Multi-locus species tree for the Amazonian peacock basses (Cichlidae: Cichla): Emergent phylogenetic signal despite limited nuclear variation

- Twenty-one nuclear loci with a median of 14 variable sites per locus were used to infer a phylogeny of Cichla.
- There was low resolution, allele sharing, and significant discordance among the observed gene trees.
- Three different analytical techniques recovered similar phylogenies of delimited species.
- Loci with few variable sites still contributed important phylogenetic signal.
- Confidence internals for node ages in the coalescent tree were wide despite many loci.

The inference of phylogenies of closely related species is obstructed by phenomena such as porous species boundaries and deep coalescence, and is often exacerbated by low levels of nucleotide variation among most loci surveyed in phylogenetic studies. We investigated the utility of twenty-one nuclear loci that had a range of 5-40 (median of 14) variable sites per locus to estimate the phylogeny of the genus Cichla, a group of 15 Neotropical cichlid fishes that began to diverge in the early to mid Miocene. We found that under a concatenated approach, the least variable loci, while contributing less to the overall phylogenetic signal (posterior node support), nevertheless provided information that increased support for the final tree. Moreover, this was not a result of misdirection by mutational noise, as the inference from all data was far superior to those from reduced datasets (those with more variable loci) in terms of the relative precision of posterior tree space. Phylogenetic methods that allowed each locus to have a separate genealogy, including Bayesian concordance analysis and a multispecies coalescent model, provided phylogenies that were also compatible with the concatenated tree in terms of the eight recently delimited species of Cichla, albeit with somewhat diminished support for some branches. In contrast, described species that still regularly exchange genes showed unstable relationships among analyses: not a surprising result from analyses that assume that gene tree heterogeneity results from incomplete lineage sorting and not gene flow. Importantly, we also observed that the confidence intervals for node ages in the coalescent analyses were quite wide, and likely susceptible to influence of the prior on node density (e.g. birth-death).