It is hard work to reconstruct a phylogenetic tree. This point has been made many times (for example, see Felsenstein 2004), but bears repeating here. There are an enormous number of ways to connect a set of species by a phylogenetic tree – and the number of possible trees grows extremely quickly with the number of species. For example, there are about 5 × 1038 ways to build a phylogenetic tree1 of 30 species, which is many times larger than the number of stars in the universe. Additionally, the mathematical problem of reconstructing trees in an optimal way from species’ traits is an example of a problem that is “NP-complete,” a class of problems that include some of the most computationally difficult in the world. Building phylogenies is difficult.
The difficulty of building phylogenies is currently reflected in the challenge of reconstructing the tree of life. Some parts of the tree of life are still unresolved even with the tremendous amounts of genomic data that are now available. Accordingly, scientists have devoted a focused effort to solving this difficult problem. There are now a large number of fast and efficient computer programs aimed solely at reconstructing phylogenetic trees (e.g. MrBayes: Ronquist and Huelsenbeck 2003; BEAST: Drummond and Rambaut 2007). Consequently, the number of well-resolved phylogenetic trees available is also increasing rapidly. As we begin to fill in the gaps of the tree of life, we are developing a much clearer idea of the patterns of evolution that have happened over the past 4.5 billion years on Earth.
A core reason that phylogenetic trees are difficult to reconstruct is that they are information-rich2. A single tree contains detailed information about the patterns and timing of evolutionary branching events through a group’s history. Each branch in a tree tells us about common ancestry of a clade of species, and the start time, end time, and branch length tell us about the timing of speciation events in the past. If we combine a phylogenetic tree with some trait data – for example, mean body size for each species in a genus of mammals – then we can obtain even more information about the evolutionary history of a section of the tree of life.
The most common methods for reconstructing phylogenetic trees use data on species’ genes and/or traits. The core information about phylogenetic relatedness of species is carried in shared derived characters; that is, characters that have evolved new states that are shared among all of the species in a clade and not found in the close relatives of that clade. For example, mammals have many shared derived characters, including hair, mammary glands, and specialized inner ear bones.
Phylogenetic trees are often constructed based on genetic (or genomic) data using modern computer algorithms. Several methods can be used to build trees, like parsimony, maximum likelihood, and Bayesian analyses (see Chapter 2). These methods all have distinct assumptions and can give different results. In fact, even within a given statistical framework, different software packages (e.g. MrBayes and BEAST, mentioned above, are both Bayesian approaches) can give different results for phylogenetic analyses of the same data. The details of phylogenetic tree reconstruction are beyond the scope of this book. Interested readers can read “Inferring Phylogenies” (Felsenstein 2004), “Computational Molecular Evolution” (Yang 2006), or other sources for more information.
For many current comparative methods, we take a phylogenetic tree for a group of species as a given – that is, we assume that the tree is known without error. This assumption is almost never justified. There are many reasons why phylogenetic trees are estimated with error. For example, estimating branch lengths from a few genes is difficult, and the branch lengths that we estimate should be viewed as uncertain. As another example, trees that show the relationships among genes (gene trees) are not always the same as trees that show the relationships among species (species trees). Because of this, the best comparative methods recognize that phylogenetic trees are always estimated with some amount of uncertainty, both in terms of topology and branch lengths, and incorporate that uncertainty into the analysis. I will describe some methods to accomplish this in later chapters.
How do we make sense of the massive amounts of information contained in large phylogenetic trees? The definition of “large” can vary, but we already have trees with tens of thousands of tips, and I think we can anticipate trees with millions of tips in the very near future. These trees are too large to comfortably fit into a human brain. Current tricks for dealing with trees – like banks of computer monitors or long, taped-together printouts – are inefficient and will not work for the huge phylogenetic trees of the future. We need techniques that will allow us to take large phylogenetic trees and extract useful information from them. This information includes, but is not limited to, estimating rates of speciation, extinction, and trait evolution; testing hypotheses about the mode of evolution in a group; identifying adaptive radiations, key innovations, and other macroevolutionary explanations for diversity; and many other things.