Data from: Model choice, missing data and taxon sampling impact phylogenomic inference of deep Basidiomycota relationships

Resolving deep divergences in the tree of life is challenging even for
analyses of genome-scale phylogenetic datasets. Relationships between
Basidiomycota subphyla, the rusts and allies (Pucciniomycotina), smuts and
allies (Ustilaginomycotina) and mushroom-forming fungi and allies
(Agaricomycotina) were found particularly recalcitrant both to traditional
multigene and genome-scale phylogenetics. Here, we address basal
Basidiomycota relationships using concatenated and gene-tree based
analyses of various phylogenomic datasets to examine the contribution of
several potential sources of bias. We evaluate the contribution of
biological causes (hard polytomy, incomplete lineage sorting) versus
unmodeled evolutionary processes and factors that exacerbate their effects
(e.g. fast-evolving sites and long-branch taxa) to inferences of basal
Basidiomycota relationships. Bayesian MCMC and likelihood mapping analyses
reject the hard-polytomy with confidence. In concatenated analyses,
fast-evolving sites and oversimplified models of amino acid substitution
favored the grouping of smuts with mushroom-forming fungi, often leading
to maximal bootstrap support in both concatenation and coalescent
analyses. On the contrary, the most conserved data subsets grouped rusts
and allies with mushroom-forming fungi, although this relationship proved
labile, sensitive to model choice, to different data subsets and to
missing data. Excluding putative long branch taxa, genes with high
proportions of missing data and/or with strong signal failed to reveal a
consistent trend toward one or the other topology, suggesting that
additional sources of conflict are at play. While concatenated analyses
yielded strong but conflicting support, individual gene trees mostly
provided poor support for any resolution of rusts, smuts and
mushroom-forming fungi, suggesting that the true Basidiomycota tree might
be in a part of tree space that is difficult to access using both
concatenation and gene-tree based approaches. Inference-based assessments
of absolute model fit strongly reject best fit models for the vast
majority of genes, indicating a poor fit of even the most commonly used
models. While this is consistent with previous assessments of
site-homogenous models of amino acid evolution, this does not appear to be
the sole source of confounding signal. Our analyses suggest that
topologies uniting smuts with mushroom-forming fungi can arise as a result
of inappropriate modeling of amino acid sites that might be prone to
systematic bias. We speculate that improved models of sequence evolution
could shed more light on basal splits in the Basidiomycota, which, for
now, remain unresolved despite the use of whole genome data.