Combining Data in Phylogenetic Systematics: An Empirical Approach Using Three Molecular Data Sets in the Solanaceae

?Data from sequences of two chloroplast DNA (cpDNA) genes, ndhV and rbcL, and from restriction site mapping studies of the entire chloroplast genome were analyzed for the same set of 17 species of Solanaceae (potato/tomato family) and the outgroup Ipomoea (Convolvulaceae, morning glory family). ndhV is a more rapidly evolving gene and yields more phy? logenetic information than does rbcL. The results of analyses of each data source independently are congruent with respect to certain major features of Solanaceae phylogeny but are in conflict with respect to the branching order within the monophyletic subfamily Solanoideae. This conflict resides in a portion of the phylogeny where relatively few characters are available in any of the three data sets. Combining the data into a single analysis yielded a more fully resolved tree, in which some clades are supported that are unique to each individual data set. However, in each such case, these resolved clades also were obtained when the two alternative data sets were analyzed together. The analysis of the combined data set provides substantial support for in? creased resolution within the Solanoideae and for the rest of the tree. The X = 12 dade, so called because of a base chromosome number synapomorphy, comprises the subfamily Solanoide? ae, tribe Anthocercideae, and part of tribe Nicotianeae (Nicotiana) and is the most well-supported dade within the family. Solanaceae; rbcL; ndhV; chloroplast DNA; congruence; combining data sets. Analyzing a single, large data set de? rived by combining different data sets ver? sus analyzing the individual data sets sep? arately and then examining the resulting trees for congruence is a topic of contem? porary debate in systematics. It has been argued (e.g., Swofford, 1991) that incon? gruence in results from analyzing logically separable data sets may lead to insight re? garding nonindependence of characters (e.g., convergent evolution may affect one gene sequence and not another), a viola? tion of an assumption of phylogenetic analysis. The potential for bias introduced by nonindependence argues powerfully for analyzing data sets separately. If data from two sources provide strong support for conflicting hypotheses of relationship, character nonindependence may be re? sponsible. Advocates of the principle of 1 E-mail: olmstead@spot.colorado.edu. 2 Present address: Laboratory of Molecular System? atics, National Museum of Natural History, Smith? sonian Institution, Washington, D.C. 20560, USA. total evidence (e.g., Kluge, 1989; Dono? ghue and Sanderson, 1992) argue that the combined inference from many sources of data will yield the strongest estimate of phylogeny. If conflict in results from sep? arate data sets exists only in portions of the resulting trees where character support is weak, then little evidence exists for char? acter nonindependence and stochastic pro? cesses provide an equally likely explana? tion. This is the most appropriate situation for combining data into one analysis (de Queiroz, 1993). There is no way to establish what con? stitutes the best evidence of phylogenetic relationship without first knowing the pat? tern of evolutionary history within a group (e.g., Hillis et al., 1992). Several methods have been used, individually or in concert, to assess the efficacy of systematic data within a cladistic parsimony framework. The number of characters and the congru? ence among characters within a data set will determine the degree of resolution possible from any source of data (Swofford, 1991). A data source that yields many char-