Deep genomic-scale analyses of the metazoa reject Coelomata: evidence from single- and multigene families analyzed under a supertree and supermatrix paradigm

Abstract

Solving the phylogeny of the animals with bilateral symmetry has proven difficult. Morphological studies have suggested a variety of alternative hypotheses, of which, Hyman's Coelomata hypothesis has become the most established. Studies based on 18S rRNA have failed to endorse Coelomata, supporting instead the rearrangement of the protostomes into two new clades: the Lophotrochozoa (including, e.g., the molluscs and the annelids) and the Ecdysozoa (including the Panarthropoda and most pseudocoelomates, such as the nematodes and priapulids). Support for this new animal phylogeny has been attained from expressed sequence tag studies, although these generally have a limited gene sampling. In contrast, deep genomic-scale analyses have often supported Coelomata. However, these studies are problematic due to their limited taxonomic sampling, which could exacerbate tree reconstruction artifacts. Here, we address both of these sampling limitations; we study the effect of long-branch attraction (LBA) in deep genomic-scale analyses and provide convincing evidence, using both single- and multigene families, that Coelomata is an artifact. We show that optimal outgroup selection is key in avoiding LBA and identify the use of inadequate outgroups as the reason previous deep genomic-scale analyses found strong support for Coelomata.

FIG. 1.—

Testing outgroup choice in minimally sampled data sets. Majority rule consensus trees derived from ML gene trees. Bootstrap support from both multigene families and single-gene families is shown for each node. The following core ingroup species are common to all: Homo sapiens, Drosophila melanogaster, and Caenorhabditis elegans. Outgroups used are (A) the yeast Saccharomyces cerevisiae (B) the cnidarian Nematostella vectensis. Bootstrap support values are shown for each combination of protein family identification and alignment method. Bootstrap support is displayed for single-gene families, multigene families, and combined single-gene families and multigene families, respectively.

FIG. 2.—

Phylogenomic supertrees of the Bilateria. (A) A tree derived using only the fungal outgroup. This tree is based on 2,164 from 41 species. (B) A tree derived using fungal and animal (nonbilaterian) outgroups. This tree is based on 1,949 genes from 43 species. The monophyly of Ecdysozoa, Lophotrochozoa, and Protostomia is recovered in (B), whereas (A) supports Coelomata. Numbers at the nodes represent bootstrap support. Full circles indicate 100% bootstrap support for a node.

FIG. 3.—

Phylogenomic supertree of the Bilateria recovered using only animal (nonbilaterian) outgroups. This tree is based on 2,216 genes from 42 species. High support for the monophyly of Ecdysozoa, Lophotrochozoa, and Protostomia can be observed. Numbers at the nodes represent bootstrap support. Full circles indicate 100% bootstrap support for a node.

FIG. 4.—

Results of the supermatrix analyses. (A) The effect of LBA is obvious if one roots the tree using Nematostella vectensis, as a tree essentially consistent with the new animal phylogeny is recovered, but Saccharomyces cerevisiae is incorrectly nested within the Protostomia. (B) A tree illustrating that Ecdysozoa is easily recovered when analyses are performed using only nonbilaterian animals as outgroups. Numbers at the nodes represent posterior probabilities. Full circles indicate a posterior probability of 1. Posterior probabilities lower than 1 have only been reported for nodes that are relevant to the Ecdysozoa versus Coelomata problem. Urochordata is collapsed in a basal polytomy because the posterior probability of Deuterostomia is less than 0.5.