Horizontal transfers confuse the prokaryotic phylogeny based on the HSP70 protein family

Sir, Recently, Gupta has argued that archaebacteria and Gram-positive eubacteria are closely related to the exclusion of Gram-negatives and proposed a new classification of prokaryotic organisms into these two major groups (Gupta, 1998, Mol Microbiol29: 695–707). This idea is supported mainly by several sequence signatures, of which the most important is the absence of a segment of about 20 amino acids in the HSP70 genes of Gram-positives and archaebacteria (see Fig. 1). Such indels can indeed provide valuable phylogenetic information, even more reliable than the phylogenies based on complete gene sequences (e.g. Hashimoto et al., 1998, Proc Natl Acad Sci USA95: 6860–6865). However, they can also be misleading, especially in prokaryotes, in which horizontal transfers are frequent (Doolittle and Logsdon, 1998, Curr Biol12: R209–R211; Lawrence and Ochman, 1998, Proc Natl Acad Sci USA95: 9413–9417). We will demonstrate that this is the case for the HSP70 genes. . Unrooted maximum likelihood phylogenetic tree showing representatives of the major groups belonging to the HSP70 protein family. Well-characterized groups appear in colour, whereas black branches correspond to sequences of uncertain origin. Arrowheads indicate the two groups carrying the deletion discussed in the text. Note that these are not closely connected in the tree. Minor classes with few representatives, such as the STCH and NST-1 (only found in metazoans), and YHM4 (only found in fungi), have not been included. ER, endoplasmic reticulum; F, fungi; M, metazoans; P, plants; Schizo, Schizosaccharomyces; BYV, beet yellow closterovirus; CTV, citrus tristeza virus; LChV, little cherry virus; LIYV, lettuce infectious yellow closterovirus. Figure 1 shows a phylogenetic tree for the major classes of the HSP70 protein family. There are four groups in eukaryotes: cytosolic, endoplasmic reticulum, GRP170 and SSE. Cytosolic and endoplasmic reticulum sequences cluster together, as do GRP170 and SSE sequences. In prokaryotes, two groups are clearly defined, MREB and DNAK, which are both present in all eubacteria but scarcely represented in archaebacteria. For instance, Methanobacterium and, perhaps, Archaeoglobus have both genes, whereas Halobacterium and Methanosarcina possess only DNAK (Lange et al., 1997, Curr Microbiol35: 116–121). An additional major phylogenetic group is composed of viral sequences. Gupta considers the absence of about 20 amino acids to represent the ancestral state because it is observed in MREB as well as in DNAK of Gram-positives, concluding that an insertion occurred at the base of Gram-negatives. However, the two groups that display this putative ancestral character are very distant in the phylogenetic tree (Fig. 1). This opens the possibility that this ‘shared’ feature is caused by two independent deletions, making the presence of the segment of about 20 amino acids the ancestral state. The latter hypothesis is favoured by the difficulty of aligning this region, which makes it impossible to determine whether the deletions in Gram-positives, archaebacteria and MREB occurred exactly at the same position and, therefore, whether they have the same size. In fact, size equivalency can only be achieved by introducing several additional gaps in the region surrounding the indel in the MREB sequences (Fig. 2). We have re-examined this alignment using a larger number of MREB sequences, instead of only two as in the original work (Gupta and Golding, 1993, J Mol Evol37: 573–582). A more parsimonious alignment of this region can be obtained with more matches and fewer gaps (Fig. 2). In this case, the deletions for the indel under study show very different sizes (27 amino acids for Gram-positives and 41 for MREB). Although overall sequence conservation is high in the HSP70 family, the ambiguity of the alignment in this region does not clearly support the polarization of the indel proposed by Gupta. On the contrary, it suggests that the two deletions are the result of a convergence affecting this highly variable region of the proteins. . Sequence alignment of the protein region containing the indel discussed in the text (highlighted in grey). Flanking regions, which are clearly aligned among all groups, are not shown. Consensus sequences obtained from 10 sequences from each group are shown with the exception of Escherichia 3 and Borrelia 2 sequences. The MREB consensus sequence appears twice: aligned as proposed by Gupta (MREB Gupta) or by us (MREB). Bold letters indicate amino acid positions perfectly conserved, capital letters those in at least 50% of sequences, dots those variable in more than the 50% of sequences, and dashes indicate gaps. Black boxes show the conserved positions useful to perform the alignment between the different groups. Cyto + ER, cytoplasmic and endoplasmic reticulum. On the other hand, our phylogenetic analysis shows that four sequences from Gram-negative eubacteria emerged far from the major DNAK group in the tree (Fig. 1). Interestingly, two of them, Borrelia 2 (accession number AE001137) and Escherichia 3 (accession number AE000169), display exactly the typical deletion of Gram-positives (Fig. 2). These sequences have no orthologues in closely related organisms whose complete genomes have been sequenced (Treponema and Haemophilus respectively; Fleischmann et al., 1995, Science269: 496–512; Fraser et al., 1998, Science281: 375–388). Thus, two independent recent horizontal transfers from Gram-positive species are the most parsimonious interpretation. Their phylogenetic location far away from the Gram-positive cluster is probably the result of a long branch attraction artifact, as these recently transferred sequences evolve considerably faster. Interestingly, their codon usages are very deviant from the standard values of the respective genomes (data not shown), which strengthens the hypothesis of recent lateral transfer. Two additional eubacterial sequences, from Myxococcus (accession number AC000101) and Escherichia hsc66 (accession number D90883), do not display the deletion but exhibit similar unusual phylogenetic positions. They probably arose by horizontal transfers from other Gram-negative species. However, at least in the case of the hsc66 sequence, the transfer was much more ancient than in the previous cases, as orthologues are present in several β- and γ-proteobacteria (Budin and Philippe, 1998, Mol Biol Evol15: 943–956). Therefore, contrary to the claim of Gupta, horizontal transfers of HSP70 genes seem likely, allowing us to propose another interpretation of the archaebacterial HSP70 genes. At present, HSP70 genes have only been characterized in four genera of euryarchaeota and were not detected in the complete genomes of Methanococcus and Pyrococcus (Bult et al., 1996, Science273: 1058–1073; Kawarabayasi et al., 1998, DNA Res5: 55–76). Despite an intensive search, HSP70 genes have never been found in other archaebacteria, including crenarchaeota (Lange et al., 1997, Curr Microbiol35: 116–121). This narrow taxonomic distribution suggests a recent origin of the known archaebacterial HSP70 genes, as the four genera containing them form a robust monophyletic group in the ribosomal RNA tree (Olsen et al., 1994, J Bacteriol176: 1–6) to the exclusion of all the archaebacteria devoid of HSP70. The simplest explanation, therefore, is that the archaebacterial HSP70 genes have been acquired through horizontal gene transfer from Gram-positive eubacteria, as suggested previously (Gogarten et al., 1996, Symp Soc Gen Microbiol54: 268–292). Structural clues, such as the presence of typical bacterial-like Shine–Dalgarno sequences in these archaebacterial genes (Macario et al., 1991, Gene108: 133–137), also point to this. We propose that the archaebacterial genes issued from a unique ancient transfer event predating their divergence, instead of several independent transfers, as suggested previously. If this is the case, all the archaebacterial sequences should emerge as a monophyletic group in the HSP70 tree. Although this result has not been observed (Gupta, 1998, Mol Microbiol29: 695–707), this is not conclusive evidence, as we will demonstrate here. The phylogenies analysing this problem have always been constructed using neighbour joining (NJ) and maximum parsimony (MP) methods, which are less efficient than maximum likelihood (ML), especially when evolutionary rates are highly variable (Hasegawa and Fujiwara, 1993, Mol Phylogenet Evol2: 1–5). We have carried out a ML analysis of the sequences from archaebacteria and Gram-positive bacteria, with some sequences from Gram-negative bacteria (spirochetes and chlamydiales) included as an outgroup (data not shown). Although the best ML tree shows the polyphyly of archaebacteria in a similar way to the NJ and MP trees, their monophyly is retrieved in trees differing in the standard deviation of likelihood by the statistically insignificant value of 1 (Kishino and Hasegawa, 1990, J Mol Evol30: 151–160). Therefore, the monophyly of the archaebacterial HSP70 sequences cannot be rejected with the available data, and a unique horizontal transfer from Gram-positive eubacteria to an ancient euryarchaeota, masked by the strong differences in evolutionary rate shown by these sequences, is an explanation with equivalent statistical support. The above considerations demonstrate the difficulties inherent in attempts to reconstruct the early history of life even from seemingly impeccable evidence. Conclusions based on sequences of other macromolecules are equally problematic because of the occurrence of horizontal gene transfers (Gogarten et al., 1996, Symp Soc Gen Microbiol54: 268–292) and the limitations of tree reconstruction methods (Philippe and Laurent, 1998, Curr Opin Genet Dev, in press). In our opinion, the relationships between archaebacteria, eubacteria and eukaryotes remain unresolved. Improved analysis of genes for which horizontal transfers have been demonstrated to be very rare promises to provide a real answer. We thank Hervé Le Guyader, Philippe Lopez and Miklós Müller for critical reading of the manuscript. D.M. is supported by a postdoctoral fellowship from the Spanish Ministerio de Educacion y Cultura.