Seabird Supertrees: Combining Partial Estimates of Procellariiform Phylogeny

—The growing use of comparative methods to address evolutionary questions has generated an increased need for robust hypotheses of evolutionary relationships for a wide range of organisms. Where a phylogeny exists for a group, often more than one phylogeny will exist for that group, and it is uncommon that the same taxa are in each of the existing trees. The types of data used to generate evolutionary trees can also vary greatly, and thus combining data sets is often difficult or impossible. To address comparative questions for groups where multiple phylogenetic hypotheses already exist, we need to combine different hypotheses in a way that provides the best estimate of the phylogeny for that group. Here, we combine seven seabird phylogenies (based on behavioral, DNA–DNA hybridization, isozyme, life history, morphological, and sequence data) to generate a comprehensive supertree for the Procellariiformes using matrix representation with parsimony. This phylogeny contains 122 taxa and represents a conservative estimate of combined relationships presented in the original seven source trees. We compared the supertree with results of a combined sequence data supermatrix for 103 seabird taxa. Results of the two approaches are broadly concordant, but matrix representation with parsimony provides a more comprehensive and more conservative estimate of the phylogeny of the group because it is less influenced by the largest of the source studies (which uses a single, relatively quickly evolving gene). Genetic data sets that can be combined in a supermatrix approach are currently less likely to be available than phylogenies that can be combined using some form of supertree approach. Although there are limitations to both of those approaches, both would be simpler if all phylogenetic studies made both their data sets and trees they generate available through databases such as TREEBASE. Received 8 December 2000, accepted 21 September 2001. RESUMEN.—El uso creciente de métodos comparativos para abordar preguntas evolutivas ha incrementado la necesidad de hipótesis robustas sobre relaciones evolutivas para una amplia variedad de organismos. Cuando existe una filogenia para un grupo, a menudo existirá más de una filogenia para ese grupo, y es poco frecuente que los mismos taxa estén presentes en cada uno de los árboles. Los tipos de datos usados para generar árboles evolutivos también pueden variar enormemente, por lo que generalmente es difı́cil o imposible combinar estos datos. Para responder a preguntas comparativas en grupos para los cuales existen múltiples hipótesis filogenéticas, necesitamos combinar diferentes hipótesis de manera que obtengamos la mejor estimación de la filogenia de estos grupos. Aquı́ combinamos siete filogenias (basadas en comportamiento, hibridización de ADN–ADN, isoenzimas, historia de vida, morfologı́a y datos de secuenciamiento) para generar un super-árbol integrador para los Procellariiformes usando representación de matrices combinada con parsimonia. Esta filogenia contiene 122 taxa y representa una estimación conservativa de las relaciones presentadas en los siete árboles originales combinados. Comparamos el super-árbol con los resultados de una matriz de datos de secuencias combinada para 103 taxa de aves marinas. A grandes rasgos, los resultados de las dos aproximaciones concuerdan, pero la representación de matrices combinada con parsimonia brinda una estimación más integral y más conservativa de la filogenia del grupo, debido a que está menos influenciada por uno de los estudios utilizados (la fuente con más datos, que a su vez usa un único gen de evolución relativamente rápida). Los juegos de datos genéticos que pueden ser combinados en una super-matriz están por lo general menos disponibles que las filogenias que pueden ser combinadas usando alguna aproximación de tipo super-árbol. Aunque existen limitaciones para estas dos aproximaciones, ambas serı́an más simples si todos los estudios filogenéticos pusieran a dispo1 E-mail: [email protected] January 2002] 89 Seabird Supertrees sición tanto sus juegos de datos como los árboles que generan a través de bases de datos como TREEBASE. RELATIONSHIPS WITHIN THE tubenose seabirds (Procellariiformes) are of general interest to many biologists, at least in part because they are a diverse and wide-ranging group, and because they are commonly found in most oceanic regions of the world. Nunn and Stanley (1998) generated a phylogeny for 85 species of tubenose seabirds for their discussion of effects of body size on rates of molecular evolution. By being such a diverse group (e.g. showing such a great range in body size), these birds allow comparative questions like that of Nunn and Stanley (1998) to be readily addressed when a phylogeny is available. As well as ranging greatly in size, the tubenose seabirds are particularly interesting because they are behaviorally and ecologically very diverse, and they provide a model system for investigating cospeciation (e.g. Paterson et al. 2000). In addition to Nunn and Stanley’s (1998) phylogeny for tubenose seabirds, there are several other phylogenies available for that group. Some of those phylogenies include taxa not present in Nunn and Stanley’s (1998) study, whereas some also disagree with the relationships found in their study. The storm petrels (Hydrobatidae), for example, unexpectedly do not form a monophyletic group in Nunn and Stanley’s (1998) phylogeny. Given that there are a number of phylogenies available for that group, precisely what the best estimate of the phylogeny is remains uncertain. Within the last few decades, there has been a dramatic increase in number of studies using phylogenies to address a wide range of issues. Those issues include, for example, behavior (e.g. Zyskowski and Prum 1999), biogeography (e.g. Kennedy and Spencer 2000), coevolution (e.g. Paterson et al. 2000), genetic systems (e.g. Cruickshank and Thomas 1999), language (e.g. Gray and Jordan 2000), rates of molecular evolution (e.g. Johnson and Sorenson 1998), speciation (e.g. Friesen and Anderson 1997), and taxonomy (e.g. Kennedy et al. 1999). Because biologists are becoming more convinced of the utility of taking a phylogenetic approach to questions they wish to address, robust hypotheses about phylogenetic relationships for the taxa of interest are required. Even with ongoing advances in molecular technology, phylogenetic hypotheses do not exist for most of the world’s taxa. When phylogenies do exist for a group, they will often not include all taxa of interest to the researcher. A single phylogeny for the taxa of interest is not always available, and studies have sometimes had to combine two or more phylogenies to obtain a tree that contains all the taxa (Sanderson et al. 1998). Kennedy et al. (1996), for example, had to combine four different source trees (two generated from morphological data and two from DNA–DNA hybridization data) to investigate homology of pelecaniform behaviors by mapping them onto the best estimate of that group’s phylogeny. A tree that results from combination of multiple source-tree topologies has been termed a ‘‘supertree’’ (Sanderson et al. 1998). Intuitively, the ideal way of combining several source trees would appear to be to combine all source phylogenies’ data matrices into a single ‘‘supermatrix’’ that could then be analyzed to estimate the phylogeny. Sanderson et al. (1998) note, however, that the supermatrix approach will often not be tenable because of the cost involved with filling in gaps in the data as well as difficulties associated with combining some types of data. They point out, for example, that DNA–DNA hybridization data would not be able to be included in a supermatrix, and that homologizing characters would become increasingly difficult as the size of the matrix increased as more distantly taxa were added (Sanderson et al. 1998). The supertree approach offers an alternative to the supermatrix approach. One method for constructing supertrees is matrix representation with parsimony (MRP; Baum 1992, Ragan 1992). Matrix representation with parsimony converts topologies of individual source trees into a data matrix (for a general explanation, see Sanderson et al. 1998). Once matrices for each of the source trees are combined, supertrees can be found using parsimony analysis. Because matrices are derived from the source trees’ topologies, MRP allows different data types (e.g. sequences, morphology, behavior, allozymes, DNA–DNA hybridization) to be combined (Bininda-Emonds and Bryant 1998). 90 [Auk, Vol. 119 KENNEDY AND PAGE