PHYLOGENETIC STUDIES OF A LARGE DATA SET .1. BAMBUSOIDEAE, ANDROPOGONODAE, AND POOIDEAE (GRAMINEAE)

Large data sets, with several hundred terminal taxa, are becoming increasingly common in phylogenetic studies, but are proving very difficult to analyze because existing algorithms cannot explore the enormous number of trees efficiently. This article presents the results of an ongoing project to carry out phylogenetic analyses on a data base with 760 terminal taxa, the genera of the grass family (Gramineae), initially scored for more than 400 morphological and anatomical characters. The approach consists of three steps: (1) Using a small number of highly consistent characters, determine which large groups are demonstrably monophyletic and which may be polyphyletic. Treat the large monophyletic groups as single terminal taxa, and focus on the overall structure of the entire group. This results in a tree that links the large monophyletic ''black boxes'' with smaller basal groups; the latter can then be taken as provisional outgroups. (2) Use the outgroups defined in step 1 to analyze the cladistic structure of the big monophyletic groups, with a much larger set of characters. (3) Use a cladistically-guided sample of basal taxa from each large clade to redo the family-level analysis. Kellogg and Campbell (1987) carried out step 1 for the grass family and defined four monophyletic groups (subfamilies) that were derived from within a highly polyphyletic assemblage of genera. This article reports on step 2, analyses of three of the four monophyletic groups, the pooid clade (I 84 genera), the bambusoid clade (166 genera), and the Andropogonodae plus Arundinelleae (121 genera). 150-220 characters per clade, a much larger number than commonly used in morphological studies, were chosen from the comprehensive database. The initial descriptions of characters, their division into states, and their application to particular genera were the result of 20 years of work on the family by one of us (LW). Subsequent choice of characters for cladistic analysis was done by the other author (EAK) using only the pattern of variation of the character rather than the morphological descriptor, thus eliminating possible bias from a priori ideas of a character's value. Each clade was analyzed in two ways, (1) with all terminal taxa for which there were adequate data, and (2) for only mono- and ditypic taxa; the latter analysis was to minimize the effect of possibly polyphyletic genera. In all cases, the reduced data set produced groups similar to those of the entire data set. The bambusoid clade consists of several well-defined subclades corresponding approximately to previously-recognized tribes. The relationships among the subclades are not resolved by these data. The Andropogonodae is made up of two major groups, an awned group and an awnless group; the latter includes taxa previously included in the Maydeae and also genera conventionally assigned to the Rottboelliinae. The pooid clade exhibits high homoplasy and no robust cladistic relationships. This is not likely to be caused by problems with generic circumscription, but may reflect extensive lateral gene flow (hybridization), rapid radiation followed by extensive anagenetic change, or true parallelism in morphological characters. The traditional tribes of the Pooideae are, with the exception of the Triticeae, apparently not monophyletic. Morphological cladograms are evaluated in light of data from molecular characters; while the results are generally consistent, there are too few molecular data yet to make meaningful comparisons.