Improving Orthologous Signal and Model Fit in Datasets Addressing the Root of the Animal Phylogeny

Abstract

There is conflicting evidence as to whether Porifera (sponges) or Ctenophora (comb jellies) comprise the root of the animal phylogeny. Support for either a Porifera-sister or Ctenophore-sister tree has been extensively examined in the context of model selection, taxon sampling, and outgroup selection. The influence of dataset construction is comparatively understudied. We re-examine five animal phylogeny datasets that have supported either root hypothesis using an approach designed to enrich orthologous signal in phylogenomic datasets. We find that many component orthogroups in animal datasets fail to recover major lineages as monophyletic with the exception of Ctenophora, regardless of the supported root. Enriching these datasets to retain orthogroups recovering ≥3 major lineages reduces dataset size by up to 50% while retaining underlying phylogenetic information and taxon sampling. Site-heterogeneous phylogenomic analysis of these enriched datasets recovers both Porifera-sister and Ctenophora-sister positions, even with additional constraints on outgroup sampling. Two datasets which previously supported Ctenophora-sister support Porifera-sister upon enrichment. All enriched datasets display improved model fitness under posterior predictive analysis. While not conclusively rooting animals at either Porifera or Ctenophora, we do see an increase in signal for Porifera-sister and a decrease in signal for Ctenophore-sister when data are filtered for orthologous signal. Our results indicate that dataset size and construction as well as model fit influence animal root inference.

Solving an enigma in the tree of life, at the origins of teleost fishes

“…This work evidenced similarities between the organisation of the genomes of sponge and other animals, thus supporting a phylogeny in which ctenophores constitute the sister group of a clade bringing together sponges and the rest of animals. This surprising result, which contradicts the most recent molecular phylogeny analyses [8,31,32], is galvanising research efforts to better understand the methodological and/or biological reasons underlying these incongruences. The thousands highquality genomes that will become available in the near future offer unprecedented opportunities to understand the mechanisms that govern the evolution of genome organisation, and re-examine many controversial branches of the tree of life.…”

Reply to: Available data do not rule out Ctenophora as the sister group to all other Metazoa

“…Other simulation studies have supported recoding 12 , or are at least ambivalent 13 , and as we advocated in R&M, 'a fuller understanding of the implications of recoding is needed' 1 . Second, our claim that REA and WEA15 contain paralog contamination referenced other work 7 and personal communication was limited to WEA17 1 , which has now been shown to support Porifera-sister without recoding when orthogroups with poor orthologous signal (i.e., inability to recover major animal lineages at the gene tree level) are excluded 14 . We concede that personal communication was less than ideal, particularly as sorting orthologs from paralogs is at least somewhat dependent on the approach b Length in substitutions/site of the ancestral branch of the Ctenophore clade for each dataset as analysed at the amino acid level using standard partitioned phylogenomics with site-homogeneous models (i.e., analysis level 'L1' of R&M 1 ).…”

Phylogenomic Inference, Divergence-Time Calibration, and Methods for Characterizing Reticulate Evolution

“…Subsampling is typically done by selecting, for example, half of the loci in a full data matrix with the desirable feature associated with phylogenetic signal. These features-such as alignment length, average bipartition support, treeness divided by relative composition variability, and the number of parsimony and variable sites, among others (Phillips and Penny 2003;Shen et al 2016Steenwyk et al 2019bSteenwyk et al , 2020aSteenwyk et al , 2021Steenwyk et al , 2022bMongiardino Koch 2021;McCarthy et al 2023;Redmond et al 2023)-aim to capture the information content of a locus.…”