600 
Telopea Vol. 6(4): 1996 
The characters used are discussed in detail in the Appendix, and the distribution of 
the states given in Table 1. 
Approximately 70 species are members of Rytidosperma sensu Zotov; of these, both 
morphological and anatomical data are available for 54 species, as well as for the 
species of Erythranthera and Pyrrhanthem. Closely related genera are represented by 
several species each (Karroochloa, hvo of four; Schismus, two of five; and Triboliuiii, 
three of ten) representing the variation within these genera. The 23 species of Danthoiiia 
are represented by six species. The tree was rooted to Pentaschistis curvifolia. 
Pentaschistis has haustorial synergids (Philipson & Connor 1984, Verboom et al. 1994), 
which is a synapomorphy for the Danthonieae, but is not a member of the Danthonin- 
Rytidosperma clade as it lacks the distinctive obovate caryopsis: it is thus a suitable 
outgroup taxon (Nixon & Carpenter 1993). 
Parsimony analysis was performed using Hennig86 (Farris 1988), run as a subroutine 
of DADA (Nixon 1993), and PAUP vs. 3.1 (Swofford 1993). Cladograms were prepared 
using CLADOS (Nixon 1992). Several analyses were performed. An initial analysis 
included all 69 species, too large a set for detailed analysis. Initial trees for branch 
swapping were generated by multiple random input, using both PAUP and DADA. 
The set of shortest trees was then branch-swapped using TBR in PAUP and bb* in 
Hennig. However, in both searches the available memory was exhausted before 
completion of the search, so there might be shorter, as yet undiscovered, trees. The 
second analysis included a subset of 33 species selected to reflect variation within 
the group; these species were chosen to represent groups of similar species. 
Successive weighting (Carpenter 1988) was used to choose among the set of most 
parsimonious trees. However, the results produced were generally at variance with 
the set of minimally parsimonious trees, and consequently these results were used to 
provide an assessment of the robustness of the various nodes. In addition, a bootstrap 
analysis with 100 replicates was conducted. Tree topologies and the costs of modifying 
tree topologies were explored using MacClade vs. 3 (Maddison & Maddison 1992); 
Characters were mapped on the strict consensus tree, using Accelerated 
Transformation (Swofford & Maddison 1987), which prefers reversals to parallelisms 
when both optimisations are equally parsimonious. Anderberg and Tehler (1990) 
argued that it is illogical to plot characters on a consensus tree, as consensus trees 
are derived from the set of most parsimonious trees. However, this is not the issue: 
the characters are plotted on the the tree to illustrate which characters can be 
interpreted as synapomorphies for which clades. 
Results 
Analysis 1 could not be completed, due to lack of memory. Nine hundred and nine 
trees of length = 332 steps, consistency index = 0.18 and retention index = 0.62 were 
found. The strict consensus tree is given in Fig. 1. Analyses with different starting 
trees, and with starting trees designed from the results of later analyses, produced 
essentially the same strict consensus tree, but with variation indicating that the set 
of trees located is somewhat different. The second analysis found four trees 
(length = 190, consistency index = 0.295 and retention index = 0.561), with the only 
polychotomy being within Danthoiiia (Fig. 2). 
Successive weighting did not lead to greater resolution, but provided some indication 
of the support for the various nodes (Fig. 3). The bootstrap results confirmed the 
low consistency index: very few nodes were retrieved in more than 50% of randomised 
data sets (Fig. 3). 
