Can molecular data really shake the mammalian tree? The quality of evidence from complete mitochondrial DNA sequences

In recent years mtDNA sequences have been used to resolve long-standing and sometimes controversial phylogenetic problems. Among the most intractable have been those of the eutherian mammals that evolved with at least two episodes of rapid divergence. These gave rise to modern orders at least 65 million years ago and then 25 million years later to recognizably modern families. The difficulty arises because sequences must contain enough change to reveal closely spaced branching events without accumulating homoplasies that obscure the true branching pattern; thus sequences must be conserved and they also must be relatively long.; In principle, mtDNA sequences are ideal for these kinds of analyses because the genes evolve at different rates but have a single evolutionary history. It has become clear, however, that single genes (<2-kb) can generate different trees. To determine the reason for the conflicting trees, I use maximum likelihood methods to estimate topologies and branch lengths from 19 sequences with known subordinal relationships and comprising all 13 mtDNA protein-coding genes. I compare these optimal estimates from single genes (0.2–1.8-kb) and from seven combinations of genes (2–3.5 kb) and use maximum likelihood tests (Chapter 1) and spectral analytic (Chapter 2) methods to show that random error, rather than systematic error explains conflicting gene trees. I show that the proper choice of genes to sequence enables reliable estimation of relationships among mammalian families. Even the entire 11-kb sequence, however, cannot to unambiguously resolve ancient interordinal relationships among these 19 eutherians.; Moreover, I use real (Chapters 3) and simulated sequences (Chapter 4) to demonstrate that a single model cannot adequately describe sequence evolution among these species. Instead, the best strategy is to accommodate significant interdependence among model parameters (base frequency, substitution type, invariant sites and gamma rate heterogeneity) with maximum likelihood estimates (MLE) that are fully-stratified by codon position. Failure to do so results in misleading branch length and topology estimates and provides a false level of confidence that the true tree has been identified.; Finally, I use the same techniques on 56 complete mtDNA sequences (Chapter 5) to confirm that densely sampled, complete sequences (10.6-kb) resolve relationship among modern eutherian families with high confidence. The same sequences, however, are almost certainly too short and uninformative to describe unambiguously ancient evolution among modern mammalian orders.