Computational protocol: The evolution of giant flightless birds and novel phylogenetic relationships for extinct fowl (Aves, Galloanseres)

Similar protocols

Protocol publication

[…] Parsimony analyses of the morphological data matrix were performed using PAUP_4.0b10 [], using heuristic searches with TBR branch swapping and 1000 random addition replicates per search. Inapplicable characters (coded as gaps ‘-’) were treated as missing data. Strict consensus trees were computed from the set of most parsimonious trees (MPTs), and clade support was assessed by bootstrapping [] using the same settings and 1000 replicates. To prevent the bootstrap analyses from getting stuck on replicates with huge numbers of equally parsimonious trees, nchuck was set to 2000. The tree was rooted between extant palaeognaths and extant neognaths (see electronic supplementary material for nexus file).Parsimony analyses were initially performed without any constraints (i.e. no molecular backbone), which resulted in topologies robustly contradicted by molecular data, and which raised suspicions of homoplasy artefacts (e.g. large-bodied taxa clustering together despite being morphologically dissimilar and/or geographically remote). To examine what may be driving the discordance, we conducted analyses with selective weighting of characters in unconstrained analyses, which showed that those of pectoral girdle elements (linked to loss of flight) were homoplasious and driving the contradictory topologies. This resulted in improved congruence with molecular analyses, supporting the view that the morphological data were extensively affected by homoplasy; the improved congruence also increased confidence in the molecular topology. As the primary aim of the analyses was to ascertain the relationships of fossil taxa in the context of the most likely phylogenetic relationships of extant taxa, we focused on the parsimony analyses with a molecular backbone. We examined the effect of the poorly known Brontornis by deleting this taxon and repeating the analysis. Also, because parsimony analyses robustly retrieved flightlessness as primitive for palaeognaths (an inference likely erroneous as abundant molecular data show that several ratite lineages each derive from a volant ancestor [,]), we repeated the analysis removing the ratite taxa to preclude their flightless condition causing problems with the optimization of a flightless ancestor for neornithines.Diet, flight and body size were optimized on the primary trees in the ‘backboned’ analysis using parsimony via Mesquite [].Body mass was optimized on the MPT(s) using linear parsimony in Mesquite []; discrete phenotypic traits were optimized using parsimony (ordered or unordered as applicable) using PAUP. [...] Bayesian analyses co-estimate topologies, branch lengths (anagenetic and chronological), ancestral states, divergence dates and evolutionary rates (see Lee & Palci [] for a review of the benefits of this approach). The morphological, body size and ecological data were simultaneously analysed using tip-dated approaches that employ the ages of the fossil taxa [,], as implemented in BEAST 1.8.4 []. The root age was constrained to be broadly consistent with Prum et al. ([]: analyses with Vegavis included). No other node age constraints were imposed, and the retrieved dates are generated from the phenotypic and stratigraphic information (geological age) contained in the fossil taxa (tips). Bayes Factors were used to test the need to accommodate rate variability among-characters (i.e. γ parameter) and among-lineages (i.e. relaxed clock). Each Bayesian analysis was repeated four times to confirm stationarity, with the post-burnin samples of all four runs combined for statistical analyses and consensus trees.Bayesian phylogenetic analyses were performed with, and without, the molecular backbone. This is the first study to implement backbone constraints in BEAST 1; the relevant code is annotated in the xml file in the electronic supplementary material. In addition to estimating tree topology, divergence dates and morphological evolutionary rates, the Bayesian analyses also simultaneously estimated ancestral states for body size, diet and flight ability at each ancestral node, which can then be summarized on the timetree. Diet in fossil taxa was coded conservatively as ‘?’ and thus inferred using character optimization; the only exception was moa where gut contents indicate herbivory []. Our discussion focuses on the backboned analysis, as the analysis without the molecular backbone resulted in relationships among living taxa that are contradicted by a large body of genomic work. […]

Pipeline specifications

Software tools PAUP*, BEAST
Application Phylogenetics