Phylogenomic analyses clarify the pattern of evolution of Adephaga (Coleoptera) and highlight phylogenetic artefacts due to model misspecification and excessive data trimming

Abstract

Adephaga is the second largest suborder of Coleoptera and contains aquatic and terrestrial groups that are sometimes classified as Hydradephaga and Geadephaga, respectively. The phylogenetic relationships of Adephaga have been studied intensively, but the relationships of the major subgroups of Geadephaga and the placement of Hygrobiidae within Dytiscoidea remain obscure. Here, we infer new DNA-hybridization baits for exon-capture phylogenomics and we combine new hybrid-capture sequence data with transcriptomes to generate the largest phylogenomic taxon sampling within Adephaga presented to date. Our analyses show that the new baits are suitable to capture the target loci across different lineages of Adephaga. Phylogenetic analyses of moderately trimmed supermatrices confirm the hypothesis of paraphyletic ‘Hydradephaga’, with Gyrinidae placed as sister to all other families as in morphology-based phylogenies, even though quartet-concordance analyses did not support this result. All analyses conducted with site-heterogeneous models suggest Trachypachidae as sister to a clade Carabidae + Cicindelidae in congruence with results from morphological studies. Haliplidae is inferred as sister to Dytiscoidea, while a clade of Noteridae (+ most likely Meruidae) is inferred as sister to all remaining Dytiscoidea. A strongly supported clade Hygrobiidae + (Amphizoidae + monophyletic Aspidytidae) is inferred in most analyses of moderately trimmed supermatrices when a site-heterogeneous model is used. In general, we find that stringent trimming of supermatrices results in reduced deviation from model assumptions but also in reduction of phylogenetic information. We also find that site-heterogeneous C60 models provide greater stability of phylogenetic relationships of Adephaga across analyses of different amino-acid supermatrices than site-homogeneous models. Thus, site-heterogeneous C60 models can potentially reduce incongruence in phylogenomics. Lastly, we show that gene-tree errors are prominent in the data, even after sub-sampling genes to reduce these errors, but we also show that subsampling genes based on the likelihood mapping criterion in summary coalescent analyses results in higher topological congruence with the concatenation-based tree. Overall, our analyses demonstrate that moderate alignment trimming strategies, application of site-heterogeneous models and mitigation of gene-tree errors should be routinely included in the phylogenomic pipeline in order to more accurately infer the phylogeny of species.