"PRINCIPLES OF PHYLOGENETICS: ECOLOGY AND EVOLUTION"Integrative Biology 200B! ! ! ! ! ! ! ! Spring 2009University of California, Berkeley D.D. AckerlyFeb. 3, 2009. Qualitative character evolution within a cladogram – parsimony methods Assigned reading: Maddison WP, Maddison DR (2000) MacClade 4 (pdf manual), Sunderland MA: Sinauer, chapters 3, 4 and 22 (equivalent to chapters 4, 5 and 17 in published MacClade 3 manual).Maddison WP, Slatkin M (1991) Null models for the number of evolutionary steps in a character on a phylogenetic tree. Evolution 45: 1184-1197.! In this portion of the class, we are interested in examining how discrete-state characters evolve on a tree individually and together. These are characters that meet the 'discrete-state' criteria for taxonomic characters. We will focus on binary traits (those with just two states); multistate characters are often generalizations of the binary case but can get more complicated, especially for the study of trait correlations. Other quantitatively varying characters can also be studied, but they require different methods that we'll cover later.! When we focus on the question of character evolution, we will assume that we have already obtained a phylogeny (with possible uncertainties). As we map the history of particular characters, we are not reevaluating the underlying phylogenetic hypothesis. Some have argued that you must not include a character for phylogeny reconstruction and then also map its ancestral states to test evolutionary hypotheses. What do you think?! Mapping of character states on a tree is an essential starting point for comparative methods related to phenotypic traits. The analysis of ancestral/derived states is central to phylogenetic approaches to questions of adaptation, divergent/convergent evolution, and correlated evolution. A typical adaptive hypothesis posits that trait X is an adaptation to selective pressure Y. For example: tough, evergreen leaves, termed sclerophylls, are an adaptation to semi-arid environments. This hypothesis may be tested in various ways, but one of the key predictions is that lineages with sclerophyll leaves first encountered semi-arid environments, and then evolved sclerophylly. If they evolved sclerophylly before encountering the hypothesized selective environment, then this trait must have evolved (and thus be an adaptation) in response to some other factor. This is one type of question that would lead you to conduct a comparative test.I. Ancestral states! The simplest approach to mapping ancestral states is the parsimony principle: we seek the reconstruction that requires the fewest evolutionary 'steps' or transitions between character states. We don't know that evolution proceeded in this way, but the basic argument for parsimony is that we should not assume any additional evolutionary events, beyond the minimum number necessary to explain the observed patterns. After calculating the ancestral states for a character, we can then test a number of hypotheses regarding the number of steps, their distribution on a tree, the polarity of character change, and the sequences or associations of change in two or more characters.The exact algorithm for calculating ancestral states by parsimony depends on the type of character and the assumed 'cost' of transition among different states:1) Unordered traits have 2 or more states, and all transitions require only a single step;2) Ordered traits have 3 or more states on an ordinal scale, and the number of steps is equal to the difference in state values (i.e., from 1 to 3 requires 2 steps);3) Dollo traits: change is only allowed from ancestral to derived state (reversals require 'infinite' steps);4) Arbitrary step matrices allow user to assign any"desired cost structure to transitions.In many cases, most parsimonious reconstructions (MPR) can be estimated visually, though it is easy to miss alternative equally parsimonious reconstructions when there are several gains and losses. To illustrate the mechanics of parsimony reconstruction, the formal algorithm for Fitch parsimony for unordered states is provided on the last page. Note: In contrast with the maximum likelihood methods we will study later in the class, branch lengths are not incorporated in parsimony algorithms. One can calculate a branch length after the fact based on the number of characters that change on a branch, but they are not used in the process of determining the MPR.II. Is a trait ‘conserved’? Null models for the number of changes on a tree After reconstructing the evolutionary history of a trait, you might wish to test a hypothesis about whether the trait has undergone an ‘unusually’ small or large number of transitions between alternative states. This is the question of phylogenetic signal: if the trait exhibits a small number of transitions, then it will be conserved over large portions of the tree and close relatives will tend to exhibit the same state (high phylogenetic signal = conserved trait). If the trait exhibits a large number of changes, then close relatives will frequently have different states (low phylogenetic signal = divergent/convergent trait). Testing the hypothesis of phylogenetic signal requires that we compare the observed number of transitions to some null hypothesis, and ask whether the data is improbable, and thus significantly different from, the null. In this case, an appropriate null model could be constructed by asking how many changes would be expected in the trait if there was no phylogenetic signal, i.e. states of the trait were randomly arranged among the taxa so close relatives share the same state only by chance (not due to inheritance from a shared common ancestor) (Maddison and Slatkin 1991). This is an example of a non-parametric test (i.e., there is no elegant solution that can be derived analytically from the underlying statistical parameters). The significance of non-parametric tests can be solved using Monte Carlo methods, which simply means brute force computer simulation under the null model. The basic logic of a Monte Carlo test is as follows:I. Choose a test statistic, T, in this case the number of evolutionary steps on a tree.II. Calculate T for the observed data; we’ll call this T0.III. Choose a null model to obtain random permutations of the data in which the pattern of interest is removed, but all other relevant aspects of the data are maintained. In this case, we can permute (= sample
View Full Document
Unlocking...