Estimation of divergence times

We estimated a time-calibrated phylogeny of Velloziaceae by Bayesian inference using relaxed molecular clock models in BEAST 2.2 (Bouckaert et al., 2014). We based topological priors (i.e. the monophyly of Velloziaceae and of Pandanales) on the results from the non-clock MrBayes analysis. We excluded the two most distant outgroups (Acorus gramineus and Encholirium scrutor), as including them in exploratory runs led to very large confidence intervals for the root age of the tree, and much longer times until MCMC convergence.

There are no known fossils of Velloziaceae, so for calibration purposes we used the oldest fossil known for the order Pandanales, represented in our matrix by members of Velloziaceae and three outgroups: Cyclanthus bipartitus Poit. ex A.Rich. and Thoracocarpus bissectus (Vell.) Harling (Cyclanthaceae), and Pandanus pygmaeus Thouars (Pandanaceae). We used standard procedures (Magallón, 2004; Iles et al., 2015) to assign the oldest fossil of Pandanales to the stem node of the order. To identify this fossil, we carried out an extensive review of the literature on monocot fossils, cross-referenced with a recent review of monocot fossils suitable for molecular dating (Iles et al., 2015). We obtained two candidates: inflorescences of Mabelia and Nuhliantha (Triuridaceae) dating from the Turonian period (93.5–89.3 Mya; Gandolfo et al., 2002), and an infructescence of Pandanaceae, Gruenbachia pandanoides (Herman and Kvacek, 2010), dating from 83.6–72 Mya. The position of Gruenbachia pandanoides within Pandanaceae is still uncertain (Iles et al., 2015), and the Turonian Triuridaceae fossils are older, so we used the latter to establish the minimum age of Pandanales. We specified a lognormal prior with mean M = 15 and standard deviation S = 0.75, resulting in a median of 101 Mya.

As with the MrBayes analysis, we partitioned the data by gene and used RJMCMC to average over substitution models using the BEAST add-on package RBS (Bouckaert et al., 2013). Models and clock rates were unlinked across the four data partitions. We used a birth–death tree prior and a relaxed, uncorrelated lognormal (UCLN) molecular clock (Drummond et al., 2006). All specifications for the analysis are included in the input file (available upon request). We performed four independent runs, each for 50 million generations and a sampling frequency of 5000. To check for stationary and convergence, we examined parameter ESS values and distributions in Tracer 1.6 (Rambaut et al., 2013). Results were combined using LogCombiner (distributed with BEAST 2.2), discarding 20 % as burn-in. We used TreeAnnotator (distributed with BEAST 2.2) and FigTree 1.4.1 (Rambaut, 2009) to produce the maximum clade credibility (MCC) tree, which we used for downstream analyses. We compared node ages recovered by this analysis to those of other studies based on different calibrations (Janssen and Bremer, 2004; Mennes et al., 2013; Hertweck et al., 2015; Magallón et al., 2015).

Diversification analyses

We analysed diversification rates on the MCC tree pruned of all outgroups and seven redundant terminals of six species (two terminals representing different morphological varieties of Barbacenia flava were retained). The final tree included 151 terminals (150 species).

We used this final tree as input to BAMM 2.5, which uses Bayesian RJMCMC to infer heterogeneity in rates of speciation, extinction and trait evolution on phylogenetic trees (Rabosky, 2014). To account for incomplete taxon sampling, we included sampling fractions of each genus of Velloziaceae. We ran BAMM for 10 million generations, sampling every 10 000 generations, and analysed the results using BAMMtools 2.1.6 (Rabosky et al., 2014). We discarded the first 10 % of samples as burn-in and verified chain stationarity by inspecting plots of the likelihood trace and calculating ESS values of parameters.

We used BAMMtools to summarize differences in diversification across the phylogeny of Velloziaceae, and calculated marginal odds ratios for rate shifts on individual branches and Bayes factors (BFs) for the number of distinct rate regimes (Shi and Rabosky, 2015). We recorded the distinct shift configurations in the 95 % credible set and identified the core shifts with BFs exceeding 5. We summarized the number of shifts in the 20 shift configurations that together account for > 50 % of the posterior distribution. We considered a marginal odds ratio > 50 as strong evidence favouring a shift in net diversification rate on a given branch. We performed macroevolutionary cohort analysis (Rabosky et al., 2014), identifying sets of taxa likely to share the same diversification regime. We calculated mean rates of diversification across the Velloziaceae phylogeny and generated rates-through-time plots for specific clades.

To cross-check these results we used MEDUSA, a maximum-likelihood approach of estimating diversification rate shifts (Alfaro et al., 2009), using the R (R Core Team, 2015) package geiger v. 2.0.6 (Harmon et al., 2015). We accounted for incomplete phylogenetic sampling by adding unsampled species to their respective genera (i.e. Barbacenia, Vellozia and Xerophyta).

We also tested directly whether speciation rates differ between African and South American clades of Velloziaceae by the Hidden-State Speciation and Extinction model (HiSSE) (Beaulieu and O’Meara, 2016). HiSSE extends previous state-dependent diversification models (e.g. BiSSE; Maddison et al., 2007) by allowing an unobserved binary character to influence diversification dynamics in addition to the observed character, mitigating the risk of Type I error (Rabosky and Goldberg, 2015; Beaulieu and O’Meara, 2016). We coded geography as a binary trait (occurrence in Africa vs. South America), pruning the sole Asian species, Acanthochlamys bracteata, from the tree. We used the R package hisse 1.1 (available at https://cran.r-project.org/web/packages/hisse/index.html) to evaluate the fit of alternative models by their average Akaike weight (w_i) (Burnham and Anderson, 2002) and estimate confidence intervals for parameters of the best-fit model. We evaluated two models: BiSSE, having no hidden states; and HiSSE-partial, in which only one observed state (i.e. ‘South America occurrence’, which accounts for most species and genera of Velloziaceae) can transition between ‘hidden’ states. Model fit was assessed by w_i (Burnham and Anderson, 2002).

Diversification rates in Velloziaceae

Both BAMM and MEDUSA indicated heterogeneity in diversification rates across the phylogeny (Fig. 2 and Material S7). There is strong support for two rate shifts (Table 2 and Material S7). Although a three-shift model had the highest posterior probability with BAMM, its BF is only 3.7) over the two-shift model. Examination of the most credible rate shift configurations from BAMM (see Fig. 3A) indicates some uncertainty about the precise number and locations of rate shifts (Material S9), but it is clear that two occur within the two largest Neotropical genera, Barbacenia and Vellozia (Fig. 3B), a result also obtained by MEDUSA (Material S9). We present the mean (model-averaged) diversification rates estimated along every branch of the Velloziaceae tree by BAMM (Fig. 2). In Vellozia, the shift is to a regime with higher initial net diversification that declines through time, while for Barbacenia, the shift is of greater magnitude to a higher constant rate (Fig. 2). We refer to the clades descending from these shifts as core-Vellozia and core-Barbacenia, respectively. Apart from these two clades, rates across the rest of the family are more or less homogeneous, except for weak signals of slowing diversification rates in the monotypic Acanthoclamys (node 1) (Figs 2 and ​and3;3; Material S9). This third rate shift has low support (BF = 1.83).

Open in a separate window

Fig. 3.

BAMM results of rate shifts in Velloziaceae. (A) Top four most credible shift sets recovered by BAMM with associated probabilities (f). Grey circles indicate the location of estimated shifts. Rate values represent new lineages per million years, with rates increasing from blue to red (online version). (B) Phylogeny of Velloziaceae with branch lengths transformed to Bayes-factor support for rate shifts.

MEDUSA shows strongest support for two independent shifts in macroevolutionary regime (we report the results of the analysis including sampling fractions, because there was no difference from the analysis that omitted incomplete sampling). They are inferred to be more recent than those inferred by BAMM (Fig. 2), nesting within the core-Barbacenia and the core-Vellozia clades (Material S7).

Estimates of net diversification by BAMM are higher and more variable (uncertain) in the South American clades of Velloziaceae [lambda: 0.18, quantiles (25–75 %): 0.15–0.25; net diversification: 0.17, quantiles: 0.125–0.195] than in the African clade (lambda: 0.1, quantiles: 0.07–0.14; net diversification: 0.08, quantiles: 0.075–0.081). This is reflected in a relatively complex estimate of macroevolutionary cohorts in the family (Fig. S10), and in comparisons across genera and infra-generic clades of Velloziaceae (Fig. 4), as well as the variable rates of speciation and extinction estimated from BAMM simulations (Fig. S11). The geography-dependent diversification analysis corroborates more variable rates in South America, with the highest support for a model in which the rate of diversification there, in conjunction with a ‘hidden’ binary character (HiSSE-partial model, see Table 3), is about four times higher than the background diversification rates within Velloziaceae (i.e. the rate observed in both the African lineage and the South American representatives that are not associated with the ‘hidden’ state).

Open in a separate window

Fig. 4.

(A) Estimated speciation rates in Velloziaceae lineages and (B, C) estimated logarithm of the number of species (logN.spp) accumulation through time (in millions of years) for selected lineages of Velloziaceae. All Velloziaceae: includes all five genera of the family; core-Barbacenia + core-Vellozia: the infra-generic clades within these genera exhibiting the highest probability for having suffered shifts in their diversification rates relative to the background rates simulated by BAMM analyses (delimitated by nodes 157 and 209, respectively).

African versus South American diversification of Velloziaceae

The diversification analyses clearly support the hypothesis that the imbalance between Palaeo- and Neotropical floristic diversity results from higher speciation rates in South America (e.g. Gentry, 1982; Linder, 2014; Couvreur, 2015). However, occurrence in Africa versus South America per se does not explain much of the variation in diversification of Velloziaceae. Instead, rate heterogeneity is nested within the South American clades, with independent shifts in Vellozia around 23 Mya and in Barbacenia around 13 Mya. Similar patterns of higher speciation in Neotropical lineages have been shown in other plant groups centred in Andean and Amazon regions (e.g. Bell and Donoghue, 2005; Hughes and Eastwood, 2006; Bardon et al., 2013; Koenen et al., 2015). The main biogeographical factors proposed to explain higher speciation in the Neotropics are area, time for speciation and climatic stability (reviewed by Couvreur, 2015). However, climatic instability during the Quaternary resulting in population isolation in refugia has also been invoked to explain recent diversification in the region (Haffer, 1969; Prance, 1974). Here, neither the shifts nor average differences in the diversification rates seem to be related to overall clade ages, since the South American Vellozia–Barbaceniopsis clade and the African Xerophyta are sister clades (i.e. exhibit the same stem lineage ages) but exhibit different diversification dynamics (Fig. 2). In addition, the South American genus Barbacenia, which split from the [Vellozia–Barbaceniopsis]–Xerophyta clade approx. 10 Myr earlier (i.e. showing the older stem age between the three major clades of Velloziaceae), is the clade showing the most recent crown age and the highest diversification rate in the family (Fig. 2). The total geographical range of Xerophyta in Africa is larger than that of the South American genera of Velloziaceae, and range sizes of individual species also tend to be larger in Africa compared to the campos rupestres endemics in South America (Mello-Silva et al., 2011); however, it is difficult to speculate whether small-scale geographical range has influenced diversification rates.

Most diversification in Velloziaceae is concentrated within the Neogene (mainly in the Miocene), although the origin estimated for the stem lineages of the genera occurring in both South America and Africa dates from the early Eocene (approx. 55–40 Mya). This pattern is consistent with expectations of ancient origins and persistence of old lineages under a ‘museum’ hypothesis for the Neotropics. Even considering the alternative scenarios with more recent rate shifts (as reconstructed by MEDUSA), diversification in Neotropical Velloziaceae is still older than that predicted by higher diversification rates caused by Pliocene/Quaternary climatic events. The relatively homogeneous background rates of diversification in Velloziaceae suggest that catastrophic events did not cause higher extinction rates in Africa than in other tropical areas, as proposed by Raven and Axelrod (1974). Indeed, evidence for catastrophic climatic events in Africa is largely lacking (Couvreur, 2015) and many studies have found bursts of diversification since the early Oligocene to the Miocene in African clades (Schnitzler et al., 2011; Linder, 2014), similar to our findings in Velloziaceae. Despite being relatively old, such shifts in diversification linked to environmental changes contradict the ‘museum’ model of tropical biodiversity, which assumes steady diversification under climatic stability (see Koenen et al., 2015).

Current evidence, including our results, suggests that independent bursts of diversification across time, in different lineages, have generated spatial imbalances in tropical diversity, instead of a single factor (Linder, 2008; Hughes et al., 2013; Couvreur, 2015). Likewise, multiple nested rate shifts accounting for the high diversity of angiosperm lineages (Magallón and Sanderson, 2001) appear to be a general pattern, instead of single rate shifts explaining exceptionally species-rich plant groups (Smith et al., 2011; Koenen et al., 2013). This study is one of the first to show that faster diversification underlies higher Neotropical diversity in lineages from open areas, as most studies to date have focused on rain forests or dry forests (Hughes et al., 2013). A notable exception is Leguminosae, in which rate shifts have been documented in Neotropical lineages found in grasslands and savannas (Simon et al., 2009; Koenen et al., 2013), as well as in the campos rupestres and Andean páramos (Drummond et al., 2012; Koenen et al., 2013; de Souza et al., 2013; Rando et al., 2016). Likewise, contrasts in diversity in these legume clades are closely associated with particular distinctive ecologies, and potentially with ecological (biome) shifts, rather than with the strict geographical history of clades (Lavin et al., 2004; Koenen et al., 2013). Our results also imply that continental vicariance did not directly influence Velloziaceae diversification, as the pattern of rate shifts is more complex, independently nested within the two South American lineages, and requiring hidden states unrelated to continental areas of occurrence per se. Vicariance may have been important, however, to the diversification of the Andean representatives of the family (Mello-Silva et al., 2011): the genus Barbaceniopsis and the species Vellozia andina split from their sister lineages around 23–21 Mya, concomitant with the initial South/Central Andean uplift (Hoorn et al., 2010).

Vellozia: adaptive radiation in the campos rupestres?

The core-Vellozia clade shows a trajectory of diversification in line with a ‘mature’ evolutionary radiation (sensu Linder, 2008), i.e. an initial burst followed by declining rates. This could reflect niches filling toward an equilibrium (Benton and Emerson, 2007; Rabosky, 2014), and/or greater extinction as niches become more finely divided among proliferating species (Jansson and Dynesius, 2002; Linder, 2008). A previous study found that diversification of water-related adaptations could be important to niche divergence in Vellozia (Alcantara et al., 2015). The development of conspicuous aerial stems (as a surrogate trait for drought resistance and loss of desiccation tolerance in these species) associated with higher diversification rates in representatives of the Vellozia + Barbaceniopsis clade corroborates this (Fig. 2).

The Oligocene/Miocene transition time inferred for the diversification increase in Vellozia is coincident with a global increase in temperature, fossil angiosperm pollen diversity, particularly of Poaceae and Asteraceae, and areas of C₃ grassland dominance (Jaramillo et al., 2006). We hypothesize that in this environment, the development of conspicuous aerial stems (associated with drought resistance) allowed greater plant productivity and size (Alcantara et al., 2015), and enabled Vellozia to escape restricted occurrence in exposed outcrops and expand to open grasslands. It would have allowed the colonization of warmer microhabitats with a deeper soil layer, surrounding the exposed rock outcrop habitats characteristic of campos rupestres. In these soils, seasonal water availability is less restricted, and consequently the competition with other grassland-related species more intense (Alpert, 2006). Thus, this ecological release can be seen as a key opportunity for Vellozia radiation. A better comprehension of the adaptive radiation of Vellozia in the campos rupestres will require a deeper understanding of the origins of alternative drought resistance strategies in the family. It will also be important to consider how other ecological features, such as responses to nutrient soil availability (Ayensu, 1973; Oliveira et al., 2016) and altitude (Mello-Silva, 1995), root specializations (Oliveira et al., 2015), fire resistance (Conceição and Orr, 2012), and interactions with pollinators and herbivores (Sazima and Sazima, 1990, Landau et al., 1999; Jacobi and Sarto, 2007), might have contributed to diversification in Vellozia.

The ecological diversity and endemism of Vellozia and other plant clades of the campos rupestres have been long cited as examples of adaptive radiation (Coile and Jones, 1981; Harley, 1988). Givnish (2015) argues that adaptive radiation does not require explosive diversification in its initial phase; instead, its evolutionary signature is simply closely related to species occupying different ecological niches in sympatry, or to a convergence of adaptive life forms in distantly related species of different lineages occupying similar niches in different locations. The only previous study to test for adaptive diversification in a campos rupestres lineage found no signal of ecological divergence, despite the fact that Minaria (Apocynaceae) radiated only around 1.5 Mya (Ribeiro et al., 2014). Other recent studies have inferred dated molecular phylogenies of campos rupestres lineages and found patterns that suggest evolutionary radiations, but did not explicitly test adaptive hypotheses (Koenen et al., 2013; de Souza et al., 2013; Rando et al., 2016). Our understanding of the evolution and distribution of ecological strategies in this habitat is still in its infancy, preventing a detailed assessment of how ecological divergence has contributed to lineage diversification. Thus, further studies focusing on the evolution and characterization of life forms (e.g. Mello-Silva, 1995; Alves and Kolbek, 2010) and ecological strategies in Vellozia (Oliveira et al., 2005, 2015, 2016; Teodoro, 2014; Alcantara et al., 2015), within this background of diversification rate slowdown, are promising routes toward a better understanding of adaptive radiation in the campos rupestres.

Diversification in Barbacenia: habitat specialization and ongoing high speciation

The rate shift inferred within Barbacenia between 15 and 8 Mya (approx. 13.4 Mya) is consistent with a ‘pull-of-the-present’ effect observed in constant birth–death models (Stadler, 2013) and a high turnover rate characteristic of young radiations (sensu Linder, 2008). It does not exhibit the density-dependent signature of declining rates with the filling of ecological niche space (Morlon et al., 2011). Similar patterns have been observed in South African Aizoaceae (Klak et al., 2004) and South American Lupinus (Hughes and Eastwood, 2006), although the age of the shift in Barbacenia is notable for being relatively old compared to these and similar cases (see Linder, 2008). It coincides with the period following the Middle Miocene Climatic Optimum (approx. 15 Mya), marked by a drying/cooling climate (Zachos et al., 2008). In Hoffmannseggella (Orchidaceae), diversification was associated with the expansion of campos rupestres outcrops during the Miocene cooling, as surrounding forested areas retracted (Antonelli et al., 2010). In that case, populations restricted to the outcrops expanded their geographical range and reached secondary contact before the completion of reproductive isolation in allopatry, leading to speciation by hybridization (Antonelli et al., 2010). We have as yet no evidence that hybridization has contributed to diversification in Barbacenia. Alternatively, based on the restrictive conditions of life in exposed outcrops (e.g. Alpert, 2006; Porembski and Barthlott, 2000), we tested whether a morphofunctional trait associated with such stressing conditions is associated with higher diversification rates.

The acquisition of aquiferous parenchyma between bundles in the leaves, a trait that within Velloziaceae is mostly present in species of Barbacenia, could have improved survival in drier and colder conditions by increasing water accumulation within the plant (Warming, 1893; Menezes, 1973). The ensuing expansion of Barbacenia populations might have led to accelerated diversification by increased habitat specialization and allopatric isolation. Many species of Barbacenia exhibit very small distribution ranges (Mello-Silva, 2009), and several of those are restricted to shadow and colder rocky crevices/caves found near waterfalls, in contrast to the majority of Velloziaceae, which are heliophytes. This habitat specialization strategy might have provided special advantages in the relatively small competition pool in the outcrops (Alpert, 2006).

In summary, contrasting ecological strategies seem to have driven the radiations of Barbacenia and Vellozia. While a preference for rock outcrops generally characterizes Velloziaceae as a whole, Vellozia apparently escaped this ecological restriction by evolving drought resistance (as indicated by aerial stems). Barbacenia, by contrast, evolved the existent desiccation tolerance mixing it with water retention in the leaves (i.e. by the aquiferous parenchyma between bundles) and thus became even more specialized to the outcrops during the colder and drier late Miocene. A better characterization of the habitat and fine-scale distribution of Barbacenia and Vellozia under a phylogenetic perspective is currently underway (S. Alcantara et al., in preparation) and will investigate these hypotheses in more detail.

We thank J. R. Pirani for his inspiring work on the flora of campos rupestres and encouragement at the initial steps of this work. D. D. Ackerly, G. S. Teodoro and R. S. Oliveira provided invaluable insights about the functional traits evaluated. J. Lovo, J. Paula-Souza, K. Drequeceler and M. Sousa-Baena provided help for field and lab work; K.D. also provided the pictures of the anatomical sections. J. G. Rando collected and registered the specimens of B. celiae. W. Lobin and M. Weigend provided access to several Xerophyta specimens. IBAMA-SISBIO, IEF-MG and SEMA-BA provided the licences for fieldwork and collection; and the staff of the Parque Nacional da Serra do Cipó, PN das Sempre-Vivas, PN da Serra da Canastra, PN da Chapada Diamantina, PN da Chapada dos Veadeiros, Parque Estadual Pico do Itambé, PE de Grão-Mogol, PE do Rio Preto, PE da Serra do Cabral, PE do Morro do Chapéu and Parque Municipal do Pico das Almas kindly provided assistance and/or authorizations for fieldwork. We also thank staff and students from the Botany departments at USP (São Paulo, Brazil), FMNH (Chicago, EUA), UFSC (Florianópolis, Brazil) and from the Pritzker Laboratory of Molecular Systematics. The Handling editor (Dr G. Petersen), Dr D. Cardoso and one anonymous referee provided invaluable improvements towards the final version of the manuscript. This work was supported by the Fundação de Amparo à Pesquisa do Estado de São Paulo [FAPESP 2010/50923-0 and 2013/01780-0 to S.A. and 2011/52174-4 to R.M.S.] and the Conselho Nacional de Desenvolvimento Científico e Tecnológico [CNPq Bolsa de produtividade to R.M.S.].

Data archival location: GenBank.