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

Large data sets, with several hundred terminal taxa, are becoming incr easingly common in phylogenetic studies, but are proving very difficul t to analyze because existing algorithms cannot explore the enormous n umber of trees efficiently. This article presents the results of an on going project to carry out phylogenetic analyses on a data base with 7 60 terminal taxa, the genera of the grass family (Gramineae), initiall y scored for more than 400 morphological and anatomical characters. Th e 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 monophylet ic groups as single terminal taxa, and focus on the overall structure of the entire group. This results in a tree that links the large monop hyletic ''black boxes'' with smaller basal groups; the latter can then be taken as provisional outgroups. (2) Use the outgroups defined in s tep 1 to analyze the cladistic structure of the big monophyletic group s, with a much larger set of characters. (3) Use a cladistically-guide d 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 d erived from within a highly polyphyletic assemblage of genera. This ar ticle reports on step 2, analyses of three of the four monophyletic gr oups, the pooid clade (I 84 genera), the bambusoid clade (166 genera), and the Andropogonodae plus Arundinelleae (121 genera). 150-220 chara cters per clade, a much larger number than commonly used in morphologi cal studies, were chosen from the comprehensive database. The initial descriptions of characters, their division into states, and their appl ication to particular genera were the result of 20 years of work on th e family by one of us (LW). Subsequent choice of characters for cladis tic analysis was done by the other author (EAK) using only the pattern of variation of the character rather than the morphological descripto r, thus eliminating possible bias from a priori ideas of a character's value. Each clade was analyzed in two ways, (1) with all terminal tax a for which there were adequate data, and (2) for only mono- and dityp ic taxa; the latter analysis was to minimize the effect of possibly po lyphyletic 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 previ ously-recognized tribes. The relationships among the subclades are not resolved by these data. The Andropogonodae is made up of two major gr oups, an awned group and an awnless group; the latter includes taxa pr eviously included in the Maydeae and also genera conventionally assign ed 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 later al gene flow (hybridization), rapid radiation followed by extensive an agenetic change, or true parallelism in morphological characters. The traditional tribes of the Pooideae are, with the exception of the Trit iceae, apparently not monophyletic. Morphological cladograms are evalu ated in light of data from molecular characters; while the results are generally consistent, there are too few molecular data yet to make me aningful comparisons.