Recent assembly of the Cerrado, a neotropical plant diversity hotspot, by in situ evolution of adaptations to fire

Abstract

The relative importance of local ecological and larger-scale historical processes in causing differences in species richness across the globe remains keenly debated. To gain insight into these questions, we investigated the assembly of plant diversity in the Cerrado in South America, the world's most species-rich tropical savanna. Time-calibrated phylogenies suggest that Cerrado lineages started to diversify less than 10 Mya, with most lineages diversifying at 4 Mya or less, coinciding with the rise to dominance of flammable C4 grasses and expansion of the savanna biome worldwide. These plant phylogenies show that Cerrado lineages are strongly associated with adaptations to fire and have sister groups in largely fire-free nearby wet forest, seasonally dry forest, subtropical grassland, or wetland vegetation. These findings imply that the Cerrado formed in situ via recent and frequent adaptive shifts to resist fire, rather than via dispersal of lineages already adapted to fire. The location of the Cerrado surrounded by a diverse array of species-rich biomes, and the apparently modest adaptive barrier posed by fire, are likely to have contributed to its striking species richness. These findings add to growing evidence that the origins and historical assembly of species-rich biomes have been idiosyncratic, driven in large part by unique features of regional- and continental-scale geohistory and that different historical processes can lead to similar levels of modern species richness.

The uneven distribution of species diversity across the globe and the occurrence of biodiversity hotspots with high concentrations of species are well established (1–3). However, the underlying causes of differences in species diversity and composition among biomes and the processes that have prompted accumulation of high species diversity in some areas remain poorly understood (4–6). Correlations between species richness and annual energy input, water supply, and physiographic complexity suggest that climatic and other environmental factors are major determinants of species diversity (7). Indeed, it has been shown that these factors can accurately predict the locations of most global diversity hotspots and account for the latitudinal gradient of species richness (7). However, such insights contribute little to our understanding of the historical assembly of species-rich biomes and the larger-scale evolutionary processes that have generated global patterns of diversity (4, 8, 9). Little is known about the historical and geographical assembly of species-rich biomes (5), in terms of when, how quickly, and from where the species and lineages that make up different biomes have been recruited and how they subsequently evolved in situ. This lack of data on geographical and temporal patterns of species diversification, especially in the tropics where most diversity resides, makes it difficult to assess why there are so many species in these areas, to what extent variation in diversity can be attributed to regional and continental-scale historical contingencies (4, 10, 11), and to compare patterns among different species-rich biomes (9, 12).

Recent discussion has highlighted the potential role of phylogenetic niche conservatism in shaping regional species pools (8, 13) and explaining diversity gradients (11). Prominent examples of large-scale niche or biome conservatism have been documented for the tropical–temperate biotas (11), mangroves (14), southern hemisphere extratropical biomes (13), Andean grasslands (8, 15), and seasonally dry tropical forests (16–18). However, the extent to which this tendency to retain ancestral ecology across lineages has influenced species composition in the most species-rich tropical biomes is unknown.

Recent insights into the historical assembly of species diversity and biomes have come from time-calibrated phylogenies for biome-specific lineages (5, 9, 19–23). Phylogenies have potential for reconstructing transitions from precursor to modern biotas and identifying the underlying factors that drive these processes (4, 11, 24). However, the sparse sampling of lineages and species in such studies (5, 24) has limited these insights to a few well-studied areas such as the Cape Floristic Region (12, 20).

The Cerrado.

To address these questions, we have investigated plant diversification in the Cerrado of South America by using a comparative phylogenetic approach. The Cerrado is a floristically diverse savanna that covers more than two million km² of Central Brazil and parts of Bolivia and Paraguay (Fig. 1). This region has been recognized as a global biodiversity hotspot with more than 10,000 plant species, of which 44% are endemic (1, 25, 26). Despite the floristic and global conservation importance of the Cerrado, little is known about the origins and diversification of its flora. Indeed, hypotheses about Cerrado origins range from the early Cretaceous (27), wherein Cerrado lineages were suggested to be possible precursors of the adjacent Amazonian and Atlantic rain forests, to the Holocene (28). Here we evaluate these alternatives, but we particularly focus on the hypothesis that the origin of the biome coincided with the rise to dominance of flammable C4 grasses within the past ten million years (6, 29).

Fig. 1.

Map of major vegetation types in South America showing the location of the Cerrado surrounded by a diverse array of other biomes.

In common with other tropical savannas, the Cerrado is dominated by C4 grasses which take advantage of high light and warm wet summers to rapidly accumulate biomass, which becomes flammable in the long dry winters, prompting fires, generally several times in a decade (30, 31). The synergy between fast (re)growth and flammability allows grasses to outcompete trees and shrubs, maintaining the open canopy typical of savannas in areas where, without fire, forests would dominate (31). Diverse fire adaptations are the hallmark of the endemic Cerrado flora (32). These adaptations include functionally herbaceous or woody geoxylic suffrutices with enlarged underground xylopodia or lignotubers, thick corky bark, pachycaul rosulate shrubs and trees with sparse branching, thick shoots and leaves concentrated at shoot tips, perennial herbs, and specialized flowering and fruiting phenologies (27, 32, 33). These adaptations are largely lacking in the flora of adjacent biomes where fire has been less important over evolutionary time scales (29). C4 grasses originated and started to diversify 32–25 Mya in several independent lineages (34). However, evidence from isotopic signatures of fossil mammal teeth (paleodiets) and paleosols, charcoal particles, and pollen from sites on different continents suggests that these lineages only became ecologically dominant 4–8 Mya (35–37). This near-synchronous expansion of grass-dominated vegetation in different parts of the world is regarded as marking the onset of the modern savanna biome (38).

To test the hypothesis that the evolution of the Cerrado flora followed the ecological expansion of flammable C4 grasses and to investigate the significance of fire adaptations as a selective advantage for Cerrado diversification, we assembled time-calibrated phylogenies for plant groups that include Cerrado elements. Our sample includes four datasets representing lineages rich in Cerrado endemics (Mimosa and Microlicieae, Melastomataceae) and lineages sparsely represented in the Cerrado (Lupinus and Andira), as well as diverse life forms including trees, shrubs, and perennial herbs. Two of the phylogenies (Mimosa and Andira) are published for the first time. Central to the study is a new phylogeny for the species-rich (>500 species) legume genus Mimosa, the second-largest plant genus in the Cerrado (26, 39). In addition, new analyses are presented of published data for the legume genus Lupinus (15) and the tribe Microlicieae (40). These phylogenies are used to infer divergence times and ancestral biomes for Cerrado lineages, investigate the evolution of fire adaptations, and ultimately to build an evolutionary picture of the historical assembly of species in the Cerrado.

Results and Discussion

Our analyses identified 15 Cerrado lineages, 11 in Mimosa, two in Andira and one each in Lupinus and Microlicieae (Fig. 2A). Most of these clades are robustly supported and represent groups where fire-adapted endemics make up the majority of species (see Table 1 and SI Appendix for descriptions of each lineage). In three of the study groups, Cerrado lineages form clades that are deeply nested within genera. In two, there are multiple independent Cerrado lineages. These findings are in line with floristic data showing the preponderance of species-level endemism and lack of endemic genera in the Cerrado (32, 33). Within Mimosa, Cerrado lineages originated independently at least 11 times (Fig. 2A). Two of these clades (Mimosa 3 and 8) are notably rich in fire-adapted species resulting from in situ Cerrado diversification and together represent ≈75% of all Cerrado species of Mimosa. Notably, both these species-rich Mimosa radiations are estimated to have arisen approximately 4 Mya. The diversity of life forms in these two clades (Fig. 2 B–G), including functionally herbaceous subshrubs, prostrate herbs, wand-like subshrubs and pachycaul rosette trees, reflects the diverse ways that species have evolved to tolerate fire disturbance. Similar multiple independent origins of Cerrado species are apparent in Andira, with two recent clades both having two species. The largest Cerrado clade sampled here, representing ≈200 Cerrado endemics (40), is nested within the tribe Microlicieae. However, it is clear that this was not the only incursion of Melastomataceae into the Cerrado, with several other fire-adapted Cerrado lineages in this family (e.g., in Miconia, Cambessedesia, and Tibouchina).

Fig. 2.

Evolution of the fire-adapted Cerrado flora. (A) Chronograms for Mimosa, Andira, Lupinus, and Microlicieae showing 15 Cerrado lineages (red). Outgroups are depicted in gray. Symbols for fire adaptations: * = subshrub growing from xylopodium; ▴ = pachycaul treelet; ■ = thick corky bark. Numbered nodes correspond to Cerrado lineages. The Microlicieae phylogeny has been pruned to show just the Cerrado lineage and to fit within the time scale under investigation. Expanded phylogenies, including terminal names and support values, are presented in the SI Appendix. (B–I) Photographs illustrate the diversity of life forms and fire adaptations found in Mimosa clades 3 (B and C) and 8 (D–G), and Andira (H and I). (B) Wand-like subshrub with a xylopodium, Mimosa pseudoradula. (C) Functionally herbaceous subshrub with a horizontal underground xylopodium, Mimosa venatorum. (D) Functionally herbaceous subshrub growing from a massive xylopodium, Mimosa speciosissima. (E) Functionally herbaceous wand-like subshrub growing from xylopodium, Mimosa ulei. (F) Rosulate shrub, Mimosa oligosperma. (G) Pachycaul treelet with few thick branches, Mimosa splendida. (H) A branch of the geoxylic suffrutex or “underground tree”, A. humilis. (I) Thick corky bark, Andira cordata (scale bars = 10 cm). (J) Divergence-time estimates for 15 Cerrado lineages (crown nodes). Mean and 95% credibility intervals derived from 2.7 × 10⁴ samples of a Bayesian analysis.

Table 1.

Diversity and origin of 15 Cerrado lineages sampled in this study

Lineage	Total*/sampled species	Proportion of lineage in the Cerrado (%)	Age (Mya) (95%CI)	Life form/fire adaptation	Ancestral biome
Mimosa 1	1/1	100	0.4 (0.0–1.2)	subshrub with xylopodium	equivocal
Mimosa 2	3/3	100	1.6 (0.3–3.4)	subshrub with xylopodium	equivocal
Mimosa 3	34/16	92	4.4 (2.2–6.7)	subshrub with xylopodium	dry forest
Mimosa 4	11/6	73	1.6 (0.6–2.8)	shrub	equivocal
Mimosa 5	2/1	50	3.6 (1.2–6.0)	shrub	equivocal
Mimosa 6	4/2	100	0.9 (0.0–2.3)	shrub, tree	equivocal
Mimosa 7	8/6	87	3.2 (1.4–5.3)	subshrub with xylopodium	equivocal
Mimosa 8	50/26	100	4.1 (2.0–6.5)	xylopodium, pachycaul tree	equivocal
Mimosa 9	27/8	81	8.4 (4.3–12.8)	subshrub with xylopodium	dry forest
Mimosa 10	1/1	100	2.3 (0.4–4.8)	tree	dry forest (Caatinga)
Mimosa 11	1/1	100	4.4 (1.0–8.1)	tree	dry forest (Caatinga)
Andira 1	3/3	67	1.8 (0.5–3.4)	″underground tree″	rain forest (Amazon)
Andira 2	2/2	100	0.7 (0.0–2.0)	tree with thick corky bark	rain forest (Amazon)
Lupinus	11/5	100	1.9 (0.9–3.1)	perennial herb, ericoid shrub	subtropical grassland
Microlicieae	200/25	90	9.8 (6.2–14.0)	ericoid shrub, xylopodium	wetland

*In some cases, the total number of species in a lineage was estimated by assigning unsampled taxa to groups based on taxonomic affinities.

The estimated ages of Cerrado lineages span the late Miocene to Pliocene from 9.8 to 0.4 Mya, with the majority less than 4 Mya (Fig. 2J), broadly coinciding with the hypothesized expansion of C4 grass-dominated savanna biomes (38). Some Cerrado lineages apparently slightly predate hypothesized C4 grass dominance, but the majority are younger (Fig. 2J), and this is compatible with fire as a common underlying trigger prompting the evolution of endemic Cerrado species and lineages, although the precise dynamics of this are clearly complex (30, 31, 38). One factor influencing very recent (Quaternary) evolution of Cerrado lineages (e.g., in Andira) may have been dry glacial periods, when the Cerrado was likely to have been more extensive. The derived nature of Cerrado lineages found here is in line with analyses of the legume family as a whole that showed that savanna elements worldwide are in general recently derived from dry or wet forest ancestors (18).

The idea that the Cerrado lineages evolved in response to increased C4 grass dominance and fire frequency is reinforced by the occurrence of fire adaptations in these groups (Fig. 2, Table 1). Fire adaptations in Andira include thick corky bark and a geoxylic suffruticose growth form in Andira humilis, a subshrub with an extensive network of underground woody shoots (“underground tree”), a life form unique within this otherwise arborescent genus (41) (Fig. 2 H and I and SI Appendix). Within Mimosa, fire adaptations include: functionally herbaceous perennials with xylopodia (Fig. 2 B–E) that evolved multiple times independently from an ancestral shrubby habit; unusual pachycaul treelets with few branches, thick shoots and leaves clustered at shoot apices (Fig. 2 F and G and SI Appendix); and crowded, persistent stipules that protect shoot apices from fire. The evolution of fire-adapted life forms is strongly correlated with Cerrado occurrence in Mimosa (likelihood ratio statistic = 30.30, P < 0.01) but much rarer or absent in sister lineages growing in adjacent wet and seasonally dry tropical forest. Indeed, seasonally dry tropical forest (the “succulent biome” sensu in ref. 18), which is the ancestral biome for several Cerrado lineages (see below), although sharing a long dry season with the Cerrado, is characterized by sparse occurrence of grasses, abundant Cactaceae and other succulents, and lack of natural fires (18, 29, 37). The large, fire-adapted Microlicieae clade is also nested within a family that predominates in fire-free (rain forest and swamp) habitats (discussed in ref. 40). It is notable that similar fire adaptations are common across other plant families in the Cerrado. For example, of 301 plant species in a one-hectare plot in the southern Cerrado, 94 species across 64 genera and 37 families have xylopodia (32).

Adaptive shifts between ecological zones can play an important role in the generation of species diversity (4, 11). Our results alongside wider floristic data (32, 42) show that fire adaptation and Cerrado occurrence are phylogenetically labile with multiple independent lineages at shallow depths within species-level phylogenies and across genera in disparate plant families. These frequent adaptive shifts suggest that fire does not pose a significant adaptive barrier to shifts between biomes. In comparison, shifts from tropical into temperate biomes and mangrove vegetation have been infrequent (4, 8), suggesting that adaptations to high salt and substrate anoxia or frost are more complex to evolve. In the case of the Cerrado, the boundaries appear to have been porous to the ingress and recruitment of lineages from a range of fire-free wet and dry forest vegetation types. Of significance is the intermingling of open, fire-prone savanna with patches of closed, less fire-prone riverine gallery forest, which share few species in common but which can harbor both fire-adapted and fire-sensitive congeners and even sister species (43). Phenotypic plasticity in fire traits (e.g., the normally stunted A. humilis can sometimes form a shrub or treelet in the absence of fire; ref. 41) is a further indication of the potential ease of the evolutionary transitions associated with fire adaptation. We suggest that the ease of fire adaptation has been significant in generating high species diversity in the Cerrado, although the ingress of a lineage into this fire-prone habitat has not necessarily led to significant or rapid in situ diversification. The 10 largest genera (1,474 spp.) make up only 13% of the Cerrado flora (26), and many of the dominant Cerrado tree genera (e.g., Qualea and Caryocar) are represented by only few species.

Several authors have pointed to affinities between the woody flora of the Cerrado and the Atlantic and Amazon rain forests (32, 42–44). For example, of 121 woody species dominant in the Brazilian Cerrado, 99 belong to predominantly rain forest genera (6). These affinities are borne out by wet forest–Cerrado transitions implied by the phylogeny of Andira (SI Appendix) and undoubtedly many other groups not sampled here. However, the sister groups and putative ancestral areas and ecologies of Cerrado lineages identified in our phylogenies are diverse and include geographically adjacent seasonally dry tropical forests, subtropical grasslands, and wetlands, as well as wet forests (Table 1 and SI Appendix). In seven of 15 lineages, ancestral areas were recorded as equivocal, but even for these, the options with the highest probabilities frequently included dry and wet forest. These results agree with floristic data that suggest that neotropical savannas comprise a mixture of elements of various provenance and floristic affinities (42). The Cerrado is bordered by a diverse array of biomes (Fig. 1) including the Amazon and Atlantic wet forests, tropical and subtropical dry thorn scrub (Caatinga and Chaco), subtropical grasslands, and wetlands. The proximity of these diverse vegetation types that have all contributed to the recruitment of Cerrado lineages suggests that this privileged position in the heart of South America may have played a role in the Cerrado's striking species richness.

The in situ assembly of the endemic rich Cerrado flora via frequent recent adaptive shifts to resist fire stands in contrast to the widespread support for ideas that lineages tend to maintain their ancestral ecologies (8, 11, 13) and that dispersal of preadapted lineages has played an important role in the assembly of regional species pools (8). For example, the species-rich páramo biome in the Andes includes many northern, temperate plant lineages that were able to disperse and to take advantage of similar ecological conditions to diversify rapidly in the Andes (8, 15, 45). Similarly, many seasonally dry tropical forest clades show evidence of phylogenetic niche conservatism among disjunct areas across the tropics (16, 18). In these cases, it has apparently been easier to switch geography than ecology (8). However, this does not appear to be the case for the Cerrado, where niche evolution from adjacent closed, fire-free habitats to open, fire-prone habitats, cross-cutting other abiotic niche dimensions, has apparently been prevalent in the evolution of the Cerrado flora.

This study is an initial attempt to assemble a representative set of time-calibrated phylogenies for Cerrado plants. The lineages discovered represent just 3–4% of the Cerrado flora. Despite this sparse sampling, there are indications that our results are representative of the flora as a whole. First, the study groups include species-rich and species-poor lineages as well as diverse life forms. Second, the preponderance of species, as opposed to generic endemism and prevalence of fire adaptations amongst these endemics, support the idea that Cerrado lineages are derived and recent. Finally, preliminary phylogenies for other plant groups including Styrax (46), Viguiera (47), Ruellia (48), and Manihot (49), also show evidence of recently derived Cerrado lineages, multiple independent Cerrado lineages (in Ruellia, Manihot, and Styrax) and associated fire adaptations (e.g., xylopodia in Ruellia and Viguiera) (SI Appendix).

The emerging picture of Cerrado origins is of recent diversification of endemic plant lineages that took place during the late Miocene and early Pliocene, driven by the common trigger of fire adaptation and facilitated by ease of fire adaptation across plant groups from the diverse biomes immediately surrounding the Cerrado. This picture is very different from what is known about the origins and diversification of plants in one of the best-studied model hotspots, the Cape Floristic Province in South Africa, where plant lineages are estimated to range from less than 5–46 Mya, with no evidence for simultaneous initiation driven by a single trigger (20, 50). Linder (9) classified these older Cape lineages as mature radiations, in which stability over time has produced numerous endemic genera. In contrast, the Cerrado seems to be the product of more recent diversification with very few endemic genera. These contrasting scenarios suggest that the origins of species-rich biomes and underlying causes of high species diversity can be highly idiosyncratic, driven as much by unique features of regional- and continental-scale history and physiography as by any more universal ecological processes at more local scales (4, 9, 10). If species-rich biomes do indeed have very different histories, then understanding the differences between biomes, and hence the mechanisms responsible for the origin and maintenance of the variation in species richness over the globe, will be crucial for the interpretation of global biodiversity gradients across space (e.g., refs. 8 and 45) and time (e.g., refs. 51 and 52).

Methods

Datasets and Sampling.

Our sample of plant lineages was constrained by the paucity of species-level phylogenies that include Cerrado elements. Of the published phylogenies for plant groups that include Cerrado species, Styrax (Styracaceae) (46), Viguiera (Asteraceae) (47), Microlicieae (Melastomataceae) (40), Ruellia (Acanthaceae) (48), Lupinus (Leguminosae) (15), and Manihot (Euphorbiaceae) (49), only two (Lupinus and Microlicieae) were judged to be well enough sampled, resolved and supported, and amenable to time calibration for more intensive analysis of Cerrado origins (SI Appendix). New DNA sequence data were generated for two legume genera, Mimosa and Andira. Our largest dataset is for the species-rich genus Mimosa (chloroplast region trnD-trnT), for which ≈50% of the >500 species were sampled, including 92 of 189 species listed for the Cerrado (39), most of these being narrow, fire-adapted endemics. For Andira, a nearly fully sampled dataset of ITS sequences was assembled, including multiple accessions for some species and all four species that grow in the Cerrado. The Lupinus dataset used here (15) comprises ITS and LEGCYC1A nuclear DNA sequences for 98 species, including five of the 11 Cerrado endemics. Data for Microlicieae include sequences for three plastid loci (rbcL, rpl16, and ndhF) and 59 species. The published Microlicieae data (40) were reanalyzed with 23 additional taxa, for which only one data partition (rpl16) was available, to increase taxon sampling and accuracy of divergence time estimates. Laboratory methods and taxa sampled are presented in the SI Appendix.

Phylogenetic Analyses and Dating.

Phylogenetic trees and divergence times for Cerrado nodes were estimated by using an uncorrelated relaxed molecular clock approach implemented in BEAST, version 1.4.8 (53). All analyses were conducted by using the uncorrelated lognormal relaxed clock, assuming a general time-reversible model, with invariable sites and among-site rate heterogeneity (GTR+I+Γ). Results were assessed to have reached stationarity and convergence by using Tracer, version 1.4 (54), and data from multiple runs were combined after exclusion of burn-in trees.

Divergence-time estimates for the legume genera, Mimosa, Andira, and Lupinus were based on a new family-wide analysis of matK data containing 839 terminals. This large dataset was assembled from previous alignments (55, 56) with the incorporation of 101 additional mimosoid legume sequences (see ref. 57 and additional refs. in the SI Appendix), thereby increasing taxon sampling well beyond previous legume studies. The dating procedure was built upon previous studies (55, 56) that used a set of well-documented fossils for calibration but with the addition of two new mimosoid fossil constraints. In total, 23 fossils were used as minimum age constraints (lognormal prior distribution, mean = 0, stdev = 1, in BEAST) (SI Appendix). The legume stem node was allowed to vary between 60 and 70 Mya (uniform prior distribution). Twenty individual chains of 10⁷ generations were run, and after the exclusion of 2 × 10⁶ generations (burn-in from each chain), the results were combined. Age estimates for five nodes (mean and 95% credibility intervals) from this analysis were used as calibration points for subsequent species-level analyses of Mimosa, Andira, and Lupinus by using a normal distribution prior, with mean and standard deviation set to the corresponding estimate.

For the Microlicieae dataset, three fossils provided minimum age constraints (40): (i) the oldest Melastomataceae fossil (53 Mya, Melastomataceae crown node), (ii) the oldest tribe Melastomeae fossil (23 Mya, Melastomeae crown node), and (iii) a 30-Mya fossil leaf assigned to the Merianieae crown node. These constraints were imposed as lognormal distributions as described above. The age of the core Microlicieae was previously estimated to be 3.7 Mya (40), which is considerably younger than the divergence time reported here (9.8 Mya). This discrepancy can be attributed to the denser taxon sampling used here, given the known bias toward younger age estimates caused by undersampling (58).

Analyses of the four study group datasets comprised three independent runs of 10⁷ generations each, which all converged to the same posterior and were combined, resulting in final aggregates of 2.7 × 10⁷ generations (after exclusion of burn-in trees). Fully annotated phylogenies, including terminal taxon names are presented for each study group (SI Appendix).

Ancestral Biome Reconstruction and the Evolution of Fire Adaptation.

Ancestral character reconstruction analysis was used to identify Cerrado lineages, infer their putative ancestral biomes, and investigate the evolution of fire adaptations. Ecologies or “biomes” of occurrence (seasonally dry tropical forest, rain forest, wetland, savanna, subtropical grassland, temperate, Mediterranean, tropical montane, widespread) for sampled species were assigned based on species distribution data from taxonomic accounts (SI Appendix) and field observations, coded as discrete character states and optimized under a maximum parsimony criterion (“Unordered”) onto the 50% majority rule consensus tree by using Mesquite (59). To account for topological uncertainty, the procedure “Trace over trees” was used to summarize reconstructions over a set of 540 Bayesian trees sampled at stationarity. This approach was used to define Cerrado lineages by identifying crown nodes assigned unequivocally to the “savanna” biome. The same procedure was used to identify ancestral biomes for Cerrado lineages as the most frequent state at the stem nodes subtending each clade. Estimating ages of Cerrado lineages by using stem, rather than crown, nodes generates estimates 2.5 Mya older on average and therefore does not alter the conclusions of the study. Optimizations were also performed by using a maximum-likelihood approach in Mesquite under the Mk1 model, as used in biome reconstructions in other studies (13), but this alternative approach did not alter the results.

The evolution of morphological traits related to fire was investigated in a similar way. The presence/absence of thick corky bark in Andira and the evolution of fire-adapted life forms in Andira and Mimosa were analyzed. For these two genera, species were coded for life form (herb, shrub, vine/liana, functionally herbaceous subshrub with xylopodium, woody shrub with xylopodium, pachycaul treelet, tree), and optimized as described above. We used a comparative analysis of discrete characters (60) to test the correlation between the evolution of fire-adapted life forms and Cerrado occurrence in Mimosa. This method (implemented in Mesquite) searches for evidence of correlated evolutionary change in two discrete characters by using a likelihood ratio test statistic to discriminate between two models fitted to the data—one for independent evolution of the two characters and the other allowing for correlated evolution. We ran 100 Monte Carlo simulations to test if the likelihoods of the two models are statistically different. Data matrices for biome of occurrence and morphology, along with character-state optimizations and detailed methods, are presented in SI Appendix.