414 CHAPTER 16Likelihood and Bayesian inference have several advantages over parsimony.
They are more robust to homoplasy; unlike parsimony, they estimate the lengths
of the branches (in terms of time or the number of changes that occurred); they tell
us the relative statistical support for different trees (rather than just identifying the
most likely tree); and they can be used to simultaneously estimate other quantities
(such as substitution rates). For those reasons, likelihood and Bayesian inference
are the most widely used approaches for estimating phylogenetic trees from DNA
sequences. But they also have limitations. Like any other method, both likelihood
and Bayesian inference can give erroneous results if the assumptions are wrong.
A second problem is that they are difficult to use with morphological data. Fortu-
nately, all three methods give similar results in many cases.
Once we have estimated a phylogeny, an important question becomes how
confident we are in that estimate. When we use likelihood or Bayesian inference,
the relative confidence in two alternative phylogenies can be calculated directly.
An alternative method called bootstrapping is often used with parsimony. The
approach here is to randomly discard some of the data and then reestimate the
phylogeny. After doing that many times, if we consistently get the same phylogeny,
then we become confident that is the true phylogeny, because multiple, somewhat
different data sets yield the same answer. Bootstrapping is often used to assess the
degree of confidence in the individual branches of a tree.
There are several ways to test the validity of phylogenetic methods. One is to
apply them to phylogenies that are known with certainty: evolutionary histories that
have been simulated on a computer, allowing the lineages to branch and their char-
acters to change according to various models of the evolutionary process. The inves-
tigators then see whether or not a phylogenetic method using the final characters
of the simulated lineages gives an accurate history of their branching. Another test
is to apply the method to data on experimental populations of real organisms that
have been split into separate lineages by investigators (creating artificial branch-
ing events) and allowed to evolve. For example, David Hillis and coworkers suc-
cessively subdivided lineages of T7 bacteriophage that accumulated DNA sequence
differences rapidly over the course of about 300 generations [8, 18]. The investiga-
tors then scored the eight resulting lineages for sequence differences and performed
a phylogenetic analysis of the data. For this many populations, there are 135,135
possible dichotomous trees (in which each lineage branches into two others), but
the phylogenetic analysis correctly found the one true tree. Finally, throughout sci-
ence, the chief way of confirming a hypothesis is to see if it agrees with multiple, indepen-
dent sources of data. For phylogenetic hypotheses, these sources might be different,
unlinked gene sequences, or morphological features and DNA sequences, which
evolve largely independently of each other and thus provide independent phyloge-
netic information. These two kinds of data usually yield similar estimates of phylog-
eny. For instance, the phylogenetic relationships among higher taxa of vertebrates
inferred from DNA sequences are almost all the same as those inferred from mor-
phological features (FIGURE 16.13).PHylogENIES fRoM PHENoTyPES Throughout most of the history of sys-
tematics, relationships were inferred using morphological data. The vertebrate
phylogeny in Figure 16.13 is one of many examples that have largely stood the
test of time and the arrival of new DNA sequence data. Even though DNA
sequences are now the main source of phylogenetic data, there are situations
when DNA sequences are not available and morphology is the only source of
information. Nowhere is that more true than in understanding the relationships
of fossilized species to one another and to living species. Careful study is needed
to discriminate distinct characters and to distinguish derived from ancestral
character states.16_EVOL4E_CH16.indd 414 3/22/17 1:33 PM