Identifying and Removing Fast-Evolving Sites Using Compatibility Analysis: An Example from the Arthropoda

Abstract

Felsenstein (1978) first recognized that long branch attraction (LBA) can seriously affect the accuracy of phylogenetic reconstruction. Although Felsenstein specifically addressed this problem in the cases of parsimony and clique analyses, it is now known that LBA can affect any tree reconstruction method, including maximum likelihood (ML) and Bayesian approaches. However, LBA is only a problem for distance, ML, and Bayesian methods when the assumed substitution model is underparameterized, i.e., when it is unrealistically simple (Swofford et al., 2001; Lemmon and Moriatry, 2004). LBA should therefore be avoidable by analyzing the data using ML, Bayesian, or distance methods under the best-fitting substitution model (providing this is a good approximation of the true substitution model). However, ML and (although to a much lesser extent) Bayesian analyses are time consuming, whereas many widely used implementations of distance methods (for example Kumar et al., 2001) do not allow the specification of complex substitution models. Accordingly, estimating the relationships of fast-evolving species still represents one of the most serious problems of molecular phylogenetics. Strategies for dealing with LBA that do not necessarily rely on the use of probabilistic methods or complex evolutionary models have been suggested. These strategies can be of special utility when parsimony or distance methods are used. These include (1) increasing the taxon sampling (Hendy and Penny, 1989; Hillis, 1996; Rannala et al., 1998; Pollock et al., 2002; Poe, 2003); (2) optimal outgroup selection (Wheeler, 1990); and (3) sampling strategies specifically targeting slowly evolving species (e.g., Aguinaldo et al., 1997). Unfortunately, none of these strategies is universally applicable. For example, increasing the taxon sampling can (in some cases) exacerbate LBA (Kim, 1996; Poe and Swofford, 1999; Poe, 2003), phylogenetic uncertainty can prevent the selection of adequate outgroups, and for certain groups, it is possible that no slowly evolving species can be identified. An alternative approach to countering LBA that does not necessarily rely on the use of complex substitution models is to identify and remove fast evolving sites, which are expected to contribute substantially to it (e.g., Brinkmann and Philippe, 1999; Hirt et al., 1999). Particularly, Brinkmann and Philippe (1999) proposed a simple parsimony-based method, christened slow-fast (SF), for identifying (and then removing) fast evolving sites from an alignment. SF, as well as other methods that will not be considered in detail here (e.g., Hirt et al., 1999), can be especially useful when taxon sampling is limited, if close outgroups are unavailable, and in all cases when fast evolving species are included in the investigation. Here, the use of alternative, compatibility-based methods (see Felsenstein, 2003; Semple and Steel, 2003; Pisani, 2002; Wilkinson, 2001; Meacham and Estabrook, 1985, for an introduction) for identifying fast-evolving sites is proposed and illustrated using arthropod data in a taxonomic congruence (Miyamoto and Fitch, 1995) context. Unlike the parsimony-based SF, the methods here proposed are topology independent, allowing for their application in cases where SF cannot be applied (see below).