A South American forest legacy underlies historical biogeography and biome evolution in the Neotropical Tabebuia alliance (Bignoniaceae)

Background and aims - The Tabebuia alliance is a widespread radiation of Neotropical trees and shrubs spanning South America, Mesoamerica, the Caribbean and southernmost North America. I hypothesised when the sampled radiation originated, whether its deepest geographic history was centred in South America, and whether transitions among humid forest, seasonally dry tropical forest and savanna are robust to uncertainty in the discrete-state transition model. Materials and methods - I analysed a fixed time-scaled phylogeny with 69 ingroup terminals. Divergence times were estimated in BEAST 2 from two independent 100-million-generation chains. Historical geography was coded as South America, Central America plus Mexico, the Caribbean and North America and analysed with unstratified DEC, DIVALIKE and BAYAREALIKE in BioGeoBEARS. Biome occupancy was coded as humid forest, seasonally dry tropical forest and savanna; polymorphic observations were treated as sets of compatible tip states and fitted under ER, SYM and ARD in corHMM. Because biome models had similar AICc support, 1,000 stochastic character maps were generated under each model and key quantities were model-averaged. Sensitivity analyses tested plausible recoding of low-confidence biome assignments and an alternative equal root-state prior. Key results - The crown of the Tabebuia alliance was dated to 36.06 Ma, with a central 95% interval of 28.49-45.39 Ma. DEC and DIVALIKE together accounted for essentially all geographic-model weight, and every fitted geographic model placed virtually all root probability on ranges containing South America. Biome-model selection was unresolved: SYM, ER and ARD had AICc weights of 0.377, 0.334 and 0.289. Under the primary Yang/stationary treatment, the model-averaged root probabilities were 0.546 for humid forest, 0.424 for seasonally dry forest and 0.030 for savanna. Forest states accounted for 93.3-97.5% of reconstructed evolutionary time. Sensitivity analyses retained the same qualitative result: plausible recoding yielded forest-root probabilities of 0.952-0.987, whereas an equal root-state prior reduced model-averaged forest-root support to 0.867 without overturning forest-time dominance. Conclusions - The sampled Tabebuia alliance most likely diversified during the late Eocene from an ancestral range that almost certainly included South America. A long history of exchange between humid and seasonally dry forests is more robust than any single directional scenario for savanna evolution. Support for a forest-associated ancestor remained strong across all sensitivity analyses, although the exact root-state probability depended on root-prior specification.