Fast-evolving alignment sites are highly informative for reconstructions of deep Tree of Life phylogenies

The trimming of fast-evolving sites, often known as “slow-fast” analysis, is broadly used in microbial phylogenetic reconstruction under assumption that fast-evolving sites do not retain accurate phylogenetic signal due to substitution saturation. Therefore, removing sites that have experienced multiple substitutions would improve the signal-to-noise ratio in phylogenetic analyses, with the remaining slower-evolving sites preserving a more reliable record of evolutionary relationships. Here we show that, contrary to this assumption, even the fastest evolving sites, present in conserved proteins often used in Tree of Life studies, contain reliable and valuable phylogenetic information, and that the trimming of such sites can negatively impact the accuracy of phylogenetic reconstruction. Simulated alignments modeled after ribosomal protein datasets used in Tree of Life studies consistently show that slow-evolving sites are less likely to recover true bipartitions than even the fastest-evolving sites. Furthermore, site specific substitution-rates are positively correlated with the frequency of accurately recovered short-branched bipartitions, as slowly evolving sites are less likely to have experienced substitutions along these intervals. Using published Tree of Life sequence alignment datasets, we additionally show that both slow-and fast-evolving sites contain similarly inconsistent phylogenetic signals, and that, for fast-evolving sites, this inconsistency can be attributed to poor alignment quality. Furthermore, trimming fast sites, slow sites, or both is shown to have substantial impact on phylogenetic reconstruction across multiple evolutionary models. This is perhaps most evident in the resulting placements of Eukarya and Asgardarchaeota groups, which are especially sensitive to the implementation of different trimming schemes.Significance StatementIt is common practice among comprehensive microbial phylogenetic studies to trim fast-evolving sites from the source alignment in the expectation to increase the signal to noise ratio. Here we show that despite fast-evolving sites being more sensitive to parameter misspecifications than mid-rate evolving sites, such sensitivity is comparable, if not smaller, than what we observe among slow-evolving sites. Through the use of both empirical and simulated datasets we also show that, besides the lack of evidences regarding the noisy nature of fast-evolving sites, such sites are of core importance for the reliable the reconstruction of short-branched bipartitions. Such points are exemplified by the variations in the Eukarya+Archaea Tree of Life when subjective alignment trimming strategies are employed.