Impact of model assumptions on the inference of the evolution of ectomycorrhizal symbiosis in fungi

Ectomycorrhiza (ECM) is a symbiotic relation between plant and fungi that
is essential for nutrient uptake of many stand forming trees. There are
two conflicting views about the evolution of ECM in fungi suggesting (1)
relatively few transitions to ECM followed by reversals to non-ECM, or (2)
many independent origins of ECM and no reversals. In this study, we
compare these, and other, hypotheses and test the impact of different
models on inference. We assembled a dataset of five marker gene sequences
(nuc58, nucLSU, nucSSU, rpb1, and rpb2) and 2,174 fungal taxa covering the
three subphyla: Agaricomycotina, Mucoromycotina and Pezizomycotina. The
fit of different models, including models with variable rates in clades or
through time, to the pattern of ECM fungal taxa was tested in a Bayesian
framework, and using AIC and simulations. We find that models implementing
variable rates are a better fit than models without rate shift, and that
the conclusion about the relative rate between ECM and non-ECM depend
largely on whether rate shifts are allowed or not. We conclude that
standard constant-rate ancestral state reconstruction models are not
adequate for the analysis of the evolution of ECM fungi, and may give
contradictory results to more extensive analyses.