Widespread Use of the Glu-tRNA Gln Transamidation Pathway among Bacteria

The expression of the Rhizobium meliloti glutamyl-tRNA synthetase gene in Escherichia coli under the control of a trc promoter results in a toxic effect upon isopropyl- (cid:98) - D -thiogalactopyranoside induction, which is probably caused by a misacylation activity. To further investigate this unexpected result, we looked at the pathway of Gln-tRNA Gln formation in R. meliloti . No glutaminyl-tRNA synthetase activity has been found in R. meliloti crude extract, but we detected a specific ami-notransferase activity that changes Glu-tRNA Gln to Gln-tRNA Gln . Our results show that R. meliloti , a member of the (cid:97) -subdivision of the purple bacteria, is the first Gram-negative bacteria reported to use a transamidation pathway for Gln-tRNA Gln synthesis. A phylogenetic analysis of the contemporary glutamyl-tRNA synthetase and glutaminyl-tRNA synthetase amino acid sequences reveals that a close evolutionary relationship exists between R. meliloti and yeast mitochondrial glutamyl-tRNA synthetases, which is consistent with an origin of mitochondria in the (cid:97) -subdivision of Gram-negative purple bacteria. A 256-amino acid open reading frame closely related to bacterial glutamyl-tRNA synthetases, which probably originates from a glutamyl-tRNA synthetase gene duplication, was found

Aminoacyl-tRNA synthetases (aaRS) 1 catalyze the specific attachment of amino acids to their cognate tRNAs, ensuring the fidelity of the translation process. Even though aaRSs catalyze the same overall reaction, they can be partitioned into two classes by short consensus amino acid sequences shared by all members of the same class (Eriani et al., 1990). This division correlates well with functional and structural properties displayed in each class; for instance, class I enzymes aminoacylate the 2Ј-OH group of the terminal adenosine of tRNA, whereas class II aaRSs use the 3Ј-OH group (except phenylalanyl-tRNA synthetase, which uses the 2Ј-OH group). Moreover, the threedimensional structure of class I aaRSs (Brunie et al., 1990;Rould et al., 1989) shows that their active site is built around a Rossman fold, whereas that of class II members (Cusack et al., 1990;Ruff et al., 1991) is built around an antiparallel ␤-sheet. The evolutionary relationship found between members of the same class of aaRSs is consistent with the division of aaRSs into two classes and reveals that within each class, aaRSs specific for chemically similar amino acids tend to group in subclasses (Nagel and Doolittle, 1991). Moreover, aaRSs specific for the same amino acid from a variety of organisms preferentially group together, reinforcing the hypothesis that the complete set of 20 aaRSs, considered to be very ancient enzymes (Schimmel and Ribas de Pouplana, 1995), evolved long before the divergence leading to eukaryotes and prokaryotes (Nagel and Doolittle, 1991;Brown and Doolittle, 1995). Based on this assumption, a root of the universal tree of life was recently derived from an ancient class I aaRS gene duplication (Brown and Doolittle, 1995).
As a general rule, 20 different aaRSs, one for each amino acid, constitute the minimal set required for protein biosynthesis. The formation of Gln-tRNA Gln stands as an exception to this rule since in Bacillus subtilis, a GluRS aminoacylates tRNA Glu and misacylates tRNA Gln with glutamate (Wilcox and Nirenberg, 1968;Lapointe et al., 1986), and the latter is transformed into Gln-tRNA Gln by a specific amidotransferase (Wilcox and Nirenberg, 1968;Strauch et al., 1988). This pathway of Gln-tRNA Gln formation has also been found in archaebacteria (White and Bayley, 1972;Gupta, 1984), Gram-positive bacteria, cyanobacteria, and organelles of eukaryotes (Schön et al., 1988). In the context of the coevolution theory of the genetic code and of the amino acid biosynthetic pathways (Wong, 1975(Wong, , 1988, we suggested that the direct glutaminylation pathway that uses a GlnRS evolved from the transamidation pathway that uses a misacylating GluRS (Lapointe, 1982).
Relatedness between GluRS and GlnRS has been inferred * This work was supported by Grant OGP0009597 from the Natural Sciences and Engineering Research Council of Canada (to J. L.). The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked "advertisement" in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
‡ first on the basis of a catalytic property since they both require tRNA for amino acid activation (see ) and later on the basis of their sequence similarities (Breton et al., 1986(Breton et al., , 1990, suggesting that they evolved from a common ancestor. More recently, the amino acid sequences from GluRSs of two higher eukaryotes were shown to be more similar to the GlnRS sequences than to the bacterial GluRS sequences (Cerini et al., 1991). According to those results, it was proposed that GlnRS arose in eukaryotes from a gene duplication and was later transferred horizontally to bacteria (Lamour et al., 1994), which is also consistent with the strict distribution of GlnRS in the bacterial kingdom. Based on tRNA discrimination properties of Escherichia coli GluRS and GlnRS (Rogers and Söll, 1993) and on the difference in tRNA Gln aminoacylation pathways between Gram-negative and Gram-positive bacteria, an evolutionary scenario, consistent with the horizontal gene transfer hypothesis (Lamour et al., 1994), has been proposed by Rogers and Söll (1995). In their model, they suggest that the divergence between eukaryotic and bacterial "GlxRS" (an ancestral enzyme able to recognize all tRNA isoacceptors) reflects the process of accommodation to multiple tRNA isoacceptors (with anticodons (c/u)UX for eukaryotes and UUX for bacteria). They also assumed that this process of accommodation appeared after the split between prokaryotes and eukaryotes, but before glutamine was added to the pool of amino acids used in protein synthesis, suggesting that glutamine was not present in the last common ancestor. Recently, the resolution of the crystal structure of Thermus thermophilus GluRS (Nureki et al., 1995) and its comparison with the three-dimensional structure of E. coli GlnRS (Rould et al., 1989) revealed a striking difference in the architecture of their COOH-terminal halves. From these observations, Nureki et al. (1995) suggested that a prototypical enzyme consisting of the class-defining NH 2 -terminal half recruited an all-␣-domain in prokaryotes and an all-␤-domain in eukaryotes for better tRNA recognition, which is consistent with the accommodation hypothesis (Rogers and Söll, 1995). However, it is also suggested that the COOHterminal half might have originated from several ancestors and had evolved independently from the class-defining NH 2 -terminal half, a hypothesis consistent with the view that these two domains of class I synthetases interact with the two major domains of the tRNA molecule (Schimmel et al., 1993;Schimmel and Ribas de Pouplana, 1995).
The exceptional use of two different pathways for formation of Gln-tRNA Gln suggests that in the last common ancestor, the incorporation of glutamine into proteins was not well established, which is consistent with the proposal that glutamine was a late addition to the amino acid repertoire (Wong, 1975). Furthermore, the higher similarity found between GluRSs from higher eukaryotes and all GlnRSs than with bacterial GluRSs requestions the generality of the prevailing hypothesis that aaRSs are more related to those sharing the same amino acid substrate than to the others (Nagel and Doolittle, 1991). To elucidate the significance of the particularities of the GlxRS family in relation to the other class I aaRSs and to the establishment of the genetic code, we performed an enlarged phylogenetic analysis based on the distribution of the transamidation pathway and on amino acid sequence analysis. In particular, we focused on the pathway for Gln-tRNA Gln formation in Rhizobium meliloti, a member of the ␣-subdivision of Gram-negative purple bacteria. According to the proposed endosymbiotic origin of organelles (Gray, 1992), mitochondria originate from the ␣-subdivision of Gram-negative purple bacteria. Since mitochondria lack GlnRS whereas E. coli, a member of the ␥-subdivision of Gram-negative purple bacteria, possesses one, it was relevant to look for the presence of GlnRS in the ␣-subdivision. We examined the evolutionary relationship between eukaryotic and bacterial GluRS and GlnRS using a highly conserved region, the Rossman fold. Our biochemical data and the relative position of R. meliloti and yeast mitochondria in evolutionary trees based on GlxRS sequence alignments are consistent with an ␣ purple bacterial origin of mitochondria. Our phylogenetic analysis also suggests that contemporary GluRS and GlnRS evolved from a common ancestral GluRS that participated in Gln-tRNA Gln biosynthesis together with an amidotransferase, reinforcing the idea that the transamidation pathway is an ancestral trait. This ancestral GluRS diverged to give a GlnRS-like GluRS in the eukaryotic lineage and a GluRS-like GluRS in the bacterial lineage that both participated in a transamidation pathway. This evolutionary scenario supports the hypothesis that the GlnRS activity has appeared in the eukaryotic lineage and was later transferred to few bacteria. However, it is in contrast to the generally accepted hypothesis suggesting that aaRS evolved to include the full set of 20-amino acid specificity before the divergence leading to prokaryotes and eukaryotes.
Aminoacylation Reaction-The GluRS and GlnRS activities were assayed by measuring the rate of formation of [ 14 C]glutamyl-tRNA and [ 14 C]glutaminyl-tRNA, respectively, as described previously . The amount of aminoacyl-tRNA formed was measured in aliquots of 25 l. Unfractionated tRNA and tRNA Glu (1360 pmol/A 260 unit) from E. coli MRE600 were purchased from Boehringer Mannheim. Unfractionated tRNA from R. meliloti was prepared by the method of Zubay (1966).
Amidotransferase Assay-The amidotransferase activity was measured in a single-step reaction as described previously . Enough of the partially purified R. meliloti GluRS needed to charge the tRNA Glu and tRNA Gln contained in 0.3 A 260 units of unfractionated R. meliloti tRNA was used in the transamidation assay mixture. Before adding 10 l of the S-20 extract containing a putative amidotransferase activity, the mixture was incubated for 2 min at 37°C, and an aliquot of 5 l was used to check the level of tRNA glutamylation. After an incubation of 8 more min at 37°C, the aminoacyl-tRNAs were precipitated. The pellet obtained by centrifugation at 12,000 ϫ g for 30 min was dissolved and deacylated in 400 l of 20 mM NaOH for 1.5 h at room temperature. The deacylated tRNA mixture was neutralized with 81 l of 0.1 N HCl and concentrated with butanol to a final volume of 5-10 l. [ 14 C]Glutamate and [ 14 C]glutamine were analyzed by chromatography on cellulose thin-layer plates in isopropyl alcohol/formic acid/water (20:1:5) at room temperature. The chromatograms were dried, cut at each centimeter from the origin to the solvent front, and counted for radioactivity.
Separation of the R. meliloti GluRS and GlnRS Activities-This separation was done essentially by the procedure of Lin et al. (1992) with the following modifications. Frozen R. meliloti A2 cells (2.6 g) were thawed in 10 ml of extraction buffer (10% glycerol, 20 mM ␤-mercaptoethanol, 1 mM phenylmethylsulfonyl fluoride, and 10 mM KPO 4 , pH 8.0). The cells were disrupted by sonication in an ice bath and centrifuged at 12,500 ϫ g for 45 min to remove cell debris. A polyethylene glycol 8000/dextran T500-partition was carried out in extraction buffer containing 7% polyethylene glycol and 1.5% dextran. The two phases were separated by centrifugation at 4000 ϫ g for 20 min. The top phase was diluted four times with extraction buffer and loaded directly at a flow rate of 1 ml/min on a MonoQ HR 10/10 column (Pharmacia Biotech Inc.) equilibrated against extraction buffer. The column was washed with 6 ml of extraction buffer and then with a 70-ml gradient of 0 -0.6 M NaCl. Finally, fractions were assayed for GluRS and GlnRS activities using unfractionated E. coli tRNA.
Expression in E. coli of the R. meliloti gltX Gene Encoding GluRS-The plasmid pTrc99b (Pharmacia Biotech Inc.) (Amann et al., 1988) was used to express the gltX gene of R. meliloti in E. coli. To reconstruct the cloned gltX gene (Laberge et al., 1989), a 2.4-kilobase BamHI fragment encoding the C-terminal part of gltX was inserted into the unique BamHI site of pTrc99b to create pYGM305. To express a full gltX gene, a 800-base pair KpnI-StuI fragment encoding the N-terminal part of gltX was inserted into the KpnI-StuI unique sites of plasmid pYGM305 to obtain plasmid pYGM308. The E. coli thermosensitive strain JP1449 harboring an altered GluRS was transformed with these plasmids to check for complementation at permissive and nonpermissive temperatures with or without addition of 1 mM IPTG. For analysis of total proteins, cells picked from colonies grown on LB agar plates with or without IPTG were resuspended in 150 l of Laemmli sample buffer, boiled for 5 min, and centrifuged for 5 min to remove cell debris. Aliquots of 20 l were analyzed by SDS-polyacrylamide gel electrophoresis (Laemmli, 1970).
Computer Analysis-The programs of the University of Wisconsin Genetics Computer Group (Devereux et al., 1984) were used for nucleotide and amino acid sequence analysis. In particular, the Wordsearch and Tfasta programs were used to scan the GenBank TM (Version 89) Data Bank, and PileUp was used for multiple amino acid sequence alignment. The programs contained in the PHYLIP 3.5 package (Felsenstein, 1988) where used for phylogenetic analyses. All the sequences were obtained either from GenBank TM (Version 89) using Stringsearch and Fetch commands or from the NCBI nonredundant data base using the Tblastn program, except for Giardia lamblia GlnRS (Shields, 1993). 2

RESULTS
Toxicity of R. meliloti GluRS for E. coli-No typical E. coli promoter was found upstream of R. meliloti gltX, and a plasmid containing this gene did not complement the E. coli thermosensitive strain JP1449 altered in GluRS. 3 To express R. meliloti gltX in E. coli, we constructed plasmid pYGM308 (see "Experimental Procedures"), in which this gene is under the control of the trc promoter. We transformed E. coli strain JP1449 with pYGM308 to check for functional complementation and with pYGM305, which contains only the 3Ј-half of gltX, as a negative complementation control. Under permissive conditions (32°C), cells containing pYGM308 grow slowly on LB agar plates, but they stop growing when IPTG is added to the medium. Cells containing pYGM305 grow well with or without IPTG. No complementation was seen in the presence or absence of IPTG at nonpermissive temperature (42°C). This toxic effect is also seen when R. meliloti gltX is expressed in E. coli DH5␣, JM109, and, to a lesser extent, JM101. JM101 cells containing pYGM305 and pYGM308, grown on LB agar plates with or without IPTG, were used to prepare total protein extracts; no bands corresponding to R. meliloti GluRS could be detected when these extracts were analyzed by SDS-polyacrylamide gel electrophoresis (data not shown), indicating that this enzyme is present at a low level in E. coli under these conditions.
Absence of the GlnRS Activity in R. meliloti-To partially purify both R. meliloti GluRS and GlnRS, we used a two-step procedure previously described for E. coli GluRS (Lin et al., 1992). The first step consists of a two-phase polyethylene glycol-dextran partition followed by chromatography on a fast protein liquid chromatography MonoQ column. The R. meliloti GluRS activity was eluted by 0.25 M NaCl as a sharp peak, but no GlnRS activity was detected in the eluted fractions (Fig. 1). Since the activity was measured using a heterologous tRNA (unfractionated E. coli tRNA) and the two-phase partition might be deleterious to a putative R. meliloti GlnRS, we measured the GluRS and GlnRS activities in an S-20 extract using homologous tRNA. No GlnRS activity was detected in this crude extract using either unfractionated E. coli or R. meliloti tRNA, whereas the high level of [ 14 C]glutamate incorporation using either tRNA reveals the presence of a GluRS activity (Fig. 2). Interestingly, the fractions containing the GluRS activity aminoacylated well unfractionated E. coli tRNA, but not E. coli tRNA Glu (Fig. 3), suggesting that a tRNA other than tRNA Glu accepts glutamate from R. meliloti GluRS.
Characterization of Gln-tRNA Gln Formation in an R. meliloti Extract-The absence of a GlnRS activity in an R. meliloti crude extract and the misacylation properties of its GluRS indicated that the transamidation pathway is used for the formation of Gln-tRNA Gln . To further investigate this possibility, we characterized the amidotransferase activity of an R. meliloti crude extract. The amidotransferase assay was per- formed with unfractionated R. meliloti tRNA, which was aminoacylated first with [ 14 C]glutamate in the presence of unlabeled glutamine using the partially purified R. meliloti GluRS. The addition of glutamine, ATP, and an extract containing a putative Glu-tRNA Gln amidotransferase activity to the glutamylated unfractionated R. meliloti tRNA caused a tRNA-dependent conversion of [ 14 C]glutamate to [ 14 C]glutamine (Fig.  4). To eliminate the possibility of a contaminating activity that converts free [ 14 C]glutamate to [ 14 C]glutamine, we analyzed by cellulose thin-layer chromatography the precipitable tRNA fraction that was deacylated before chromatography. As shown in Fig. 4, ϳ35% of the [ 14 C]glutamate was transformed to [ 14 C]glutamine, which is consistent with the presence of two tRNA pools aminoacylated with [ 14 C]glutamate, [ 14 C]Glu-tRNA Glu , and [ 14 C]Glu-tRNA Gln before the transformation of the latter by the amidotransferase.
Comparison of 16 GlxRSs and a Novel Related ORF-When searching the NCBI data base for all GlxRS amino acid sequences, we found a novel ORF (accession number D26562) located at 4 min on the E. coli chromosome. This ORF was reported to be similar to E. coli gltX (Fujita et al., 1994). Interestingly, this ORF encodes a 256-amino acid sequence more similar to the Rossman fold found in GluRSs than to that of the other class I aaRSs. This ORF is located 37 base pair downstream from dskA and 368 base pairs upstream from pcnB on the E. coli chromosome. Northern analyses and S1 analyses of dskA transcripts show that ORF256 is not cotranscribed with this gene (Kang and Craig, 1990). Moreover, no typical E. coli promoter is found upstream of ORF256, and a putative rho-independent terminator for dskA is located near the first codon of ORF256. These observations suggest that ORF256 is probably not normally expressed and is probably not coding for an essential gene in E. coli. We also found downstream from the furA gene of Vibrio vulnificus (GenBank TM VIBFURA) the 3Ј-end of a GlnRS gene; this gene organization between furA and the GlnRS gene is identical to that found in E. coli (Bachmann et al., 1990).
Comparison of the GlxRS amino acid sequences clearly shows that bacterial GluRSs are more related to each other (35-50% identity) than to GlnRS (20 -25% identity). Recently, Lamour et al. (1994) reported that eukaryotic GluRS sequences were more related to the GlnRS sequences (34 -37% identity) than to bacterial GluRS sequences (20 -23% identity). The amino acid sequence comparison of ORF256 with GluRS and GlnRS shows that it is more related to bacterial GluRSs (29 -38% identity) than to eukaryotic GluRSs (20% identity), all GlnRSs (21-23% identity), or any other class I aaRS. We aligned all GluRS and GlnRS sequences with the PileUp program, which uses a progressive alignment algorithm. In the most conserved regions, which lie within the first and second halves of the dinucleotide-binding domain (Rossman fold) of class I aaRSs, near the two signature sequences HIGH and KMSKS (Fig. 5), our alignment is almost identical to the alignment obtained from comparison of the structural model of E. coli GlnRS (Rould et al., 1989) and T. thermophilus GluRS (Nureki et al., 1995). The detailed analysis of the structural model obtained from an E. coli tRNA Gln ⅐GlnRS⅐ATP crystal (Perona et al., 1993) suggests that residues constituting the glutamine pocket in E. coli GlnRS are found near the two signature sequences. As shown in Fig. 5, the residues involved in interactions with parts of the glutamine that are structurally conserved with glutamate (E. coli GlnRS Arg 30 , Pro 32 , Asp 66 , Tyr 211 , and Phe 233 , identified by asterisks in Fig. 5) are identical between eukaryotic GluRSs and GlnRSs, whereas conservative changes or identical residues are usually found in bacterial GluRSs. Systematic changes from GlnRSs to all GluRSs (His 215 3 Cys, Cys 229 3 Arg, and Gln 255 3 Ile, identified by double asterisks) are found in residues that interact with the carboxylamide side chain of glutamine, which is structurally different from the negatively charged carboxylate side chain of glutamate. Moreover, Cys 229 of E. coli GlnRS is changed to an arginine that is highly conserved among eukaryotic and bacterial GluRSs (Fig. 5). These features of our alignment are of high predictive value in discriminating between eukaryotic GluRSs and GlnRSs. Based on these sequence particularities, we identified the yeast cytoplasmic GluRS that was previously identified as a GlnRS (GenBank TM SCC7LORFS).
Phylogenetic Analysis of the GlxRS Rossman Folds-We used in our analysis an amino acid alignment of the GlxRS Rossman folds. 4 Furthermore, analyses were performed on two stretches (Fig. 5) of 64 amino acids (including the HIGH motif) and 82 amino acids (including the KMSKS motif) included in the Rossman fold. Two methods were used for phylogenetic analysis of GluRS and GlnRS alignments. A maximum parsimony analysis (PROTPARS) was performed on the two halves and on the overall alignment of the dinucleotide-binding folds, and the statistical significance of the trees obtained was assessed with the bootstrap test (SEQBOOT) using 500 replicas. Our analysis reveals two major clusters (Fig. 5): the eukaryotic cluster (subdivided into two distinct groups: one for the GluRSs and the other for bacterial and eukaryotic GlnRSs) and the 4 Sequence alignment and pairwise distance comparisons are available upon request to J. L.

FIG. 5. Partial amino acid sequence alignment of the GlxRS Rossman fold region.
The alignment is based on the overall alignment of the Rossman fold using the PileUp program of the Genetics Computer Group package. Boldface letters indicate residues that are identical for at least 12 GlxRSs. White letters in black boxes indicate stretches of amino acids that allow the identification of bacterial GluRSs, or GlnRSs, or eukaryotic GluRSs. White letters in dark gray boxes represent those that are identical within GlnRSs and eukaryotic GluRSs, but different from bacterial GluRS identities. Black letters in light gray boxes represent signature sequences characteristic of Chlamydia, Thermus, Rhizobium, and yeast mitochondrial GluRSs. Asterisks mark the positions of residues implicated in the glutamine-binding site of E. coli GlnRS (Perona et al., 1993); single asterisks mark the positions of residues interacting with regions structurally conserved in glutamine and glutamate, whereas double asterisks represent residues that interact with atoms that differ in glutamine and glutamate. Eukaryotic GluRSs (1), GlnRSs (2), bacterial GluRSs (3), and mitochondrial GluRS (4) are clustered together. Abbreviations for GluRS sequences are as follows: EBst, Bacillus stearothermophilus ; EBsu, B. subtilis ; ECps, Chlamydia psittasci (Wichlan and Hatch, 1993); EDme, Drosophila melanogaster (Cerini et al., 1991); EEco, E. coli (Breton et al., 1986); EHsa, Homo sapiens (Fett and Knippers, 1991); EMho, Mycoplasma hominis (Dybrig and Yu, 1994); ERme, R. meliloti (Laberge et al., 1989); ESce, Saccharomyces cerevisiae cytoplasm (GenBank TM SCC7LORFS); EScm, S. cerevisiae mitochondrion (GenBank TM YSCMSE1G); ESsp, Synechocystis sp. (Zhao et al., 1993); ETth, T. thermophilus (Nureki et al., 1992). Abbreviations for GlnRS sequences are as follows: QEco, E. coli (Yamao et al., 1982); QHsa, H. sapiens (Lamour et al., 1994); QSce, S. cerevisiae cytoplasm (Ludmerer and Schimmel, 1987); QGla, G. lamblia (Shields, 1993). The abbreviation for ORF256 is as follows: EORF, E. coli (Fujita et al., 1994). bacterial cluster (subdivided into two statistically significant groups: one for Bacillus sp. and Mycoplasma GluRSs and the other for Rhizobium, Thermus, and Chlamydia GluRSs). This arrangement has been observed for all the trees constructed and is supported by high bootstrap values (500 replicas). The relative branching order of the two groups of bacterial GluRSs with each other and with E. coli GluRS and ORF256 varied depending on which part of the Rossman fold was used for the tree construction. E. coli and Synechocystis GluRSs usually branch together as a third group among bacterial GluRSs, but this topology is not well supported by bootstrap values. As shown by Lamour et al. (1994), yeast mitochondrial GluRS branches differently depending on whether the HIGH or KM-SKS alignment is used for tree construction, but we found that it branches in the R. meliloti group (Fig. 6) if the entire Rossman fold or a combination of the HIGH and KMSKS alignments (data not shown) was used. Moreover, signature sequences characteristic of the R. meliloti group, which includes Chlamydia and Thermus GluRSs, were also found in yeast mitochondrial GluRS (Fig. 5). ORF256 branches most of the time as an outgroup of the two or three major groups within the bacterial GluRS cluster, but it is also found with the R. meliloti group. We suggest that this ORF is probably a pseudogene and accumulated more mutations than other bacterial gltX genes that are essential for cell viability. A distance matrix was generated for each alignment with the Protdist program using the Dayhoff matrix (Dayhoff et al., 1978). The tree topology obtained using the neighbor-joining method (Saitou and Nei, 1987) of the Neighbor program (data not shown) is mostly similar to that obtained with the maximum parsimony approach (Fig. 6). Moreover, the Kitch program, which assumes that an evolutionary clock is valid, reveals a tree topology (data not shown) almost identical to the tree obtained with the maximum parsimony method (Fig. 6). Although our trees are based on a limited number of genes, they are consistent with the relative branching order found in trees based on 16 S rRNA (Olsen et al., 1994). DISCUSSION No R. meliloti GlnRS activity was eluted from a MonoQ column (Fig. 1) or was detected in an S-20 cell extract (Fig. 2) using either unfractionated E. coli or R. meliloti tRNA, whereas GluRS activity was detected in both cases. Surprisingly, R. meliloti GluRS is unable to ligate glutamate to E. coli tRNA Glu , but is able to ligate it to unfractionated E. coli tRNA (Fig. 3) suggesting that one or several E. coli tRNAs are mischarged with glutamate. Since this misacylating activity of R. meliloti GluRS toward E. coli tRNA probably leads to errors in protein biosynthesis, it explains the in vivo toxic effect observed following expression of the R. meliloti gltX gene in E. coli. These characteristics of R. meliloti GluRS are strikingly similar to those observed for B. subtilis GluRS since it does not glutamylate E. coli tRNA Glu , but does glutamylate efficiently E. coli tRNA 1 Gln in vitro . The toxicity of B. subtilis GluRS for E. coli was suggested by the lack of success of many attempts at reconstructing the entire B. subtilis gltX gene in E. coli (Breton, 1990). The misacylation properties of these GluRSs are consistent with the transamidation pathway since they have to aminoacylate both tRNA Glu and tRNA Gln to compensate for the lack of GlnRS activity. Therefore, the absence of GlnRS and the misacylating activity of R. meliloti GluRS strongly suggest that R. meliloti uses the transamidation pathway for the formation of Gln-tRNA Gln . To further investigate this possibility, we looked for a Glu-tRNA Gln amidotransferase activity in an R. meliloti crude extract. As shown in Fig. 4, the specific transformation of Glu-tRNA Gln to Gln-tRNA Gln by an R. meliloti crude extract confirms the presence of the transamidation pathway.
R. meliloti is the first member of Gram-negative purple bacteria reported to use the transamidation pathway for the formation of Gln-tRNA Gln . Since it has been suggested that mitochondria arose from an endosymbiotic capture of a member of the ␣-subdivision of the purple bacteria (Gray, 1992) and since organelles use the transamidation pathway for Gln-tRNA Gln formation (Schön et al., 1988), the occurrence of this pathway in R. meliloti is consistent with this affiliation. Moreover, the relative branching order of yeast mitochondrial GluRS within the Rhizobium, Thermus, and Chlamydia group (Fig. 6) indicates that they shared a common ancestor, but no clear affiliation with a particular member of this group can be deduced. Previous phylogenetic analyses are also consistent with this grouping since those based on cytochrome c (Dickerson, 1980), rRNA (Yang et al., 1985), HSP60 (Viale and Arakaki, 1994), and HSP70 (Falah and Gupta, 1994) relate the mitochondrial origin to the ␣-subdivision of the purple bacteria, and those based on elongation factor Tu (Cousineau et al., 1992) indicate an affiliation with Chlamydia, an obligate intracellular parasite. Therefore, the endosymbiotic origin of mitochondria from an ancestor of the ␣-subdivision of the purple bacteria is reinforced by our biochemical and phylogenetic data.
Our trees show that eukaryotic GluRSs and GlnRSs branch together as an outgroup of bacterial GluRSs, implying that FIG. 6. Phylogenetic trees of GlxRS. The amino acid alignment of the GlxRS Rossman fold was analyzed for the most parsimonious tree with the PROTPARS program of the PHYLIP package. Confidence levels were determined by the bootstrapping method (500 replicas) and are shown as percentages above the branches. GluRSs and GlnRSs are identified as described for Fig. 5. they share a more recent common ancestor. Moreover, the topology of yeast cytoplasmic GluRS and GlnRS and that of yeast mitochondrial GluRS, which constitute the complete set necessary for formation of Glu-tRNA Glu and Gln-tRNA Gln in yeast, clearly show that mitochondrial GluRS has a bacterial origin and that cytoplasmic GluRS and GlnRS share a common ancestor (Fig. 6). This close evolutionary relationship between eukaryotic GluRS and GlnRS is consistent with the prevailing hypothesis suggesting that they evolved from a gene duplication (Lamour et al., 1994) and contrasts with the specificity hypothesis (Nagel and Doolittle, 1991), which states that all aaRSs specific for the same amino acid are more similar to their counterpart from different organisms than to any other synthetase specific for a different amino acid. Furthermore, the strong conservation between eukaryotic GluRSs and GlnRSs of residues proposed to interact with their respective amino acid substrates (Fig. 5) (Perona et al., 1993) implies that a small number of changes were sufficient to convert the specificity from glutamate to glutamine. Therefore, our phylogenetic analysis confirms that cytoplasmic GluRS and GlnRS evolved from a GlnRS-like GluRS that was used in a transamidation pathway (Fig. 7) and suggests that the specificity for glutamine evolved from a glutamate-specific enzyme.
Even if the horizontal gene transfer hypothesis is the most probable one for the emergence of GlnRS in bacteria, it does not explain how it was included in the protein biosynthetic apparatus. We suggest that the stable acquisition of GlnRS has been beneficial if it occurred under conditions favoring the elaboration of a GluRS specific to tRNA Glu , without affecting the integrity of protein biosynthesis. Recently, the complete sequence of the 0.4 -4-min region of the E. coli chromosome (Fujita et al., 1994) revealed the presence of an ORF of 256 codons, similar to the Rossman fold of GluRSs and probably originating from the duplication of an E. coli GluRS gene. The emergence of a more specific GluRS required for the stable acquisition of a transferred GlnRS was probably facilitated by a GluRS gene duplication; one GluRS would still aminoacylate both tRNA Glu and tRNA Gln to make possible the transition to the direct pathway without affecting protein synthesis, while the other GluRS would be free to restrict its specificity only to tRNA Glu . Finally, the capture of the tRNA Gln by the GlnRS would eliminate the necessity of the transamidation pathway and of the GluRS involved in it. Therefore, we suggest that ORF256 is a relic of a GluRS that was part of an ancient transamidation pathway present in the purple bacteria ancestor. Moreover, the facts that this ORF is probably not normally expressed in E. coli and that ORF256 clusters with some significance in the R. meliloti group are consistent with this suggestion.
The widespread distribution of the transamidation pathway in bacteria and particularly its presence in the purple bacteria division suggest that the presence of a GlnRS is exceptional within the bacterial kingdom. Moreover, the relative branching order of the purple bacteria within a bacterial tree based on 16 S rRNA sequences (Olsen et al., 1994) suggests that GlnRS is probably restricted to its ␥and ␤-subdivisions. Thus, the common ancestor of the purple bacteria probably did not possess a GlnRS, but instead used a transamidation pathway (Fig. 7). In contrast, GlnRS is widely distributed in the eukaryotic kingdom and probably present in all eukaryotes since it was found in G. lamblia, a flagellated protozoan devoid of mitochondria (Shields, 1993) that is known to be one of the deepest branches in the eukaryotic tree (Sogin et al., 1989). Therefore, we suggest that GlnRS appeared as an early event in the eukaryotic lineage, before the first major known divergence in this kingdom. Since archaebacteria and eukaryotes are known to share a common ancestor (Iwabe et al., 1989;Brown and Doolittle, 1995) and since the transamidation pathway was found in a member of the archaebacteria (White and Bayley, 1972), it is likely that this common ancestor used the transamidation pathway. Thus, the transamidation pathway is likely to be an ancestral feature that was retained in the contemporary bacteria and archaebacteria and replaced in the eukaryotic lineage by the direct glutaminylation pathway with the emergence of GlnRS (Fig. 7).
We propose an evolutionary scenario for the GlxRS family ( Fig. 7) that differs from the previous model obtained for the evolution of aminoacyl-tRNA synthetases Doolittle, 1991, 1995;Brown and Doolittle, 1995). As shown in our model (Fig. 7), divergence of an ancestral GluRS led to a GluRS-like GluRS (bacterial ancestor) and to a GlnRS-like GluRS (eukaryotic and archaebacterial ancestor) that were used in a transamidation pathway. The GlxRS tree (Fig. 6) confirms that the GlnRS-like GluRS gene duplication, from which evolved the contemporary eukaryotic GluRS and GlnRS, occurred after the divergences leading to the three kingdoms. To explain the emergence of a GlnRS in some bacteria, we suggest that the stable acquisition of a GlnRS gene, following its horizontal transfer (Lamour et al., 1994), was favored by a GluRS gene duplication. Therefore, the GluRS-like sequence and the Gl-nRS-like sequence do not depict the adaptation to different specificities for glutamate or glutamine, but instead represent the wide divergence of primary structure of the ancestral GluRS. It is possible that this wide divergence reflects a longer evolutionary history of GluRS or the adaptation of a glutamatespecific enzyme to different tRNA environments as suggested by Rogers and Söll (1995). Thus, the transamidation pathway was the first solution for incorporation of glutamine into proteins. This scenario is the most parsimonious one since it assumes that the genetic code was established and that glutamine was already included in the amino acid pool before the FIG. 7. Hypothetical scheme for GlxRS evolutionary history. Unbroken arrows show a gene transfer event and its direction, and the X on the line represents the inactivation of a gene function. The dashed arrow refers to the capture by a eukaryotic cell of an endosymbiotic ␣ purple bacteria (␣), leading to mitochondria. Gene duplications are shown with a D on the node. This model, detailed under ''Discussion,'' is based on the distribution of the GlnRS activity among organisms and on the evolutionary relationships among those organisms as inferred from GlxRS phylogenetic analyses. emergence of the three domains.
Although we assumed in our model that a close evolutionary relationship between archaebacteria and eukaryotes exists as previously suggested (Iwabe et al., 1989;Brown and Doolittle, 1995), alternative relationships between the three domains have been proposed by others (Lake, 1988;Golding and Gupta, 1995). In fact, rooting of the universal tree and therefore evolutionary relationships between the three domains have not yet been clearly resolved. The characterization of archaebacterial GlxRS genes will be very important to address this question.