bisection-reconnection) branch swapping, and the MulTrees option (saving all optimal trees) in
effect. A second heuristic search was conducted using 1000 random addition replicates with the
above settings and saving no more than 10 trees with a score greater than or equal to the best tree
score from the first replicate in the previous analysis. In all analyses, gaps were treated as
missing data. Strict consensus trees were generated from analyses with multiple equally
parsimonious trees. For all MP analyses, statistical support for nodes was estimated using
maximum parsimony bootstrap (BS) replicates (Felsenstein 1985). For the combined data set,
BS estimates were obtained using 1,000 replicates, each with 100 random taxon addition
replicates and saving no more than 1,500 trees per bootstrap replicate, TBR branch swapping and
the MulTrees option in effect.
All data were also analyzed by Bayesian inference (BI) methods with MrBayes v3.1.2
(Huelsenbeck and Ronquist 2001; Ronquist and Huelsenbeck 2003). An appropriate model of
evolution (under the AIC criterion) was selected for each data partition using the program
Modeltest v3.4 (Posada and Crandall 1998). All Bayesian analyses (individual loci and
combined data) were conducted while retaining the appropriate model for each data partition.
Markov Chain Monte Carlo was implemented with four heated chains and trees were sampled
every 1,000th generation for one million generations. The first 25 percent of the total number of
generations was discarded as bum-in. A 50 percent majority rule consensus tree was generated
from the remaining trees, in which the percentage of nodes recovered represented their posterior
probability (PP).
Congruent nodes resulting from the NJ, MP, and BI analyses of the combined molecular
data was used to assign isolates to a phylogenetic lineage (PL). Only isolates that fell within
clades of high support (BS value >70 and PP value > 95) were assigned to a PL.