Replicate radiations provide powerful comparative systems to address questions about the interplay between opportunity and innovation in driving episodes of diversification and the factors limiting their subsequent progression. However, such systems have been rarely documented at intercontinental scales. Here, we evaluate the hypothesis of multiple radiations in the genus Lupinus (Leguminosae), which exhibits some of the highest known rates of net diversification in plants. Given that incomplete taxon sampling, background extinction, and lineage-specific variation in diversification rates can confound macroevolutionary inferences regarding the timing and mechanisms of cladogenesis, we used Bayesian relaxed clock phylogenetic analyses as well as MEDUSA and BiSSE birth–death likelihood models of diversification, to evaluate the evolutionary patterns of lineage accumulation in Lupinus. We identified 3 significant shifts to increased rates of net diversification (r) relative to background levels in the genus (r = 0.18–0.48 lineages/myr). The primary shift occurred approximately 4.6 Ma (r = 0.48–1.76) in the montane regions of western North America, followed by a secondary shift approximately 2.7 Ma (r = 0.89–3.33) associated with range expansion and diversification of allopatrically distributed sister clades in the Mexican highlands and Andes. We also recovered evidence for a third independent shift approximately 6.5 Ma at the base of a lower elevation eastern South American grassland and campo rupestre clade (r = 0.36–1.33). Bayesian ancestral state reconstructions and BiSSE likelihood analyses of correlated diversification indicated that increased rates of speciation are strongly associated with the derived evolution of perennial life history and invasion of montane ecosystems. Although we currently lack hard evidence for “replicate adaptive radiations” in the sense of convergent morphological and ecological trajectories among species in different clades, these results are consistent with the hypothesis that iteroparity functioned as an adaptive key innovation, providing a mechanism for range expansion and rapid divergence in upper elevation regions across much of the New World.
Lupinus.online.FigureS1.04.26.11
SUPPLEMENTARY FIGURE S1. Ancestral geographic ranges of Lupinus inferred under a dispersal–extinction cladogenesis (DEC) model in lagrange, based on the maximum clade credibility tree in Figure 2 of the main text. Circles correspond to the maximum likelihood estimate of ancestral ranges, with divided circles indicating ancestral ranges split by dispersal and/or extinction on subtending branches. The DEC estimates are very similar to the BEAST continuous-time Markov chain (CTMC) phylogeographic results shown in the main text, with the main difference being that DEC tends to reconstruct widespread ancestral ranges at nodes with uncertain posterior distributions in the CTMC analysis. We obtained similar estimates for DEC analyses based on 1000 trees randomly selected from the posterior distribution of Bayesian MCMC tree searches in BEAST (results not shown), indicating that the DEC analyses were robust to phylogenetic uncertainty.
Lupinus.online.FigureS2.04.26.11
SUPPLEMENTARY FIGURE S2. Net diversification rates (r = lineages/million years) in Lupinus estimated under the MEDUSA birth–death likelihood model. MEDUSA analyses were conducted using 1000 trees randomly selected from the posterior distribution of constrained pure–birth Bayesian MCMC tree searches in BEAST, pruned to the infrageneric skeleton topology in Figure 3 of the main text. Histograms show the frequency (ƒ) of estimates from replicate MEDUSA likelihood runs: a) ΔAIC scores for pure–birth versus birth–death models; b) background rates of net diversification in Lupinus; c) ΔAIC scores for a rate shift in eastern South America; d) net diversification rates in eastern South America; e) ΔAIC scores for a rate shift in western North America; f) net diversification rates in western North America; g) ΔAIC scores for a rate shift in Mexico/Andes; h) net diversification rates in Mexico/Andes. ΔAIC scores indicated that a pure–birth model was strongly favored over a model incorporating extinction (Fig. S2a). The strongest support for a primary rate shift was recovered in western North America (Fig. S2e), although the fastest rates were estimated in the Mexican/Andean clade (Fig. S2h).
Lupinus.online.FigureS3.04.26.11
SUPPLEMENTARY FIGURE S3. Net diversification rates (r = lineages/million years) in Lupinus estimated under the MEDUSA birth–death likelihood model, including relative extinction (ε = birth/death) as a free parameter. MEDUSA analyses were conducted using 1000 trees randomly selected from the posterior distribution of constrained pure–birth Bayesian MCMC tree searches in BEAST, pruned to the infrageneric skeleton topology in Figure 3 of the main text. Surface plots show the frequency (ƒ) of estimates from replicate MEDUSA likelihood runs, including the joint relationship between r and ε: a) background rates in Lupinus; b) eastern South America; c) western North America; d) Mexico/Andes. Estimates of net diversification were generally similar to estimates obtained under a pure–birth MEDUSA model, with the primary differences including lower r and high ε in the eastern South American clade (Fig. S3b), and a bimodal distribution of ε in the Mexican/Andean clade (Fig. S3d). However, ΔAIC scores shown in Figure S2a indicate that the pure–birth model was strongly favored over the birth–death results shown here. We obtained similar results for pure–birth and birth–death MEDUSA analyses based on the posterior distribution of trees from birth–death Bayesian MCMC tree searches in BEAST (results not shown), indicating that the MEDUSA analyses were robust to the choice of tree prior used for phylogenetic inference. Note that estimates of ε for eastern South America should be considered unreliable, since unresolved terminal clades do not contain sufficient information to infer this parameter (Rabosky et al. 2007; Alfaro et al. 2009).
Lupinus.online.FigureS4.04.26.11
SUPPLEMENTARY FIGURE S4. BiSSE analyses of correlated rates of net diversification (r = lineages/million years) for life history (annual versus perennial) and habitat (lowland versus montane) in Lupinus. BiSSE analyses were conducted under a pure–birth model using 1000 trees randomly selected from the posterior distribution of constrained pure–birth Bayesian MCMC tree searches in BEAST, pruned to the infrageneric skeleton topology in Figure 3 of the main text. Histograms show the frequency (ƒ) of estimates from replicate BiSSE likelihood runs: a) ΔAIC scores for unconstrained versus constrained life history rate models; b) rates of net diversification in annual species; c) rates of net diversification in perennial species; d) ΔAIC scores for unconstrained versus constrained habitat rate models; (e) rates of net diversification in lowland species; (f) rates of net diversification in montane species. The weight of evidence indicates strong support for higher rates of net diversification in perennial and montane lineages.
Lupinus.online.Appendix1.15.08.11
Supplementary methods and Identities, botanical vouchers, GenBank numbers and sources of DNA sequence data
Lupinus.online.Appendix2.04.26.11
DNA sequence alignment
Lupinus.online.Appendix3.04.26.11
Classification and assignment of extant diversities to infrageneric groups in Lupinus. The lettered notations below correspond to monophyletic terminal clades shown in Figure 3 of the main text.
Lupinus.online.Appendix4.04.26.11
Life history and habitat data for ancestral state reconstructions of character states in Lupinus.
