resembles K. pneumoniae in its possession of *pheC but its absence of *aroQ. Other Enterocluster-1 members (Escherichia, Shigella, Enterobacter agglomerans (ATCC 29915), and Citrobacter,) lack both periplasmic enzymes. On the other hand, all members of Enterocluster-2 possess both *aroQ and *pheC, except for Erwinia carotovora which lacks *pheC. These include Serratia marcescens, Serratia rubidae, Enterobacter agglomerans (ATCC 27155), Erwinia herbicola, E. milletiae, E. ananas, E. tracheiphila, E. chrysanthemi and E. amylovora. Members of Enterocluster-3 (species of Cedecea, Kluyvera, Edwardsiella, Hafnia, Yersinia, Proteus, Providencia and Morganella) lack both *aroQ and *pheC.

All four species of Mycobacterium in the finished genomes resemble Salmonella in having *aroQ but lacking *pheC. The physiological function of a partial pathway is unknown; perhaps *AroQ interfaces with a different pathway. In fact, it must be conceded that even the function of the intact periplasmic pathway to phenylalanine is unknown. An important clue presumably is the spatial location at the pole(s) of the cell. The polar, periplasmic location of a number of protein systems is receiving increasing attention [33].

Most aminotransferases that are capable of functioning as aromatic aminotransferases are members of family I within the aminotransferase protein superfamily [3]. These can usually be assigned to one of three subfamilies, Subfamily Iα partitions into narrow-specificity aspartate aminotransferases and broad-specificity aspartate aminotransferases that can accommodate the aromatic ring. Likewise, subfamily Iβ partitions into narrow-specificity imidazole acetol phosphate (IAP) aminotransferases and broad-specificity IAP aminotransferases that utilize aromatic amino/keto acids. Subfamily Iγ partitions into subgroups having a variety of substrate specificities, with aspartate aminotransferases being the most prominent subcluster. Among these are broad-specificity aspartate aminotransferases with ability to function with aromatic amino acids.

The periplasmic aminotransferase of P. aeruginosa belongs to the Iβ subfamily and exhibits catalytic activity as IAP aminotransferase in vitro assayed as described by Gu et al. [34]. It exhibits signature motifs corresponding to the regions of the active-site lysine and the interdomain interface [3]. Like its close relative, HisH of Bacillus subtilis, it has the potential to function in both histidine and aromatic amino-acid biosynthesis. Of course, the spatial location of *Aat_Iβ in the periplasm would seem to rule out in vivo function in histidine biosynthesis. This expectation is accommodated by the observation that P. aeruginosa possesses a paralog encoded by gene PA4447 [19] that must function in vivo as IAP aminotransferase as it is closely linked with other genes of histidine biosynthesis. Yet another paralog is encoded by gene PA3165 and carries out aromatic amino-acid biosynthesis in the cytoplasm. Thus, three subfamily Iβ paralogs of aminotransferase are expressed in P. aeruginosa: one for aromatic biosynthesis in the cytoplasm, one for histidine biosynthesis in the cytoplasm, and one for phenylalanine synthesis in the periplasm. Previous enzymological work had demonstrated the existence of five aminotransferases in P. aeruginosa that were capable of functioning as aromatic aminotransferases in vitro [35].

All bacterial strains and plasmids used in this study are listed in Table 6. S. typhimurium tyrA19 was grown on Luria-Bertani (LB) enriched medium or M-9 minimal media [36] plus growth-limiting L-tyrosine (0.5 μg/ml). E. coli DH5α and E. coli BL21 were grown on LB medium plus kanamycin when transformed with plasmid. Klebsiella pneumoniae 62–1 was grown according to the method of Gibson [37] for accumulation and purification of chorismate as the free acid form (about 91% pure). Prephenate and L-arogenate were obtained as specified by Bonner et al. [31]. Genomic DNA prepared from S. typhimurium was a gift from Thomas Bobik, University of Florida. Where indicated, ampicillin was added to media at 100 μg/ml, kanamycin at 35 μg/ml, and thiamine at 17 μg/ml. Medium was solidified with 1.5% (w/v) agar.

Sequences from the public databases are identified with their gene identification (GI) numbers in Table 7.