Molecular and morphological datasets have similar numbers of relevant phylogenetic characters

Abstract

Most groupings on the tree of life were first inferred using morphological characters, which have been widely used since Darwin's time. Most of the million-plus described species are known only from morphology, and their inferred phylogenetic positions are based on these characters; with between 90 and 99 percent of life still to be formally recognised (usually initially from morphological features alone), it is likely that this situation will persist for some time. A recent paper (Scotland & al., 2003), however, downgrades the utility of morphology, because of (1) subjectivities in morphological character coding and (2) a paucity of informative characters. The first proposed reason is well founded. Unlike morphological studies, where there are subjectivities in atomising an organism into and in dividing continuous variation within a character into discrete character states (e.g., Wagner, 2001), in molecular studies nucleotide positions and character states can be recognised unambiguously, sequencing error and alignment issues notwithstanding (e.g., Lee, 2004). Scotland & al.'s second point, that the pool of morphological is much smaller than molecular characters, and often for robust phylogenetic resolution, is a common view (e.g., Hillis, 1987; Givnish & Sytsma, 1997) that will be assessed empirically here. There are, of course, increasing examples of large morphological datasets comprising several hundred informative (e.g., Lee & Scanlon, 2002; Ruta & al., 2003; Lambkin & al., 2003); the insufficient characters argument would not apply to these datasets, which contain as many informative as do datasets of complete genes routinely sequenced (e.g., 18S rRNA, rbcL, cyt b). However, many if not most morphological datasets are much smaller, comprising fewer than 50 characters. I will argue, however, that even these apparently small datasets often contain as much relevant phylogenetic signal as typical molecular datasets that have orders of magnitude more characters. If these arguments are true, morphology deserves to retain a central place in inferring phylogenies (Jenner, 2004). Consider a typical morphological dataset aimed at resolving relationships between several major lineages (traditionally termed families) of uncontested monophyly. These datasets will typically have families as basic taxa (OTUs), and contain only parsimony-informative that vary between families. Invariant characters, autapomorphies (uniquely derived traits) diagnosing individual families, and that vary only within a family would usually not be included (see Yeates, 1992). There might be relatively few in this morphological dataset, but every character would be informative in resolving relationships between these families. Now consider a molecular dataset for the