To ensure that the prior branching times of the starting tree fulfilled the constraints imposed by the calibration priors we transformed branch length into absolute time using penalized likelihood [ 99] with the chronopl command in the R [ 100] package ape [ 101].
Default priors set by MrBayes were used except for the branch-length priors; branch-length priors were set as exponential with means of 1, 0.1, 0.01 and 0.001 in four separate runs to test prior sensitivity, since a large branch-length prior can result in unrealistically long trees [ 35].
The formation of a minimum of 18 leaves was required prior to branch initiation, and an apical portion devoid of branches was maintained at a similar length throughout development.
To determine whether this prior affects branch length estimates, we repeated our previous analyses of the 100 4-taxon datasets simulated on equal branch length trees, specifying both an exponential prior with mean equal to 1 and a uniform prior (lower bound of 0, upper bound of 1).
For a comparison, we also used an exponential prior on branch lengths.
Assuming an exponential Bayesian prior on branch lengths did not improve, and under certain extreme conditions significantly diminished, performance.
We used a coalescent prior on branch lengths (which involves forcing molecular clock), and default settings for other parameters.
This exponential prior for branch lengths is the default in MrBayes because a uniform prior has been suggested to result in overestimated posterior probabilities for clades [ 44].
