Robustness of RADseq for evolutionary network reconstruction from gene trees

Although hybridization has played an important role in the evolution of many species, phylogenetic reconstructions that include hybridizing lineages have been historically constrained by the available models and data. Recently, the combined development of high-throughput sequencing and evolutionary network models offer new opportunities for phylogenetic inference under complex patterns of hybridization in the context of incomplete lineage sorting. Restriction site associated DNA sequencing (RADseq) has been a popular sequencing technique for evolutionary reconstructions of close relatives in the Next Generation Sequencing (NGS) era. However, the utility of RADseq data for the reconstruction of complex evolutionary networks has not been thoroughly discussed. Here, we used new molecular data collected from diploid perennialMedicagospecies using single-digest RADseq to reconstruct evolutionary networks from gene trees, an approach that is computationally tractable with datasets that include several species and complex patterns of hybridization. Our analyses revealed that complex network reconstructions from RADseq-derived gene trees were not robust under variations of the assembly parameters and filters. Filters to exclusively select loci with high phylogenetic information created datasets that retrieved the most anomalous topologies. Conversely, alternative clustering thresholds or filters on the number of samplesperlocus affected the level of missing data but had a lower impact on networks. When most anomalous networks were discarded, all remaining network analyses consistently supported a hybrid origin forM. carstiensisandM. cretacea.