 
Summary: Phylogenetic analyses
Phylogenetic analyses were performed with PAUP v. 4.0b10 (ref. 27) and MrBayes v. 2.01
(ref. 28). Sites of known RNA editing were excluded from the analyses. Maximum
likelihood trees were constructed with a HKY85 substitution model; trees in Fig. 2 used a
transitiontotransversion ratio of 2.0. For alternative topology tests (see below for
ShimodairaHasegawa tests, and Supplementary Information for parametric bootstrap
analyses), transitiontotransversion ratios and gamma distribution parameters were first
estimated from the data. Bootstrap support values are from maximumparsimony
analyses of 1,000 bootstrap replicates and 100 random addition replicates. Posterior
probability for clade support was estimated with Markov chain Monte Carlo as
implemented in MrBayes. Four Markov chains were run for 105
to 106
generations after
burnin, using random initial trees and a general timereversible (GTR) codonsite
specific substitution model for coding sequences and GTR with gamma distribution for
noncoding sequences. The ShimodairaHasegawa29
test favoured the horizontal
placements shown in the unconstrained trees of Fig. 2 over alternative topologies based on
vertical transmission: first, the unconstrained rps2 tree (Fig. 2a) was favoured over the
constrained tree grouping Actinidia with Grevillea/Platanus as a monophyletic group
