Circumscription and biogeographic patterns in the eastern North American-east Asian genus Stewartia (Theaceae: Stewartieae): Insight from choloroplast and nuclear DNA sequence data

Castanea, Sep 2002 by Prince, Linda M

Computational Methods

All analyses and tests were conducted in PAUP* Version 4.0b8 (beta test version; Swofford 2001) and run to completion unless otherwise noted. Maximum Parsimony (MP) used Fitch (1971) equal weight, branch and bound analyses. Parsimony bootstrap analyses (Felsenstein 1985) used 10 random addition replicates, TBR branch swapping, saving a maximum of 10 trees for 1,000 bootstrap replicates. The partition homogeneity tests were conducted with 10,000 replicates of 1 random addition (TBR branch swapping, saving one tree per replicate). Combined data matrix analyses were conducted as described for the individual data sets above.

Maximum likelihood (ML) scores were calculated for one of the MP trees under one of the simpler models, the Jukes-Cantor (Jukes and Cantor [JC] 1969) model, and the more complex model, the General Time Reversible model ([GTR] with gamma rate estimation and proportion of invariant sites estimation; Lanavae et al. 1984, Rodriguez et al. 1990) for each data partition with subsequent successive approximation analyses (Swofford et al. 1996).

RESULTS

Statistics for all data partitions and analyses are summarized in Table 3. In general, tree topologies were identical for MP and ML analyses, regardless of the ML model selected. The only qualification to this statement is in reference to rooting of the ingroup relative to outgroup taxa. Alignment of the ITS matrix for outgroup taxa was ambiguous in several regions. Alternative alignments and exclusion of some characters shifted rooting within the ingroup, as did the use of alternative outgroup sequences. Rooting of the tree was also sensitive to the ML model selected for data analysis. Correct rooting is an extremely important issue for each of the questions this study wishes to address. Instability of the root in initial exploratory MP analyses (Prince 2000) prompted sequencing of the less variable chloroplast DNA regions included in this paper.

ITS

Branch and bound MP analyses of the 734 characters of the aligned ITS data matrix produced 120 shortest trees. The strict consensus tree is shown in Figure 2A with bootstrap values for branches >= 50% indicated. Strongest support was for the sister relationships of all Harta samples (99%), and for the monophyly of Stewartieae (100%). There was moderate support for a Stewartia gemmata S. monadelpha S. serrata S. sinensis clade (88%), and for the sister relationship of S. ovata and S. malacodendron to all Hartia samples (80%). Two of the sequences were incomplete, S. malacodendron [LP1241 and S. ovata [LP295], but the missing data did not result in an unexpected placement of the partially sequenced taxa.

ML analyses under the JC model produced a tree of -In = 2427.21469 which was similar in topology to the MP tree (not shown), but with higher resolution. The JC model rooted the tree in the same location as the MP analysis. The GTR G I model with all parameters estimated from the data (see Table 3 for parameter details) produced a tree of -In = 2293.53362. The topology was similar to the strict consensus of the MP analysis but with a different root location (see Figure 2). The GTR G I analyses divided the ingroup into two similar-sized Glades, one of Hartia S. ovata S. malacodendron and one of all other Stewartia samples.