The Root of the Tree of Life in the Light of the Covarion Model

A few duplicated genes have been found useful to root the universal tree of life. Despite controversial results, the consensus led to locate the root in the eubacterial branch. However, we demonstrated (Philippe and Forterre 1999) that all these markers were in fact unsuitable for any firm conclusion, mainly because of their high level of mutational saturation, which masks a major part of the phylogenetic signal. But then, the very persistence of signal for events as early as the separation of the three domains becomes puzzling. This paradox was studied here for translation elongation factor proteins, EF-1α and EF-2, which appeared to be one of the least confusing markers. We showed that these proteins do not conform to a classical rate-across-sites pattern, as those modeled by a gamma law, but rather to a covarion-based model, because the evolutionary rate of a given position often changes between taxonomic groups. Conservation of the very ancient signal can thus be better explained by the covarion model: a substitution can occur in deep branches, and the position remains constant afterward, as ``fossilized'' by a change of covation. As no reconstruction method has up to now taken into account this complex model, we devised a simple method for extracting the phylogenetic signal, by considering the variability of sequence positions within predefined phylogenetic groups. We showed that noise quantitatively prevailed upon signal. Parsimony will produce erroneous topologies, because it has to minimize primarily the number of steps of the noise. In contrast, our method effectively concentrated the signal and was more suitable for inferring ancient events. We consequently found the eubacterial rooting to be presumably due to a long branch attraction artifact, because of the higher evolutionary rate of Eubacteria for these proteins. Among the two other rooting possibilities, the eukaryotic rooting appeared to be more supported, although not enough to be conclusive.