Taxon sampling unequally affects individual nodes in a phylogenetic tree: consequences for model gene tree construction in SwissTree

Medium to large phylogenetic gene trees constructed from datasets of different species density and taxonomic range are rarely topologically consistent because of missing phylogenetic signal, nonphylogenetic signal and error. In this study, we first use simulations to show that taxon sampling unequally affects nodes in a gene tree, which likely contributes to controversial conclusions from taxon sampling experiments and contradicting species phylogenies such as for the boreoeutherians. Hence, because it is unlikely that a large gene tree can be reconstructed correctly based on a single optimized dataset, we take a two-step approach for the construction of model gene trees. First, stable and unstable clades are identified by comparing phylogenetic trees inferred from multiple datasets and data types (nucleotide, amino acid, codon) from the same gene family. Subsequently, data subsets are optimized for the analysis of individual uncertain clades. Results are summarized in form of a model tree that illustrates the evolutionary relationship of gene loci. A case study shows how a seemingly complex gene phylogeny becomes increasingly consistent with the reference species tree by attentive taxon sampling and subtree analysis. The procedure is progressively introduced to SwissTree (http://swisstree.vital-it.ch), a resource of high confidence model gene (locus) trees. Finally we demonstrate the usefulness of SwissTree for orthology benchmarking.