Fig. 6 Phylogenomic trees estimated using ASTRAL based on 122 nuclear single-copy genes. Branch support was calculated using local posterior probabilities (LPP), and the LPP value less than 1.0 are shown on branches. Fabids as defined here are shown with green branches; the expanded malvid clade is shown with blue branches.
Coalescent simulations demonstrated a significant role of the ILS in the backbone of the superrosids (Fig. S5 ). The branch subtending (core rosids + Saxifragales), the critical node for examining the relative placement of Vitales versus Saxifragales with respect to the core rosids, had a clade probability of 0.31. This value is close to the theoretical minimum probability (0.33 for the species tree clade under the ILS in the three-taxon case (Pamilo & Nei, 1988 ), suggesting that this branch is in the anomaly zone (Degnan & Rosenberg, 2006 ). Therefore, these results are consistent with the primary role of the ILS in conflicting interpretations of superrosid relationships. However, the expected clade probabilities were lower (≤ 0.1) in the backbone of the rosids, suggesting high discord compared to ILS expectations. Statistical tests suggest that ILS alone cannot explain gene tree incongruence; observed gene tree Robinson-Foulds distances were significantly higher than expected (p < 1e-20), and the observed gene tree clade probabilities were lower than the ILS expectation (p = 0.0043). These results are therefore consistent with a potential role for early reticulation in the radiation of rosids.