The Phylogenomic Landscapes of Genes are Shallow
Speaker: Eric J. Alm
Lawrence David1 and Eric J. Alm1,2,3,4,5
Program in Computational and Systems Biology1 and Departments of Biological2 and Civil & Environmental3 Engineering, Massachusetts Institute of Technology, Cambridge, MA; The Virtual institute of Microbial Stress and Survival4, Berkeley, CA; The Broad Institute of MIT and Harvard 5, Cambridge, MA.
Gene trees are often discordant to species phylogenies, and this is especially true for bacterial genes. Whether this is the result of horizontal gene transfer, gene duplication and loss, or simply statistical uncertainty in phylogenetic reconstruction remains unclear. To investigate the origin of phylogenetic incongruence, we implemented an exact algorithm called AnGST (ANalysis of Gene and Species Trees) to infer maximum parsimony ‘phylogenomic’ reconstructions of the gene content of 60 prokaryotic genomes broadly distributed across the tree of life. In its simplest form, the algorithm computed the most parsimonious ‘phylogenomic’ reconciliation of gene and species trees (consisting of duplication, loss, and transfer events). A more sophisticated version allows for an ensemble of gene trees as input, which are combined to produce the optimal ‘hybrid’ reconciled tree giving the most parsimonious evolutionary reconstruction. We then tested the methods on their ability to reconstruct simulated evolutionary scenarios as well as on real data. Surprisingly, even the more sophisticated approach that considered uncertainty in gene trees generated a large number of evolutionary scenarios nearly equal in score, indicating that the ‘phylogenomic landscape’ of genes is relatively flat, with multiple distinct evolutionary scenarios equally compatible with observed trees. Our results suggest that the ‘phylogenomic method’ of tree reconciliation, considered the ‘gold standard’ for inferring gene histories, can lead to a high rate of false predictions even if trees are manually curated.