We used data on global palm species distributions from an exhaustive, authoritative checklist of the World’s palm species (Govaerts et al., 2014, data used here accessed March 2009). Though coarse in resolution, this dataset currently represents the most complete and reliable source on species distributions of all palms worldwide. The dataset records palm species presences and absences across the world within the level 3 geographic units as defined by the International Working Group on Taxonomic Databases (TDWG; Brummitt, 2001). These TDWG level 3 units mostly correspond to countries and/or major islands, such as Borneo, Madagascar, and New Guinea, but very large countries such as USA, Brazil, and China are subdivided into smaller units. We included only native palm occurrences, excluding introduced occurrences, and doubtful as well as erroneous records. To derive estimates of species richness, we summed all presences of palm species within each TDWG level 3 unit, and did this separately for climbing and non-climbing palms (see below).

The dataset contained a total of 2,445 accepted palm species names and 5,027 native occurrence records within 194 TDWG level 3 units (Kissling et al., 2012a). It does not include a recently described climbing species of Dypsis (Rakotoarinivo and Dransfield, 2010) which would raise the total number of liana species in this genus to two. The addition of this extra species would not impact the results presented here, but this new information was taken into account throughout the discussion and in Table 1.

We classified all palm species into two growth forms: (1) climbers, and (2) non-climbers (including all other growth forms such as stemmed and acaulescent palms). Palm species that show leaning growth forms such as some Bactris species (not really climbers or trees) were considered as non-climbers. For climbers, we also included palm species that can, within the same species, show both climbing and non-climbing habits (only 16 species). The remaining species (n = 2431) were exclusively climbers or non-climbers. Information on climbing habit was derived from the literature (Russell, 1968; Dransfield, 1979, 1986; Dransfield and Beentje, 1995; Henderson et al., 1995; Henderson, 2002, 2009; Dransfield et al., 2008; Dowe, 2010) and for a few species supplemented with expert knowledge.

We tested 14 predictor variables as potential determinants of species richness in climbing and non-climbing palms. These variables reflected contemporary climate (six variables), paleoclimate (six variables), canopy height (one variable), and biogeographic region (one variable). Present and past climates (Kissling et al., 2012a; Blach-Overgaard et al., 2013; Rakotoarinivo et al., 2013) as well as biogeographic history (Baker and Couvreur, 2013a,b) are important drivers of broad-scale species distributions and diversity patterns in palms. Extracted at a coarse resolution (averaged within TDWG level 3 units), these predictor variables allow assessing how broad-scale trends in environmental conditions are related to geographic differences in species numbers of palms worldwide. We acknowledge that fine-scale heterogeneity (e.g. in climates, canopy heights etc.) within TDWG level units might not be well captured by such coarse-grained data, but analyses with new species distribution datasets at high resolution could incorporate some of this heterogeneity in the future.

Beyond examining the determinants of species richness of climbers and non-climbers, we performed diversification analyses to test whether the evolution of climbers has an impact on diversification rates as a whole and for specific clades.

To identify how many times and when the climbing habit arose in palms above the genus level, we conducted an ancestral character reconstruction using a stochastic character (posterior) mapping approach, as implemented in the program SIMMAP (Huelsenbeck et al., 2003; Bollback, 2006). We used the generic level chronogram of Couvreur et al. (2011a; 183 tips) and the genus-level coded dataset (derived from the monomorphic dataset used for the diversification analyses above). This approach does not take into account the evolution of climbers within mainly non-climbing genera such as Dypsis and Chamaedorea (three species only). We also coded the genera Calamus and Daemonorops (subfamily Calamoideae; subtribe Calaminae) as ancestrally climbing, even though a few species within each genus are non-climbers (Table 1). Ancestral states were estimated using the make.simmap function in the R package phytools (Revell, 2012). Because specifying incorrect prior values can influence posterior mapping results (Couvreur et al., 2010) we used the empirical approach function of make.simmap where the priors (α and β) of the transformation rate from one character to another (γ) were specified as follows: β = 5 and α = β × ML(Q), with Q being the transition matrix between both states. This was achieved using the use.empirical = TRUE option in make.simmap. We undertook 10,000 generations sampling every 100 steps. We used the function ‘DensityMap’ in Phytools to depict the changes in posterior probabilities (PP) along branches of the phylogeny.

Out of 2,446 palm species in our dataset, a total of 535 species (22%) were classified as climbers. The majority of climbing palm species are found within the genera Calamus (348 species) and Daemonorops (92 species). In most cases, genera with climbing habits have a high (≥90%) proportion of climbing species (Table 1). Geographically, climbing palm species occur in all four major tropical regions (Afrotropics, Australasia, Indomalaya, Neotropics; Table 1; Figure 2) and two species in Calamus even reach Oceania. Climber species richness peaked in Indomalaya, closely followed by Australasia, with the Neotropics showing the lowest climber richness (Figure 2A). This contrasted to the species richness of non-climbing palms, which, when compared to other tropical regions, was highest in the Neotropics and lowest in the Afrotropics (Figure 2B).

Among contemporary climatic variables, TEMP (positive effect) and PREC SEAS (negative effect) were the most important variables to explain climber species richness in the minimum adequate models (Figures 3B,C; Table 2). Interestingly, PREC SEAS showed contrasting effects for climbers vs. non-climbers (negative vs. positive sign). Two additional contemporary climatic variables (PREC, TEMP SEAS) were of further importance to explain species richness of non-climbing palms, but not climbers (Table 2). Paleoclimatic changes (anomalies) since the Pliocene (PLIO_PREC) and Miocene (MIO_PREC) showed negative effects on species richness of both climbers (Figures 3D,E) and non-climbers (Table 2). This indicated that areas that were relatively wetter during the late Pliocene or late Miocene tend to have more palm species today than areas that were relatively drier in the past. In contrast to Pliocene and Miocene climate variables, precipitation and temperature anomalies since the LGM were not important to explain species richness of climbing palms, but the LGM_TEMP effect was important for non-climbing palms (Table 2).

The minimum adequate models included CANOPY as an important predictor variable for the species richness of both non-climbing and climbing palms, but the effect was stronger for climbers than for non-climbing palms (Table 2). The CANOPY effect on climbing palm species richness was positive (Figure 3A), supporting the hypothesis that richness increases with canopy height of forests. In addition to CANOPY and contemporary climate as well as paleo-climatic variables, biogeographic region (REGION) was also selected in the minimum adequate models (Table 2). Climbers showed a significantly higher species richness in Indomalaya relative to other tropical regions (Figure 3F) whereas non-climbers showed a significantly lower species richness in the Afrotropics relative to Indomalaya and the Neotropics (Table 2). These results were similar to the trends in raw species richness among realms (Figure 2), but they additionally accounted for major differences in contemporary climate, canopy height, and paleoclimate. This suggests that deep-time historical effects beyond those driven by Quaternary and late Neogene climate contribute substantially to the major differences in species richness of climbers and non-climbers among regions.

Comparing the ten ClaSSE diversification models revealed that model 7, the cladogenetic speciation and extinction model, was the best fitting model based on average AIC_c values from the 100 trees (Table 3). This was not only the case for the unconstrained analysis but also for the constrained analysis. This model suggested that the anagenetic state change rates between climbing and non-climbing palms are equal (q_1-0= q_0-1, Figure 1), while all other parameters are significantly different. Climbers had on average significantly higher speciation and extinction rates when compared to non-climbers (Table 3).

The BAMM analyses reached a stationary state well before 100,000 generations in all independent runs. The ESS values for the number of shifts and the log-likelihood were always above 200, indicating appropriate sampling of parameters from the posterior. Under the Poisson prior of 1.0, the zero rate shift model was rarely sampled from the posterior, strongly supporting the hypothesis of diversification rate heterogeneity across palms. Based on a Poisson prior of 0.1, the most probable number of rates shifts was 9 (PP = 0.135), closely followed by 8 (PP = 0.132), and 10 (PP = 0.131). This means that diversification rates have changed 9, 8, or 10 times across the history of palms.

The phylorate plot shows an increase in mean diversification rates at the crown node of subtribe Calaminae (depicted by an arrow in Figure 4A). Calaminae is the largest clade of climbing palms containing around 20% of all palms species (Dransfield et al., 2008). We can also note an increase in diversification rates for subtribe Bactridinae, which contains one genus of climbing palms (Desmoncus). Other climber dominated clades do not show such increases.

The six most probable CSS’s (with a cumulative PP of 0.614, Figure 5) show that significant rate increases (red circles) are mainly located on the branches leading to part of the tribe Areceae, most of tribe Trachycarpeae and most of subtribe Bactridinae. No significant rate shifts within theses sets were identified in relation to climbing dominated clades mainly found within Calamoideae (Figure 5). Figure 6 represents the phylogenetic tree of palms where branch lengths were scaled to their BF (a) or marginal probabilities (b) of containing a rate shift. The shift in probability for subtribe Calaminae is visible in both cases but not significant when compared to other branches across palms.

Posterior mapping identified 4.9 average transformations from the non-climbing to the climbing state across the 100 trees, and 0.61 transformations in the opposite direction (reversals). These results do not include the three climbing species within the genera Dypsis and Chamaedorea, but in terms of timing of the origin of climbers in palms this omission has no effect on the results. The ancestral state of the crown nodes of the subtribes Calaminae, Plectocomiinae, and Ancistrophyllinae were strongly supported as climbing [PP(1) = 1.00; 0.99; 0.98, respectively]. Figure 4B shows that the climbing habit first evolved in palms along the branch leading to the crown node of subtribe Ancistrophyllinae during the early Eocene (55–47 mya). A second and third origin of the climbing habit was dated to the late Eocene and early Oligocene (44–33 mya) along the branches leading to the crown nodes of the subtribes Plectocomiinae and Calaminae. The timing of the two last origins could not be estimated, having evolved along the stem node of the South American genus Desmoncus and the Southeast Asian genus Korthalsia (Figure 4B).

Our results show that climbing palm species richness is associated with present-day climate (temperature, precipitation seasonality) and paleoclimatic changes since the late Neogene (Miocene and Pliocene; Table 2; Figure 3). An increase in species richness with higher temperatures was observed for both climbers and non-climbers which is in line with our hypothesis (prediction 1a) and consistent with findings on global palm diversity patterns (Kissling et al., 2012a). This relationship reflects the limited physiological and functional adaptations of palms, which reduce their survival in areas with cold climates (Tomlinson, 2006). However, the missing effect of temperature seasonality and the observed negative effect of precipitation seasonality on species richness of climbers are opposite to that of non-climbers (Table 2). The negative correlation with precipitation seasonality also contrasts with woody climbing plants, which increase (rather than decrease) in diversity with precipitation seasonality (Schnitzer, 2005). These trends could be explained by climbing palms being mostly found within Calamoideae, a subfamily that is constrained to warm and humid environments (Baker et al., 2000c; Couvreur et al., 2011a). The effect of precipitation changes (rather than absolute levels of contemporary precipitation) is also supported by the paleoclimatic effects in our regression models, which showed that Miocene and Pliocene precipitation anomalies had negative effects on climbing palm species richness. Hence, areas that were relatively wetter during the late Miocene or late Pliocene today have more climbing palm species than areas that were drier in the past. This suggests that multimillion-year non-equilibrium dynamics in diversity–climate relationships play a role in explaining present-day diversity of palms (Blach-Overgaard et al., 2013).

Our results further support the hypothesis that climbing palms are more diverse in tall-stature forests than in lower canopy ones (Figure 3A; hypothesis and prediction 1b). Forest canopy is not uniform across continents, being highest in Southeast Asia and Africa when compared to South America and Australia (Banin et al., 2012). Indeed, higher canopies and larger trees could arguably provide a three dimensional space that is more effectively exploited by certain kinds of climbing plants, particularly climbing palms. First, higher forests have a more patchy canopy because they have larger gaps in between tree trunks and fewer large trees reaching maximum heights (e.g., dipterocarp forests, see below, Richards, 1996). Second, larger trees can lead to larger tree-fall gaps, which has been shown to maintain a higher diversity of climber species in general (Putz, 1984; Schnitzer and Carson, 2001). However, in contrast to non-monocot lianas (Schnitzer, 2005) the majority of climbing palms might be less tolerant of strong disturbances. This is probably also reflected in the negative relationship of palm species richness with PREC SEAS (Table 2), which is opposite to dicotyledonous lianas (Schnitzer, 2005). Third, climbing palms have developed particularly long and light stiff “searcher-stems” compared with many other woody climbers, and they also produce long and sharp-toothed cirri and flagella. These climbing traits are particularly effective for spanning large gaps between large and tall canopy trees.

As indicated by Isnard and Rowe (2008), there is relatively little developmental and anatomical difference between climbing and non-climbing palms compared with clades of woody species in which climbers can develop highly derived and complex stem structures via anomalous secondary growth. Since the arguably more “simple” palm organization can develop effective “liana”-sized climbers it is intriguing that palms have not evolved the climbing habit more often and more widely, especially in the Neotropics. With just seven species, the Neotropics contain few representatives of the subfamily Calamoideae (Dransfield et al., 2008) and the observed diversity anomaly among regions could thus be related to phylogenetic constraints within palms (Bjorholm et al., 2006). Compared with the Calamoideae, few climbers have evolved within the Arecoideae (the largest palm subfamily in general and also within the Neotropics, Pintaud et al., 2008; Figure 4). This is reflected in genera such as Chamaedorea and Dypsis (Madagascar), which each contain one and two climbing species out of 110 and 140 species, respectively (Dransfield et al., 2008). The failure to diversify as climbers might be related to the fact that they are non-spiny, in contrast to the majority of other climbers in the Calamoideae. The only exception in Arecoideae is the Neotropical spiny genus Desmoncus (Figure 4), which has a currently reported diversity between 12 and 24 species (Isnard et al., 2005; Henderson, 2011). Other tropical plant families show a similar pattern of low climber diversity in the Neotropics compared to the Paleotropics. For example, Annonaceae, an important TRF plant family (Couvreur et al., 2011b), have a single climbing species in the Neotropics compared to ∼500 species in the Paleotropics. This might point toward potential phylogenetic constraints on certain plant families to successfully evolve and diversify as climbers in a given region. Some families might be particularly successful in their evolution toward a climbing form in one region, for example, the Neotropics (e.g., Bignoniaceae; Lohmann et al., 2013) whereas others might not (e.g., palms and Annonaceae).

Evidence suggests that the evolution of climbers promoted diversification within angiosperms (Gianoli, 2004). The diversification history of palms has not been homogenous across time (Baker and Couvreur, 2013b), and our results based on an improved diversification analysis approach (Rabosky, 2014) confirms this hypothesis. Indeed, a total of nine rate shifts across palms were detected with the highest posterior probability.

The analyses presented here suggest an important evolutionary role of climbers in explaining present-day palm diversity (hypothesis 2). Our ClaSSE analyses indicated that across palms, species with a climbing habit diversified on average 1.3 times faster when compared to species with a non-climbing habit (λ₁₁₁ > λ₀₀₀; see Table 3, prediction 2a). Elevated diversification rates for particular clades are highly dependent on young estimated crown node ages (Linder, 2008). In our case, young crown node ages of large climbing genera such as Calamus and Daemonorops will strongly pull toward higher speciation rates for climbers compared to non-climbers. An inverse result would occur if crown node ages were inferred to be very old for both these genera. Unfortunately, to date no valid estimations of crown nodes exist for genera within Calaminae. By constraining the crown node ages of all genera in our simulations to be between 20 and 80% of the stem node age (‘constrained analysis’), we avoided bias due to very old or very young crown node ages. Under these conditions, the climbing habit was also found to have significantly higher speciation rates, albeit slightly smaller in magnitude (Table 3).

Despite the overall higher diversification rate associated to climbers, the impact of the climbing trait on the diversification of specific clades is unclear (prediction 2a). Based on the analysis of the BAMM output, an increase in diversification rates (Figure 4A) is detected around the crown node of subtribe Calaminae (subfamily Calamoideae), the most species rich clade of climbing palms with around 500 Southeast Asian species (Dransfield et al., 2008). Surprisingly, however, this increase is not significant when compared to other diversification rate shifts found across the family (Figures 5 and 6). Nevertheless, a previous diversification study using the same phylogenetic tree and data but based on a ML stepwise AIC approach implemented in MEDUSA identified two significant rate increases at the stem nodes of Calamus and Daemonorops, both genera belonging to Calaminae (Baker and Couvreur, 2013b). In addition, Calamus was found to have significantly more species than expected under a constant birth–death rate model as well as have significantly higher diversification rates when compared to other genera under a high extinction rate assumption (Baker and Couvreur, 2013b). Even though these results should be interpreted with caution, taken together they support the idea that the evolution of the climbing habit in Calaminae positively impacted diversification rates in palms resulting in the speciation of a fifth of all palm species.

The increase in diversification rates in Bactridinae does not appear to be related to the climbing habit that evolved in the genus Desmoncus. Indeed, this rate increase, which was also detected with the MEDUSA analysis (rate shift 8 of Baker and Couvreur, 2013b that included Desmoncus, Bactris, and Astrocaryum), concerns most of this subtribe (excluding Acrocomia, Figure 5) and was suggested to be related to the evolution of epidermal spines, a common trait of all Bactridinae members which functions as a protection against herbivory in the Neotropics (Baker and Couvreur, 2013b).

The ancestral state reconstructions indicated that the climbing habit evolved a minimum of five independent times in palms: four times in Calamoideae and once in Arecoideae (Figure 4B). These results are consistent with the inferences for Calamoideae of Baker et al. (2000a). However, the climbing habit arose at least seven independent times because of the three climbing species within Dypsis and Chamaedorea that were not taken into account in the analysis presented here (both genera coded as non-climbing). At least in Chamaedorea, the single climbing species (C. elatior) is nested within the genus validating this assumption (Cuenca and Asmussen-Lange, 2007). In addition, these independent evolutions will be valid if our coding assumptions of Calamus and Daemonorops as ancestrally “climber” are also correct. Indeed, both these genera have a small proportion of non-climbing species (Table 1), and in this study we did not take them into account (both genera coded as climbers). The phylogenetic relations within both genera and for the subtribe Calaminae in general remain insufficiently understood and the exact placement of the non-climbing species is unresolved (Baker et al., 2000b).

Our results reveal an interesting pattern: the climbing habit appears to have had an impact on diversification rates in Calaminae, but not in other clades/genera where it evolved (Ancistrophyllinae, Plectocomiinae, Korthalsia, Desmoncus). For example, the climbing subtribes Ancistrophyllinae and Plectocomiinae (Figure 4A) have relatively old stem node ages associated with few extant species (Dransfield et al., 2008; Sunderland, 2012; Baker and Couvreur, 2013a; Faye et al., 2014). Why did the climbing habit have such an important effect on the diversification in a particular clade and not in other clades? One reason might be that different morphological adaptations are underlying the “same” trait (Donoghue, 2005), i.e., that convergent evolution (homoplasy) is constructed quite differently in different lineages and therefore has different impacts on diversification rates. For instance, in palms the climbing habit in different clades is associated with different combinations of morphological characters (Baker et al., 2000a; Isnard, 2006). Comparative studies of anatomical characters between Calaminae (Calamus and Daemonorops) and Desmoncus and subtribe Plectocomiinae (although based on the study of a few species) suggest that the evolution of a more flexible stem within subtribe Calaminae could explain the outstanding diversification of this group in terms of climbing mechanics (Rowe et al., 2004; Isnard, 2006). In addition, the evolution of a unique climbing organ (the flagellum, a modified inflorescence) has taken place within Calamus (Dransfield et al., 2008) and might have provided an additional advantage over other climbing structures in palms. Moreover, the presence of a “knee” a swelling at the junction of the leaf sheath and petiole, in most species in subtribe Calaminae and, to some extent, the African genus Eremospatha has been suggested as a potentially important trait for enhancing leaf strength at this important stress point (Isnard and Rowe, 2008). Overall, the repeated homoplasious occurrence of the climbing habit in the Calamoideae (‘clustered homoplasy’) could also indicate a more cryptic evolutionary innovation related to genetic or developmental precursors (Marazzi et al., 2012). In subfamily Calamoideae, this could be related to the tendency of organizing epidermal emergences as whorls, which manifest themselves, for example, both in the characteristic scales on the fruits as well as in organized grapnel spines on climbing organs. Finally, some life history traits might also act against increasing diversification rates. Indeed, Plectocomiinae and the genera Korthalsia (Korthalsiinae, Figure 4C) and Laccosperma (Ancistrophyllinae, Figure 4D) are hapaxanthic (individual stems die after a single flowering event), a condition which is generally not associated with high species richness across palms (Dransfield et al., 2008). Thus, several morphological novelties of subtribe Calaminae absent from other climbing genera (more flexible stems, the evolution of a flagellum in Calamus, the better mechanical role of the leaf sheath under stress, and the presence of a knee) possibly played decisive functional roles in explaining the diversification of this group when compared to other climbing palms.

Explaining geographic differences in diversification rates (e.g., numerous climbing species in Southeast Asia vs. few in the Neotropics) might be related to the fact that a particular trait only increases diversification rates under certain environmental conditions (de Queiroz, 2002). For instance, a highly dynamic geographic setting linked to a complex biogeographic history of Southeast Asia (Hall, 2009) might be important in explaining the extraordinary diversification of rattans in this region (Baker and Couvreur, 2012). The estimated increase in diversification rates around the crown node of subtribe Calaminae (34 Ma, Figure 4A) coincides with a period of important geological activity (Oligocene, early Miocene) induced by the collision of Sundaland and Australia (Hall, 2009). The Oligocene–Miocene boundary was also an important climatic transition going from a seasonally dry to an everwet aseasonal climate (Morley, 2007). In line with our results, numerous other studies have underlined the importance of the Miocene in the diversification of the Southeast Asian flora (Morley, 2007; Su and Saunders, 2009; Lohman et al., 2011; Nauheimer et al., 2012; Thomas et al., 2012; Bacon et al., 2013; Buerki et al., 2013; Richardson et al., 2014).

One important and characteristic plant family in Southeast Asia is Dipterocarpaceae (Appanah and Turnbull, 1998), with major peaks in diversity in Malaysia and mainland Southeast Asia. Its species are generally large emergent trees, which often dominate the upper canopy of the region’s forests, contributing to the much taller stature of the forests in this region relative to, e.g., the Neotropics (Richards, 1996). Interestingly, the first record of the family Dipterocarpaceae in Southeast Asia (Borneo) after its dispersal from India (Dutta et al., 2011) as well as the start of its estimated radiation in the region (Morley, 2000) correspond to the timing of increased diversification rates within the Calamoideae (Oligocene to early Miocene). We therefore hypothesize that the diversification of dipterocarps in combination with the evolution of several morpho-anatomical traits in Calaminae species could have triggered the radiation of climbing palms in this region. Ecological opportunity responses of one clade based on the success of another have been suggested in other cases such as ferns (Schneider et al., 2004) or ants (Moreau et al., 2006) during the radiation of angiosperms.