Succulent plants are widely distributed, reaching their highest diversity in arid and semi-arid regions. Their origin and diversification is thought to be associated with a global expansion of aridity. We test this hypothesis by investigating the tempo and pattern of Cactaceae diversification. Our results contribute to the understanding of the evolution of New World Succulent Biomes.
We use the most taxonomically complete dataset currently available for Cactaceae. We estimate divergence times and utilize Bayesian and maximum likelihood methods that account for nonrandom taxonomic sampling, possible extinction scenarios and phylogenetic uncertainty to analyze diversification rates, and evolution of growth form and pollination syndrome.
Cactaceae originated shortly after the Eocene–Oligocene global drop in CO2, and radiation of its richest genera coincided with the expansion of aridity in North America during the late Miocene. A significant correlation between growth form and pollination syndrome was found, as well as a clear state dependence between diversification rate, and pollination and growth-form evolution.
This study suggests a complex picture underlying the diversification of Cactaceae. It not only responded to the availability of new niches resulting from aridification, but also to the correlated evolution of novel growth forms and reproductive strategies.
Arid and semi-arid environments currently comprise the most extensively distributed terrestrial biomes on Earth (McGinnies, 1979; Peel et al., 2007). Under these climatic conditions, drought-stress represents a strong selective pressure on lineages to evolve morphological, physiological and ecological characteristics that allow them to successfully survive and reproduce (Axelrod, 1972; Gibson, 1996). Succulent plants reach their highest diversity under these climatic regimes, as conspicuous elements of a worldwide distributed biome of tropical dry forest, bushland and thicket vegetation poor in grasses and lacking tolerance to fire disturbance – the succulent biome (Schrire et al., 2005). Examples of the succulent biome are the Succulent Karoo in southern Africa, the spiny thicket in Madagascar and the Caatinga in Brazil (Lavin et al., 2004). Although xeric-adapted plants have evolved a wide diversity of strategies to respond to limited water availability, succulent plants most clearly display the relationship between their morphological traits and climatic conditions. It is perhaps for this reason that species-rich succulent lineages (i.e. lineages within families Cactaceae, Aizoaceae and Agavaceae) have have been suggested to originate and radiate in response to a global aridification trend during the late Miocene/Pliocene leading to present-day arid and semi-arid climates (Arakaki et al., 2011; and references therein). Nevertheless, large succulent radiations are not restricted to arid environments. Possibly the richest lineage of succulents is the epiphytic Epidendroideae (Orchidaceae), a group most typical of humid, tropical or temperate conditions (Silvera et al., 2009; Nyffeler & Eggli, 2010). In addition, not all succulent lineages are species rich or morphologically diverse, as exemplified by Halophytaceae (1 species), Moringaceae (13 species) and Fouqueriaceae (11 species; Nyffeler & Eggli, 2010). Hence, succulent plant diversification may not only respond to aridification.
Different biomes most likely originated in response to aridification trends during the Neogene, for example, the Australian arid zone biota (Byrne et al., 2008), the California floristic province (Calsbeek et al., 2003) and the Succulent Karoo (Verboom et al., 2009), given that the origin and expansion of arid conditions can constitute a source of novel ecological niches. Yet, few studies have provided detail on the possible ecological mechanisms driving the diversification of species-rich lineages inhabiting these environments, specifically proposing mechanistic explanations for the relation between aridification and an increase in species diversity, or providing tests for alternative explanations (but see Ellis & Weis, 2006; Ellis et al., 2006; Schnitzler et al., 2012). Aridity by itself poorly explains the dramatic increase in species numbers and the morphological and ecological diversity in certain succulent lineages, such as subfamily Ruschioideae (Aizoaceae; Klak et al., 2004), the genus Euphorbia (Euphorbiaceae; Bruyns et al., 2011; Horn et al., 2012) or the cactus family (Hernández-Hernández et al., 2011), where additional biotic and abiotic factors may have played important roles. For instance, studies of certain genera within Ruschioideae suggested that diversification occurred primarily via local adaptations to edaphic microhabitats (Ellis & Weis, 2006; Kellner et al., 2011). Additionally, Good-Avila et al. (2006) suggested that the evolution of monocarpy and bat pollination drove a second burst of diversification in Agave (Agavaceae), leading to its present diversity.
The importance of the evolution of particular pollinating systems in the radiation of lineages with increased diversification rates has been stressed (Johnson, 2010), and it may be also relevant for the diversification of several lineages within Cactaceae (Mandujano et al., 2010; Schlumpberger, 2012). The overlapping geographic distribution of Agave and columnar Pachycereeae (Cactaceae) may indicate a common pattern of diversification in response to the evolution of similar pollination mechanisms (Valiente-Banuet et al., 1996; Molina-Freaner & Eguiarte, 2003). General floral evolutionary trends in Cactaceae involve shifts from a plesiomorphic bee pollination to other pollination syndromes (Reyes-Agüero et al., 2006; Mandujano et al., 2010; Schlumpberger, 2012; and references therein). In late-diverging members of Cactaceae, the floral pericarpel forms a hypanthium, promoting the appearance of a large diversity of bird-pollinated flowers and allowing adaptation to a wider array of pollinator guilds, such as sphingids and bats (Schlumpberger, 2012, p. 304). Members of subfamily Cactoideae also exhibit the widest range of growth forms, from barrels and small globoses to massive columnars or epiphytic shrubs (Hernández-Hernández et al., 2011). The combination of particular growth forms and pollination guilds might confer an advantage for survival, reproduction and geographical expansion.
Although cacti are present in most biomes throughout the New World, there are geographical hotspots that host numerous species, and some of the most species-rich clades occupy distinct geographical areas. For example, tribe Cacteae is mainly distributed in the southern Chihuahuan Desert (Hernández & Gómez-Hinostrosa, 2005); Hylocereeae in southern Mexico and Central America; Rhipsalideae in southeastern Brazil (Calvente et al., 2011); core Pachycereeae in the Sonoran Desert; and Trichocereeae in South America (Nyffeler, 2002; Hernández-Hernández et al., 2011). A biogeographic analysis of Cactaceae using a well-sampled phylogeny at a continental level can help to elucidate whether there is geographic structure in their evolutionary history, indicating if particular biogeographic histories played a substantial role in the diversification of certain lineages.
In this study, we use Cactaceae as a system to investigate the diversification of plant lineages in arid and semi-arid regions of North and South America by testing specific hypotheses about the mechanisms promoting speciation. Insights about the origin and diversification of Cactaceae can also provide clues regarding the origin and expansion of arid biomes in the New World. Our aim is to evaluate if diversification within the family proceeded mostly as synchronous bursts – potentially reflecting a global colonization trend of arid habitats – or if they occurred at different times – for example, if there is a time lag between origination and rapid radiation in different clades, or a lag between aridification and further colonization (Guerrero et al., 2013). Arakaki et al. (2011) concluded that major cactus radiations were contemporaneous with radiations in core Ruschioideae in South Africa and agaves in North America, suggesting an association with the global expansion of arid and semi-arid environments. In this study, we investigate the diversification of Cactaceae in greater detail by conducting independent analyses using a phylogeny with a considerably improved taxonomic sampling, and evaluating the potential relevance of growth form, pollination syndrome and biogeographic history in the distribution of species richness within the family. Our results contribute to better understanding the evolutionary history and complex diversification mechanisms of cactus lineages, and provide specific hypotheses for further studies of the possible ecological mechanisms leading to increases in diversification rates in species-rich lineages in New World arid environments.
Materials and Methods
Phylogenetic analysis and divergence time estimation
We used the most taxonomically complete chloroplast dataset currently available for Cactaceae at the genus level, which includes a taxonomic sample of 224 species belonging to 108 genera, representing c. 85% of the generic diversity in the family (Hernández-Hernández et al., 2011). It comprises concatenated DNA sequences data from four plastid molecular markers: rpl16 intronic region, trnL-trnF and trnK-matK intergenic spacers, and the protein coding matK, for a total of 5590 base pairs, which were recently used to elucidate phylogenetic relationships in the family (Hernández-Hernández et al., 2011). Four additional species of Anacampserotaceae, the closest relatives of Cactaceae (Nyffeler, 2007), were included as outgroups. See Supporting Information Table S1 for the list of species, as well as accession and voucher information.
Sequences of each region were aligned automatically using MUSCLE (Edgar, 2004), followed by manual refinement using BioEdit v5.0.6 (Hall, 1999; dataset available in TreeBase, Study ID 11087). To corroborate phylogenetic relationships reported previously (Hernández-Hernández et al., 2011) and evaluate clade support, maximum likelihood (ML) phylogenetic analyses were conducted with RAxML v7.0.4 (Stamatakis, 2006). Each of the four plastid regions were assigned independent general time reversible (GTR) substitution models with 25 rate categories to model gamma-distributed site-specific rate heterogeneity. Clade support was assessed with 1000 replicates of a nonparametric bootstrap analysis, also conducted with RAxML.
Because fossils that could provide reliable absolute age calibrations near Cactaceae are not available, we implemented a two-step approach to estimate divergence times. First, the age of Cactaceae was estimated based on a representative dataset of eudicots (Eudicotyledoneae) and outgroups. Second, the credibility interval around the estimated age of Cactaceae obtained in the first step was used to calibrate the phylogenetic tree of Cactaceae described earlier. In the first step we used a dataset of 109 species belonging to 34 eudicot orders (including 21 families of Caryophyllales), and representatives of two eudicot outgroups: Ceratophyllales and Monocotyledoneae. We assembled a dataset with nucleotide sequences of the platstid protein-coding genes atpB, rbcL and matK, which were downloaded from GenBank, for a total of 4510 bp (Table S2). Divergence times were estimated with the uncorrelated lognormal (UCLN) relaxed clock available in BEAST v1.6.1 (Drummond & Rambaut, 2007; Methods S1). Twenty three internal nodes were calibrated with critically evaluated fossil-derived ages (Table S3), implemented as lognormal priors in which the mean was equal to fossil age + 10%, and the zero offset was equal to fossil age – 5 million yr (age given in units Million years ago, Ma). Divergence dates within Cactaceae were obtained with the plastid dataset used for phylogenetic estimation (see earlier), using the UCLN relaxed clock in BEAST, with equivalent conditions as described above. The prior age of the root node (stem group Cactaceae) was given a uniform distribution between 22.71 and 42.43 Ma, corresponding to the 95% highest posterior density (HPD) of the age of this node obtained in the eudicot-level analysis (Methods S1).
The absolute rates of diversification of strongly supported clades within Cactaceae were calculated using methods described in Magallón & Sanderson (2001). These estimators consider the standing species richness and the stem or crown age of a clade in the context of different extinction scenarios, providing absolute estimates of clade net diversification (r = speciation (λ) – extinction (μ)) conditional on its survival to a given time t (the present). Because absolute extinction rates for clades are unknown, diversification rates were alternatively estimated assuming no extinction (ε = 0.0), and a high relative extinction rate (ε = μ/λ = 0.9).
Assignment of species richness to clades within Cactaceae was not straightforward due to conflicts between classic taxonomic treatments and recent molecular-based phylogenies. We estimated the number of living species in well-supported major clades recovered in our Cactaceae phylogenetic tree and consistently reported in the literature with high support values (i.e. Gibson & Horak, 1978; Gibson, 1982; Wallace, 1995, 2002; Wallace & Cota, 1996; Porter, 1999; Porter et al., 2000; Butterworth et al., 2002; Nyffeler, 2002; Wallace & Dickie, 2002; Arias et al., 2003, 2005; Butterworth & Wallace, 2004, 2005; Edwards et al., 2005; Ritz et al., 2007; Griffith & Porter, 2009; Bárcenas et al., 2011; Calvente et al., 2011; Hernández-Hernández et al., 2011; Majure et al., 2012; Schlumpberger & Renner, 2012), based on the species richness of included genera (Hunt et al., 2006; Table S4).
In order to analyze cactus diversification in a phylogenetic and temporal context we implemented MEDUSA (Alfaro et al., 2009), an extension of the birth–death likelihood model (Rabosky, 2006) that allows clade-specific birth–death models, and can detect regions in a time-calibrated phylogeny where diversification rate shifts likely occurred. MEDUSA allows performing diversification analyses in phylogenies lacking a complete taxonomic sampling, as it incorporates a taxonomic likelihood for unresolved terminal clades. We pruned our original 224 taxon phylogeny to a genus-level tree, and assigned diversity to each terminal. To account for phylogenetic uncertainty, we report mean diversification rates and the most frequent rate shifts detected within Cactaceae after running MEDUSA on each of 1000 trees randomly selected from the BEAST posterior distribution.
In order to maximize the congruence with other studies, we followed the biogeographic scheme of Posadas et al. (1997) and Morrone (2001, 2002, 2006) to define operational areas. However, we slightly modified these schemes by observing the geographic ranges of cacti species, and considering regions with high species richness and endemism (Ortega-Baes & Godínez-Alvarez, 2006). We designated the following areas (Fig. 1): (A) Sonoran Desert and Sierra Madre Occidental; (B) Chihuahuan Desert; (C) Central Mexico; (D) the Antilles; (E) Central American and South American Tropical areas; (F) the Andean region of Peru; (G) Andean region of Chile and Argentina; (H) the Caatinga; (I) the Chaco area; and (J) Northern Subantarctic region. For details see Methods S2.
We reconstructed ancestral geographic ranges by utilizing parsimony, ML and Bayesian methods. For the ML framework we used the Dispersal-Extinction-Cladogenesis geography-based method (DEC; Ree, 2005; Ree & Smith, 2008) implemented in LAGRANGE, using the maximum clade credibility (MCC) tree obtained in the BEAST analyses. We used a uniform dispersal matrix to avoid overparameterization. For the parsimony and Bayesian frameworks we (respectively) performed S-DIVA and BBM (Bayesian Binary MCMC) analyses implemented in RASP v2.0b (Yu et al., 2010, 2011; Ali et al., 2012). These methods accommodate phylogenetic uncertainty by averaging the ancestral reconstructions over a sample of user-supplied trees, in this case, the 1000 randomly selected trees obtained with BEAST (see earlier). The ancestral ranges estimated at each node on the MCC chronogram were obtained. The number of maximum unit areas allowed for nodes was set to six. In the BBM analyses we set a null distribution for the ancestral range of the root of the tree, and we ran ten MCMC chains simultaneously for 5 × 104 generations, sampling a state every 100 steps, and discarding the first 100 samples as burnin.
Ancestral states, character coevolution, and diversification correlates
We estimated ancestral states of growth form and pollination syndrome under ML and Bayesian frameworks using BayesMultiState implemented in BayesTraits v1.0 (Pagel & Meade, 2006). For ML analyses, we used the BEAST MCC tree, while for the Bayesian analyses we implemented a series of reversible-jump hyperprior (RJHP) MCMC analyses (Pagel & Meade, 2006) on the 1000 randomly selected BEAST trees. The RJHP approach approximates the posterior distribution of ancestral character states while accounting for phylogenetic uncertainty. We ran three independent chains of 1×108 generations with a conservative initial burnin of 25% generations. We used Tracer v1.5 (Rambaut & Drummond, 2007) to check for chain convergence and ESS values.
Growth-form character states analyzed by Hernández-Hernández et al. (2011) were here re-scored as binary characters. We assigned (0) to species with a globose solitary, globose caespitose or barrel growth form; and (1) to species with an arborescent, shrubby or columnar growth form. Species within subfamilies Opuntioideae, Maihuenioideae and Pereskioideae lack a growth form comparable to those of members of subfamily Cactoideae; hence, their growth form was scored as nonapplicable. The most widespread, and possibly ancestral, pollination syndrome of Cactaceae is mellitophily or bee pollination (Mandujano et al., 2010; Schlumpberger, 2012), with chiropterophily (bats), ornithophily (birds) and sphingophily (moths) pollination syndromes appearing in derived clades. We thus assigned a pollination guild to species within our phylogeny by reviewing the literature and by examining floral characters in specimens and photographs. We then coded these as binary characters, by assigning (0) to mellitophilic species and (1) to species possessing derived pollination syndromes (chiropterophily, ornithophily and sphingophily).
We conducted a character correlation test for growth form and pollination syndrome using the Discrete module in BayesTraits v1.0 (Pagel, 1994, 1997; Pagel & Meade, 2006), to test for a possible correlation between the appearance of derived pollination syndromes and of a columnar, arborescent or shrubby growth form. Likelihood scores of competing models were obtained with BayesDiscrete (Pagel & Meade, 2006) on the BEAST MCC tree, and their relative fit was compared using a likelihood ratio test (LRT). We also evaluated correlation in a Bayesian context, calculating Bayes factors as the difference between the harmonic mean of marginal log-likelihood (logeL) scores from MCMC runs for the competing models (Newton & Raftery, 1994). For these, we ran the correlated and independent models for 1 × 108 generations each over the 1000 BEAST trees, with a conservative initial burnin of 25% generations. These analyses also implemented a RJHP under the same parameters as in the ancestral character reconstruction analyses. A Bayes factor of 5 or greater can be considered as strong evidence for correlated evolution, whereas a Bayes factor smaller than zero supports the independent model (Kass & Raftery, 1995; Pagel & Meade, 2006).
In order to test if diversification rates (as well as speciation and extinction separately) are correlated with character states of growth form and pollination syndrome, we used BiSSE (Binary State Speciation and Extinction) in Diversitree v0.9-1 (Maddison et al., 2007) in a ML framework. This method explicitly incorporates character state change directly into the likelihood assessment of speciation and extinction rates (Maddison et al., 2007). BiSSE includes six state-specific parameters (for states 0 and 1): two speciation rates (lambda0 and lambda1), two extinction rates (mu0 and mu1), and two rates of character state change (q01 from state 0 to 1, and q10 from state 1 to 0). To test hypotheses we compared the AIC scores obtained implementing six different models on the BEAST MCC tree. If diversification rates are correlated with character states, unconstrained models should be favored over the constrained model where parameters lambda and mu are set to be equal.
Phylogeny, divergence dates and ancestral areas
The estimated phylogeny of Cactaceae is shown in Fig. S1. The ML analysis resulted in a tree with strong support for major Cactaceae clades, congruent with previous studies (e.g. Nyffeler, 2002; Bárcenas et al., 2011; Hernández-Hernández et al., 2011). Only strongly supported monophyletic lineages were further considered in diversification analyses. The inferred phylogenetic relationships among eudicot orders also conform to relationships found in independent studies (e.g. Wang et al., 2009; Soltis et al., 2011; see Fig. S2).
Cactaceae is estimated to have split from its sister group (i.e. the stem group age) at 32.11 Myr ago (Ma), with a credibility interval (95% highest posterior density (HPD)) spanning between 42.43 and 22.71 Ma. The age of the onset of branching leading to living lineages (i.e. the crown group age) was estimated at 26.88 Ma (37.1–16.67 Ma 95% HPD). The eudicot-level chronogram is shown in Fig. S2 and the Cactaceae chronogram is shown in Fig. 1. The ages of major, well-supported Cactaceae clades are shown in Table 1.
Table 1. Estimated ages and diversification rates of clades supported by high bootstrap values
Stem group age (Ma)
Crown group age (Ma)
Eqn 6, stem Epsilon = 0
Eqn 6 stem Epsilon = 0.9
Eqn 7 crown Epsilon = 0
Eqn 7 crown Epsilon = 0.9
Numbers in bold indicate the highest estimated diversification rates.
Ancestral areas estimated with among parsimony, ML and Bayesian methods were congruent (Table S5), and we selected the results from the Bayesian analyses (Fig. 1) for further discussion. The Andean region of Chile and Argentina, including the southern Andean region of Bolivia, was found to be the most probable area of Cactaceae origin, as well as of Opuntioideae and Cactoideae (Fig. 1b, Table S5). According to our results, expansion towards unoccupied regions occurred gradually within each lineage. The ancestor of Cacteae was inferred to occupy the Chihuahuan Desert, after a presumed long dispersal from the Andean region of Chile, Argentina and Bolivia.
The absolute diversification rates (r) estimated for major, well-supported Cactaceae clades, considering no extinction (ε = 0.0) and a high relative extinction rate (ε = 0.9), are shown in Table 1. The average diversification rates estimated for crown Cactaceae (r0.0 = 0.244 speciation events per million years (sp Myr−1) for ε = 0, and r0.9 = 0.182 sp Myr−1 for ε = 0.9) are considerably higher than those estimated for Caryophyllales (r0.0 = 0.068 and r0.9 = 0.084 sp Myr−1), and for angiosperms as a whole (r0.0 = 0.042 and r0.9 = 0.048 sp Myr−1; Magallón & Castillo, 2009). Lineages with the highest diversification rates are Opuntieae (particularly the Opuntia + Nopalea clade), the Mammilloid clade (particularly the core Mammilloid clade), the PHB clade (particularly the core Pachycereeae, Pachycereinae, Stenocereinae and Hylocereae clades, and the genus Echinocereus), and the BCT clade (particularly the Trichocereeae clade and the genus Gymnocalycium; see Table 1). These clades, together with several other Cactaceae major groups, exceed the upper 95% confidence interval of the species diversity through time expected for a clade that diversifies with a rate equal of that of Caryophyllales or Cactaceae as a whole, and under a scenario of high extinction rate (Fig. 2).
MEDUSA detected several shifts in diversification relative to background levels (Fig. 3). The most frequently detected increases in diversification rates occur in the Opuntia + Nopalea clade, at the base of the Cactoideae subfamily, in the Mammilloid clade with high rates at the Mammillaria + Coryphanta clade, and in the BCT clades. Shifts occurring at lower frequency (at least 50% of the random trees) include the terete-stemmed Grusonia + Cylindropuntia clade and the PHB clade.
Character evolution and diversification correlates
Ancestral character state reconstructions of pollination syndrome and growth form in major Cactaceae clades are shown in Table S6, and results from the ML analyses are illustrated in Fig. 4. Those ancestral character states reconstructed with the highest probability values were consistent in both the ML and the Bayesian analyses. A mellitophilic pollination syndrome was found to be ancestral in Cactaceae (P = 0.73/0.99 for ML/Bayesian analyses, respectively), with shifts towards other pollination syndromes in the opuntioid flattened-stemmed clade (including Tacinga, Opuntia and Nopalea) (P = 0.99/0.99), and in the core Cactoideae only with Bayesian analyses (P = 0.96, P = 0.46 in ML). However, these two clades include mellitophilic members. The ancestral growth form of the family could not be reconstructed because representatives of subfamilies Opuntioideae and Pereskioideae lack comparable growth forms with members of Cactoideae. In Cactoideae, an arborescent, shrubby or columnar growth form was reconstructed as ancestral (P = 0.9/1.0), with shifts towards a globose or barrel form early in Blossfeldia, in the North American Cacteae clade (P = 0.99/0.99) and the South American core Notocacteae clade (P = 0.99/0.99), with independent shifts scattered throughout the BCT clade (Fig. 4). To statistically evaluate a possible concordance in pollination guild and growth-form shifts along the Cactaceae phylogeny we conducted a correlation test with BayesTraits. With the ML approach, we conducted LRTs for each of the 1000 randomly selected trees. Probability values obtained were very low (lower than 5 × 10−4; see Fig. S3), suggesting a significant improvement with the correlated model, and thus a possible correlation between the evolution of growth form and pollination syndrome. Similar results were obtained with the Bayesian MCMC approach. The correlated model yielded an harmonic mean of c. −136.38, while that of the independent model was c. −213.59, leading to a Bayes factor of 154.4, supporting the correlated evolution between growth form and pollination syndrome.
In order to investigate a correlation between character change and diversification in Cactaceae, we implemented variants of the BiSSE model. Results on parameter estimates and the comparison of AIC scores obtained for growth form and pollination syndrome models are shown in Tables S7 and S8, respectively. Results show a clear state-dependence in diversification rates, as speciation rates were on average two times higher in lineages with an arborescent, shrubby or columnar growth form, and with a derived pollination syndrome such as chiropterophily, ornithophily and sphingophily. In both traits there is a clear difference in transition rates, which are on average six times higher in favor of a transition rate from a globose to an arborescent or columnar form, and from a mellitophilic towards a derived pollination syndrome.
Cactaceae diversification is congruent with a Miocene origin of the American Succulent Biome
The Succulent Biome comprises highly fragmented, globally distributed patches of vegetation on a climate characterized by erratic, unpredictable rains, with succulents as one of the predominant growth forms together with sclerophyllous shrubs that do not resist fire (Lavin et al., 2004; and references therein). Plant groups in this biome show a strong geographic phylogenetic structure, and a considerable number of endemic and geographically-restricted species (Lavin et al., 2004; Schrire et al., 2005; Thiv et al., 2011). This biome is characterized by low immigration rates and the presence of local specialists; many genera shared among patches, but few shared species (Thiv et al., 2011). To understand the origin and expansion of the succulent biome in the New World, we investigate the geographic phylogenetic structure and biogeographic history of Cactaceae, and evaluate if clades with distinct geographic distributions originated contemporaneously, possibly as a response to a global aridification trend.
The cactus family has been previously used as a model to understand the origin of arid biomes. Arakaki et al. (2011) provided an estimation of divergence dates and diversification rates of Cactaceae by using a similar approach to the one presented here. They also used a two-step approach for the estimation of dates. First, they used an assembled order-level matrix to estimate Cactaceae crown dates with a similar number of taxa, but including 83 chloroplast genes and 13 calibration fossil dates. In spite of that effort to increase the number of genomic regions, general relationships within eudicots and outgroups obtained by Arakaki et al. are similar to the ones presented here, and to relationships reported in previous studies (e.g. Wang et al., 2009; Soltis et al., 2011). Consequently, we focused our efforts on the calibration strategy by increasing the number of temporal calibrations to include 23 divergence events obtained from the fossil record. Additionally, we used a lognormal distribution for each calibration to accommodate uncertainty in the paleontological information in the priors rather than relying upon single data points. Our smaller gene dataset allowed utilization of an uncorrelated lognormal clock method that has the advantage of considering differences in the rates of substitution among lineages without assuming rate autocorrelation (Drummond & Rambaut, 2007). Furthermore, we included more than twice the number of Cactaceae representatives with respect to the previous study, which allowed a more detailed investigation of diversification dynamics within the family.
Although our estimated dates for crown and stem Cactaceae are similar to those obtained by Arakaki et al. (2011), our estimated dates for clades within Cactaceae are generally younger (Table 1), and consequently our inferred diversification rate estimates are higher. According to our estimates, Cactaceae became differentiated c. 32.11 (42.43–22.71) Ma, slightly after the Eocene–Oligocene boundary, and the onset of the diversification of the family into extant lineages is estimated in the late Oligocene, c. 26.88 (36.85–22.71) Ma. These dates are younger than the large global drop in atmospheric CO2 concentrations occurring in the Eocene (Zachos et al., 2008), indicating a possible scenario for the origin of the ancestors of modern succulent lineages with crassulacean acid metabolism. Moreover, the estimated times of origin of other succulent groups are generally similar. For example, the origin of Euphorbia (Euphorbiaceae) was recently estimated at 36.59 (47.24–28.99) Ma (Bruyns et al., 2011); Aizoaceae and Didiereaceae (Caryophyllales) at c. 32.3 and 28.25 Ma, respectively (Hernández-Hernández, 2010), and Agavaceae (Asparagales) at c. 35–30 Ma (Good-Avila et al., 2006). However, extreme succulence and other specialized adaptations to dry habitats are derived conditions within each of these groups (Applequist & Wallace, 2000; Klak et al., 2003; Bruyns et al., 2011; Hernández-Hernández et al., 2011; Horn et al., 2012), suggesting that these attributes evolved from nonsucculent ancestors that were presumably preadapted to xerophytic habitats in the Oligocene, under warmer, more stable climates (Graham, 2011).
Although evidence for increasing aridity since the Cretaceous is available (Ziegler et al., 2003), floras of arid environments apparently are substantially younger, mostly being no older than the late Miocene or early Pliocene (Moore & Jansen, 2006; and references therein), especially in the New World. Modern vegetation of the Sonoran Desert has been estimated to date from the Pleistocene (2.59 Ma; Axelrod, 1979); the Pliocene–Pleistocene (5.33–2.59 Ma; Phillips & Comus, 2000); or the middle Miocene (13.82 Ma; Van Devender, 2000). Arid conditions leading to what is now the Chihuahuan Desert are thought to have appeared in the middle Miocene (Morafka, 1977), and fossil grasses from Mojave Desert suggest that Mediterranean chaparral-type grasslands were established there also by the Miocene (Tidwell & Nambudiri, 1989). By the end of the early Miocene, New World ecosystems included early versions of the desert, shrubland, savanna and grassland biomes, derived from drier elements present in older habitats (Graham, 2011, and references therein).
Our dating analysis indicates a time lag between the origin and diversification of Cactaceae, with the latter taking place mainly during the last 15–10 Myr (Table 1, Fig. 1), in agreement with a proposed Miocene expansion of New World arid and semi-arid vegetation. The estimated time of diversification of other (nonsucculent) North American xerophytic plant genera also agree with increasing aridity during the Miocene. For instance, Prosopis (Fabaceae) – the mesquite – originated during the late Miocene, but diversified during the Pliocene (Catalano et al., 2008); the strongest diversification phase of Bursera (Burseraceae) took place during the Miocene (De-Nova et al., 2012); Triquila (Boraginaceae) diversified in the early to late Miocene (Moore & Jansen, 2006); and Nolana (Solanaceae; Dillon et al., 2009), Agave (Agavaceae; Good-Avila et al., 2006) and Ephedra (Gnetophyta; Loera et al., 2012) diversified in the middle to late Miocene. The apparent contemporaneous origin of the largest clades, as well as species-rich lineages in Aizoaceae and Agavaceae, has been invoked as an indicator of a global aridification trend (Arakaki et al., 2011). Nonetheless, it is difficult to reject an independent origin of each cactus lineage given the large 95% HPD intervals that were obtained for estimated dates (see Fig. S4). Unfortunately, given the scarcity of fossil evidence in arid biomes, it is difficult to obtain more reliable calibration dates to improve analyses and date estimates by reducing HPD intervals. An adequate test for the hypothesis of synchronous origin and diversification would require further statistical analyses with the inclusion of other xeric-adapted lineages radiating independently in different geographical regions.
Major cacti genera radiated during their colonization of North America
Our biogeographic analyses allowed us to evaluate whether radiating lineages with the highest diversification rates are geographically concentrated sharing particular areas of origin, or are dispersed through the family's distribution. In agreement with previous studies (i.e. Buxbaum, 1969; Leuenberger, 1986; Wallace & Gibson, 2002; Ocampo & Columbus, 2010), our results support a South American origin for Cactaceae, in the central Andean region of northern Chile, north-west Argentina, Bolivia and Peru (Fig. 1; Table S1). Many species of Cactaceae are endemic to this region, which has long been regarded as the source of numerous angiosperm lineages (Raven & Axelrod, 1974). Aridity in the central Andes can be traced back to the late Jurassic, and sedimentological records of the Atacama Desert reveal climate stability for the region even during the Pleistocene climatic fluctuations (Hartley et al., 2005). The Andean uplift did not commence until 30 Ma (Hartley et al., 2005 and references therein), reinforcing the arid conditions by excluding moisture from the Amazon Basin (Placzek et al., 2009). Additionally, the rise of the Andes could provide novel niches favoring cactus diversification (Hoorn et al., 2010). We estimated several independent expansions within all major Cactaceae clades from this area into different geographic regions (Fig. 1). However, the lineages that expanded into North America contain some of the most species-rich genera, which also exhibit the highest rates of diversification (Table 1).
Two opuntioid clades were associated with a diversification rate increase in the MEDUSA analysis: the flat-stemmed Opuntia + Nopalea clade and the terete-stemmed Cylindropuntia and Grusonia. Most species belonging to these clades are distributed in North America, especially in the Chihuahuan Desert (Gómez-Hinostrosa & Hernández, 2000; Hernández et al., 2001; Powell & Weedin, 2004; Griffith & Porter, 2009; Majure et al., 2012). Other increases in diversification rates were detected in clades possessing the most elevated absolute diversification rates within the family: Echinocereus and the Mammilloid clade. Echinocereus includes c. 67 species of short, cylindrical-stemmed cacti (Hunt et al., 2006), and has the highest diversification rate in the family (r0.0 = 1.01 and r0.9 = 0.57 sp Myr−1). These species are distributed in deserts and semideserts of central and northwestern Mexico, and southwestern USA. Mammillaria, which belongs to the Mammilloid clade, is the largest genus in the family, with c. 163 species (Butterworth & Wallace, 2004; Hunt et al., 2006). It also reaches its maximum species richness and morphological diversity in arid regions of Mexico, with numerous microendemic species in the Chihuahuan Desert. All of these clades represent independent expansions into North American arid biomes (Table S5, Fig. 1).
Different hypotheses have been suggested to explain the large number of species in the Opuntioid, Mammilloid and Echinocereus clades (i.e. apomixis: see Pinkava, 2002; polyploidy: see Cota, 1993; Cota & Philbrick, 1994; Cota & Wallace, 1995; Pinkava, 2002; the presence of latex and resin canals in Mammillaria: see Farrell et al., 1991); however, detailed studies on the ecological mechanisms driving speciation in these lineages are needed to confirm them.
Novel pollination syndromes and growth forms occur in lineages with high diversification rates
Although aridification might provide environmental conditions fostering the origin of xerophytic plant lineages, other mechanisms promoted elevated diversification rates in some of them. Our taxonomic sample allowed us to provide more detail on clades and lineages associated with increased diversification rates within Cactoideae, the most diverse Cactaceae subfamily, than previously reported (Arakaki et al., 2011). We detected the BCT and PHB clades (and major lineages within them) as having significantly high species richness (Table 1, Fig. 2), as well as being associated with significant increasing shifts in diversification rate (Fig. 3). The BCT clade includes South American columnar, arborescent and shrubby species from tribes Trichocereeae and Cereeae. The PHB clade includes species with similar growth forms in the North American tribes Pachycereeae and Hylocereeae, which also includes epiphytes (Hernández-Hernández et al., 2011).
The derived conditions of bird, bat and moth pollination tend to occur in members of the PHB and BCT clades (Grant & Grant, 1979; Gibson & Nobel, 1986; Barthlott & Hunt, 1993; Cota, 1993; Nobel, 2002; Fleming et al., 2009). Derived pollination syndromes can provide effective barriers to gene flow, contributing to the origin of new lineages (Xu et al., 2012), with an impact on diversification rates (Smith et al., 2008; van der Niet & Johnson, 2012 and references therein). The evolution towards bat or bird pollination might provide important benefits to plants, because these animals deposit a large amount and variety of pollen genotypes on stigmas and, compared with pollinators such as ants or bees, are long-distance dispersers (Fleming et al., 2009).
Although derived Cactoideae clades most conspicuously include shrubs (Fig. 4 and see Hernández-Hernández et al., 2011), members of the North American PHB and the South American BCT clades convergently evolved towards arborescent and columnar growth forms, with a shift towards epiphytic habit in Hylocereeae (Hernández-Hernández et al., 2011). In addition to particular floral pollinating systems, columnar or arborescent growth forms in the context of the relatively short vegetation in dry forests and semi-arid regions might facilitate pollination by moths, bats or birds (Fleming et al., 2009). A correlation between derived pollination syndromes and an arborescent, shrubby or columnar growth form in BCT and PHB members has already been suggested (Schlumpberger, 2012), and here we tested this hypothesis of correlation under ML and Bayesian frameworks.
We found strong evidence suggesting that bat, bird or moth pollination syndromes are associated with a columnar, shrubby or arborescent growth form in the Cactaceae phylogeny. We used BiSSE to test if shifts towards novel pollination syndromes and a shrubby or arborescent growth form in the Cactoideae are coincident with shifts in diversification rates. Our results strongly indicate a correlation between character state and increase in speciation rate (Tables S7, S8). Thus, we suggest that the outstanding diversification of core Pachycereeae, Hylocereeae and Trichocereeae is related to the evolution of derived pollination syndromes, with a possible trend towards a specialization for bats in North America and moths in South America, both facilitated by an arborescent, shrubby or columnar growth form. In the context of an arid environment where plant populations have low densities, the evolution of a pollination mechanism that increases pollen-transfer efficiency can be helpful to overcome mate-finding Allee effects and to continue to reproduce successfully (Ghazoul, 2005; Gascoigne et al., 2009; Livshultz et al., 2011), particularly in long-lived and slow-growing species such as many cacti.
In conclusion, our estimated dates of origin and diversification of Cactaceae and major clades within the family are congruent with a Miocene expansion of arid biomes in the American continent. However, disparity among diversification rate estimates for clades originating at similar times suggests different underlying diversification drivers. The Opuntia + Nopalea, the Mammillaria + Coryphantha, the PHB (including the core Pachycereeae and Hylocereeae), and the BCT clades (including Trichocereeae) were identified as having higher-than-expected species richness and elevated diversification rates, and to be associated with rate increases in the phylogeny. The large morphological and ecological diversity encompassed by each of these clades, as well as by other succulent lineages, and the fact that other related xeric-adapted lineages without these diversity and species-richness originated at similar times indicates that the diversification of major succulent plant radiations might be better explained by a complex set of attributes contingent to each clade evolving in each particular arid habitat.
According to our results, the high diversification rates that characterize speciose genera such as Opuntia, Mammillaria and Echinocereus might be associated with their geographic expansion during the recent aridification of North America (particularly the expansion of the Chihuahuan Desert) during the Miocene. In the case of the core Pachycereeae, Hylocereeae and Trichocereeae, which are relatively younger lineages, their high species richness may have resulted from the origin of novel pollination syndromes associated with changes in growth forms in several clade members. We hypothesize that the presence of these characters fostered an increase in the diversification rates of lineages within the BCT and PHB clades, which include the tallest members in dry forests and semi-arid regions of South and North America. Nevertheless, the ecological mechanisms that led to increases in speciation rates in these clades require more investigation.
It has been pointed out that the origin of extant biodiversity in the Neotropical region, the most species-rich region on Earth (e.g. Prance, 1977; Gentry, 1982), cannot be attributed to the action of one or a few events during key time intervals, but rather has resulted from complex ecological and evolutionary processes including both abiotic and biotic factors (Antonelli & Sanmartín, 2011; Rull, 2011). Similarly, hypotheses for the origin of biodiversity in arid regions must take into account the complexity and diversity of possible drivers of diversification in water-limited environments. It has been shown that aridification can shape the evolution not only of functional but also of reproductive traits (Livshultz et al., 2011). However, further field studies providing ecological explanations for the mechanistic processes determining diversification in xerophytic lineages are still pending.
This work was supported by CONACyT, México grant number SEP-2004-C01-46475 to L.E.E. and S.M.; and PAPIIT-UNAM grant number IN202310 to S.M. T.H-H. conducted PhD studies in the Posgrado en Ciencias Biológicas, Universidad Nacional Autónoma de México, and received a scholarship from CONACyT. The authors thank C. Jaramillo and T. H. Fleming for valuable comments and observations to this research; and R. H. Ree, V. Sosa and L. L. Sánchez-Reyes for help in different aspects of this project. The Instituto de Biotecnología, Universidad Nacional Autónoma de México (Macroproyecto de Tecnologías de la Información y la Computación) provided computing resources.