Phylogenomic inference in extremis: A case study with mycoheterotroph plastomes.

Abstract

Phylogenomic studies employing large numbers of genes, including those based on plastid genomes (plastomes), are becoming common. Nonphotosynthetic plants such as mycoheterotrophs (which rely on root-associated fungi for essential nutrients, including carbon) tend to have highly elevated rates of plastome evolution, substantial genome reduction, or both. Mycoheterotroph plastomes therefore provide excellent test cases for investigating how extreme conditions impact phylogenomic inference. We used parsimony and likelihood analysis of protein-coding gene sets from published and newly completed plastomes to infer the phylogenetic placement of taxa from the 10 angiosperm families in which mycoheterotrophy evolved. Despite multiple very long branches that reflect elevated substitution rates, and frequently patchy gene recovery due to genome reduction, inferred phylogenetic placements of most mycoheterotrophic lineages in DNA-based likelihood analyses are both well supported and congruent with other studies. Amino-acid-based likelihood placements are broadly consistent with DNA-based inferences, but extremely rate-elevated taxa can have unexpected placements-albeit with weak support. In contrast, parsimony analysis is strongly misled by long-branch attraction among many distantly related mycoheterotrophic monocots. Mycoheterotrophic plastomes provide challenging cases for phylogenomic inference, as substitutional rates can be elevated and genome reduction can lead to sparse gene recovery. Nonetheless, diverse likelihood frameworks provide generally well-supported and mutually concordant phylogenetic placements of mycoheterotrophs, consistent with recent phylogenetic studies and angiosperm-wide classifications. Previous predictions of parallel photosynthesis loss within families are supported for Burmanniaceae, Ericaceae, Gentianaceae, and Orchidaceae. Burmanniaceae and Thismiaceae should not be combined as a single family in Dioscoreales.