Phylogenetic Reconstruction of LDFP

We reconstructed a community phylogeny for LDFP using maximum likelihood (ML) and maximum parsimony (MP) algorithms. Three different marker combinations were examined for performance in phylogenetic reconstruction: rbcL+matK, rbcL+trnH-psbA, and rbcL+matK+trnH-psbA. For all combinations of markers, 142 species were included (Heterotrichum cynosum in the Asteraceae was excluded because only sequence data for trnH-psbA was available), with six sequences of rbcLa obtained from GenBank and used in conjunction with our barcode sequences. ML analyses were conducted using RAxML [30] via the CIPRES supercomputer cluster (www.phylo.org). The different locus combinations were partitioned for independent model assessment at each marker. For all combinations of markers a single most likely tree was estimated in addition to running 200–250 bootstrap replicates depending on the marker set. The same gene combinations were used in a MP using PAUP v.4.0 [31] and also run through a local cluster in which we implemented a modification of the parsimony ratchet [32] following Carolan et al. [33] (link). This resulted in a very large number of equally parsimonious trees, with over 350,000 trees produced for the rbcLa+trnH-psbA data set and over 250,000 produced for rbcLa+matK. The three gene matrix produced many fewer trees, approximately 25,000. For both ML and MP trees, a 50% majority tree was constructed and used to quantify overall levels of support for each node within the trees, the rates of well-supported monophylly for taxonomic hierarchies (genus, family, order) and concordance with expected topologies. Analyses were conducted both with and without use of a constraint tree, as described below.