Inferring the shallow phylogeny of true salamanders (Salamandra) by multiple phylogenomic approaches

Abstract

The rise of high-throughput sequencing techniques provides the unprecedented opportunity to analyse controversial phylogenetic relationships in great depth, but also introduces a risk of being misinterpreted by high node support values influenced by unevenly distributed missing data or unrealistic model assumptions. Here, we use three largely independent phylogenomic data sets to reconstruct the controversial phylogeny of true salamanders of the genus Salamandra, a group of amphibians providing an intriguing model to study the evolution of aposematism and viviparity. For all six species of the genus Salamandra, and two outgroup species from its sister genus Lyciasalamandra, we used RNA sequencing (RNAseq) and restriction site associated DNA sequencing (RADseq) to obtain data for: (1) 3070 nuclear protein-coding genes from RNAseq; (2) 7440 loci obtained by RADseq; and (3) full mitochondrial genomes. The RNAseq and RADseq data sets retrieved fully congruent topologies when each of them was analyzed in a concatenation approach, with high support for: (1) S. infraimmaculata being sister group to all other Salamandra species; (2) S. algira being sister to S. salamandra; (3) these two species being the sister group to a clade containing S. atra, S. corsica and S. lanzai; and (4) the alpine species S. atra and S. lanzai being sister taxa. The phylogeny inferred from the mitochondrial genome sequences differed from these results, most notably by strongly supporting a clade containing S. atra and S. corsica as sister taxa. A different placement of S. corsica was also retrieved when analysing the RNAseq and RADseq data under species tree approaches. Closer examination of gene trees derived from RNAseq revealed that only a low number of them supported each of the alternative placements of S. atra. Furthermore, gene jackknife support for the S. atra - S. lanzai node stabilized only with very large concatenated data sets. The phylogeny of true salamanders thus provides a compelling example of how classical node support metrics such as bootstrap and Bayesian posterior probability can provide high confidence values in a phylogenomic topology even if the phylogenetic signal for some nodes is spurious, highlighting the importance of complementary approaches such as gene jackknifing. Yet, the general congruence among the topologies recovered from the RNAseq and RADseq data sets increases our confidence in the results, and validates the use of phylotranscriptomic approaches for reconstructing shallow relationships among closely related taxa. We hypothesize that the evolution of Salamandra has been characterized by episodes of introgressive hybridization, which would explain the difficulties of fully reconstructing their evolutionary relationships.