Phylogenetic inference
Phylogenetic resolution was first explored using IQ-TREE ultra-fast bootstrap consensus with multi-partition sequence evolution model selected via ModelFinder (see Table S2). This was done for both with and without mtg-block datasets, and a cut down version of nuclear data only for 94 taxa with >2 nuclear genes. To further investigate the consistency of support among genes for primary lineages, we employed ASTRAL consensus species tree from gene trees approach (rather than site support methods due to the patchwork supermatrix data). This used a cut down dataset of 59 taxa for 8 genes.
Bayesian relaxed-clock analyses were performed in BEAST v2.4 (Drummondet al. , 2006; Bouckaert et al. , 2014), with a lognormal clock and Yule speciation prior (each with 1/X parameterization), and using a seven partition sequence evolution model optimized by ModelFinder (nuclear gene codon positions, intron+12S, mitochondrial codon positions; Table S2). Dating calibrations comprised of two Columbidae fossil constraints and two secondary outgroup priors.Rupephaps, a ptilinopine fruit pigeon from New Zealand, was applied as exponential (offset=22.0, mean=3.0) on the stem subtendingHemiphaga , Lopholaimus and Gymnophaps (Worthyet al. , 2009). Primophaps , a phabine pigeon from northern Australia, was applied as exponential (offset=19.0, mean=3.0) on the main Australian radiation of terrestrial seed-eating pigeons (stem subtending Leucosarcia to Phaps ) (Worthy, 2012). Two normal secondary priors, based on Prum et al. (2015), were placed on the root (mean=63.0, sd=4.3) and Columbimorphs (mean=58.0, sd=4.8). Preliminary tests indicated a birth-death speciation prior was not required above the simpler Yule prior.
We ran two independent Bayesian MCMC analyses of 50 million steps sampling every 5,000, with a burnin of 20% that returned all parameters with ESS >500 (Rambaut et al. , 2014), with near identical maximum clade credibility (MCC) consensus tree topology, node support, and dating (see electronic supplementary material for the complete BEAST2 xml file). Runs were then pooled. For the purposes of biogeographic analyses, in addition to a MCC tree, a subset of 100 trees were drawn from the BEAST posterior sample to reasonably represent the topological and branch length variation of the total posterior. Outgroup taxa were then removed (Phytools R package; Revell 2012).