Evidence that eukaryotic triosephosphate isomerase is of alpha-proteobacterial origin

Escherichia coli DH-5aF′ was used for all molecular manipulations and was grown on Luria–Bertani agar or in Luria–Bertani broth under ampicillin selection. R. etli CFN42 cells and DNA were provided by E. Martínez-Romero (Centro de Investigación Sobre Fijacion de Nitrogeno), F. tularensis LVS DNA was provided by F. Nano (University of Victoria), and C. aurantiacus J-10-f1 DNA was provided by J. Lopez and R. E. Blankenship (both at Arizona State University).

Genomic DNA (50–200 ng) from R. etli, F. tularensis, and C. aurantiacus was used in PCR amplifications consisting of 10 mM Tris·HCl, 50 mM KCl, 1.5 mM MgCl₂, 0.1% Triton X-100, 0.2 mg/ml BSA (pH 9, 25°C), 10 mM each dNTP, 2 units of Taq polymerase, 0.5 unit of Pfu polymerase, and 1 μM each primer. Two sets of primers (provided by A. J. Roger, Dalhousie University) were used that match highly conserved blocks of all known TPI genes. One set corresponded to amino acids 6–11 and 232–237 (TF1, ACGTCTCGAGTTCGGTGGNAAYTGGAA, and TR1, ATCTCTAGAAGTGATGCNCCNCCNAC) and an internal set corresponded to 72–77 and 170–175 (TF4, CGAGAATTCAACGGTGCATTYACNGGNGA, and TR2, AGCTCTAGACCTGTNCCDATNGCCC), and numbering was according to the E. coli sequence.

Products of the expected size were isolated from agarose gels by freezing crushed gel slices in an equal volume of TE (10 mM Tris·HCl/1 mM EDTA, pH 8.0) and two volumes of phenol at −70°C, followed by centrifugation and ethanol precipitation of the aqueous phase. Purified products were cloned using TA-cloning vector, pCRII (Invitrogen), and sequenced by the dideoxy chain termination method using T7 polymerase (Pharmacia). All sequences reported are based on at least two independent clones and were completed on both strands, except the type 2 and type 3 R. etli clones, which are based on single clones sequenced on both strands. In no case was any discrepancy between duplicate clones observed.

New sequences were aligned with 37 existing database entries for TPI genes and the unpublished sequences from Entamoeba histolytica (provided by A. J. Roger) using the pileup program from the GCG package (25) and edited manually (Fig. 1).

Phylogenetic trees based on this alignment were inferred using distance, parsimony, and maximum likelihood methods. The type 2 and type 3 R. etli sequences were not included in phylogenetic analyses due to their short size, although we note that, in trees where all three Rhizobium sequences were included, they branched together with high affinity and had little other effect on tree topology. Corrected distance measurements were calculated for 264 sites based on the PAM 250 distance matrix, and trees were constructed by neighbor-joining using the protdist and neighbor programs from the phylip version 3.57 package (26). Statistical support for each node was assessed by conducting 100 bootstrap replicates. Unweighted parsimony trees were found by conducting heuristic searches with 50 rounds of stepwise addition with tree bisection and reconnection using paup version 3.1.1 (27). Bootstrap support was also calculated by conducting 100 random replicates. Protein maximum likelihood analyses was conducted on 213 positions using constraints as described in the results, and the JTT transition probability matrix in the protml program from the molphy version 2.2 package (28).

Additional statistical tests were performed to see if the differences observed between the tree topologies based on TPI and other markers are significant or the result of poor resolution in the TPI tree. The eukaryotic branch in both parsimony and neighbor-joining trees was moved to alternative positions in the eubacteria. These alternative topologies were then compared based on Templeton’s method (29) and evaluated statistically as described by Felsenstein (30) and Kishino and Hasegawa (31). For both starting topologies, two methods were used to evaluate the difference between alternatives and the SE: maximum likelihood and parsimony. Maximum likelihood tests were carried out on a data set of 213 positions again using protml. Tests based on parsimony were carried out by calculating the number of steps at each of the 264 positions and using these values to calculate the SE from the best tree according to Kishino and Hasegawa (31). The Templeton test built into user tree option of protpars in the phylip package were also used and the results resembled those shown.

Amplification products covering over 90% of the TPI gene were isolated from C. aurantiacus, F. tularensis, and R. etli. With the former two species, the sequence was isolated as a single 730-bp fragment using primers TF1 and TR1. In both cases, two clones were chosen and sequenced on both strands and found to be identical. With R. etli, this primer combination failed, so the same portion of the gene was amplified in two overlapping pieces, using TF1 and TR2 for the 5′ end, and TF4 and TR1 for the 3′ end. Six clones of the 3′ end were sequenced and found to be identical, but among the six clones isolated and sequenced from the 5′ end, three distinct TPI coding sequences were identified. The three variants, named type 1, 2, and 3, share between 72.5 and 65.3% identity at the amino acid level. Divergence notwithstanding, these three amino acid sequences are extremely similar to one another compared with genes from other species, and they also contain shared insertions and deletions, strongly supporting the notion that they are paralogues that diverged relatively recently. Of the three, type 1 was identical over the 282-bp overlap region with the six 3′ clones, so these have been treated as two fragments of a single gene.

Phylogenetic trees based on TPI sequences from 22 eukaryotes, 18 eubacteria, and 1 archaebacterium were inferred using parsimony, distance, and maximum likelihood methods. Parsimony and neighbor-joining trees are shown in Fig. 2 A and B, respectively, each with bootstrap proportions for nodes supported >30%. Parsimony analysis yielded eight trees of identical length (2745 steps) that differed slightly in the exact position of several eukaryotes (namely Gracilaria, Entamoeba, the order within the fungi, and the presence of Arabidopsis–Coptis and Culex–Drosophila clades). The neighbor-joining and parsimony trees also differ from one another in a few other branches, but none that is significantly supported in either analysis and none that is central to the questions posed here. Two constant features of all the preferred trees is the association between eukaryotic and proteobacterial TPI genes (in particular that of the alpha-proteobacterium R. etli) and the relatively great distance between eukaryotic and archaebacterial sequences. Protein maximum likelihood analysis was conducted by constraining the topology of the eukaryotes to match that shown in Fig. 2A and dividing the prokaryotes into the groups in the same figure: Rhizobium, gamma-proteobacteria, mycoplasmas, other low GC Gram-positive bacteria, actinomycetes, Thermotoga, Chloroflexus, and Synechocystis, and Pyrococcus. The branching order of these nine groups was then exhaustively searched, but the best 100 trees were virtually indistinguishable from one another statistically (all but a few being within 1.98 SEs from the best tree). Once again, these trees were consistent in that the eukaryotic TPI genes were always quite distant from that of the archaebacterium and nested within the proteobacteria (data not shown).

There is now a great deal of support both from molecular phylogeny and molecular biology that the archaebacteria are more closely related to eukaryotes than are eubacteria (2–4, 6, 7). The expected phylogenetic relationship of a eukaryotic cytosolic enzyme is therefore exactly the opposite of that observed here: if TPI is derived from the “host” genome, then eukaryotic genes ought to branch closer to the archaebacteria than they do to eubacteria. The highly divergent nature of the Pyrococcus TPI gene (32) raises concerns that this gene is a specialized paralogue that could conceivably be causing an erroneous phylogeny. However, excluding the Pyrococcus sequence from the alignment had little effect on the outcome of the trees (data not shown) and more importantly, the recently completed genome of Methanococcus jannaschii contains a single TPI gene that is very similar to that of Pyrococcus, suggesting further that the Pyrococcus TPI sequence is representative of the archaebacterial lineage to which these two species belong.

The relationship between eukaryotic and proteobacterial TPI genes was tested first by performing 100 bootstrap replicates on the distance and parsimony analyses. In both cases, the bootstrap support was almost universally low for all nodes except those between the most closely related taxa. TPI is a relatively small protein, and it is apparent that it has little power to resolve many details of organismal phylogeny (23). Nonetheless, the question posed here is a specific one—the significance of the position of the eukaryotic TPI genes within the eubacteria—and this was tested directly by comparing the “TPI tree” (using the topologies inferred by both parsimony and neighbor-joining) to alternative trees, according to the method described by Templeton (29). The alternative topologies were chosen by rooting the eukaryotes in each of the main intergroup branches within the prokaryotes. Fig. 3A shows the results of these tests, measured by parsimony and maximum likelihood, for the topology inferred by parsimony, and Fig. 3B shows the results for the neighbor-joining topology. In all tests the TPI tree is superior to all alternatives and is significantly so in all cases except that where the outgroup of eukaryotes is all proteobacteria. Topologies G in Fig. 3A and H in Fig. 3B are of particular interest, as these trees are a very close approximation to the topology of universal trees inferred by other molecular markers (archaebacteria as sisters to eukaryotes; refs. 2 and 3), yet these trees are between 2 and 4.4 SEs worse in both tests than the topologies actually inferred from TPI, suggesting that the phylogenetic history of TPI is different from these other molecules. Once again, these tests were repeated on the tree topologies derived by excluding Pyrococcus, and, as before, the archaebacterial sequence was apparently not distorting the analysis: the ultimate result was the same and statistical significance changed very little (data not shown).

The possibility that either the branching order defined within the eukaryotes or the root of the eukaryotic subtree had some distorting effect on the likelihood of alternative topologies was also examined by conducting 100 independent maximum likelihood replicates with a random pair of eukaryotic sequences (so that there is only one topology or root within ukaryotes) and 16 eubacteria. In every one of these 100 replicates, the TPI topology was the best, although once again the difference between the R. etli specifically and all the proteobacteria was often insignificant.