High tropical net diversification drives the New World latitudinal gradient in palm (Arecaceae) species richness

Authors


*Jens-Christian Svenning, Department of Biological Sciences, University of Aarhus, Ny Munkegade 1540, DK-8000 Aarhus C, Denmark. E-mail: svenning@biology.au.dk

Abstract

Aim  Species richness exhibits striking geographical variation, but the processes that drive this variation are unresolved. We investigated the relative importance of two hypothesized evolutionary causes for the variation in palm species richness across the New World: time for diversification and evolutionary (net diversification) rate. Palms have a long history in the region, with the major clades diversifying during the Tertiary (65–2 Ma).

Location  Tropical and subtropical America (34° N–34° S; 33–120° W).

Methods  Using range maps, palm species richness was estimated in a 1° × 1° grid. Mean lineage net diversification was estimated by the mean phylogenetic root distance (MRD), the average number of nodes separating a species from the base of the palm phylogeny for the species in each grid cell. If evolutionary rate limits richness, then richness should increase with MRD. If time limits richness, then old, relict species (with low root distance) should predominantly occur in long-inhabited and therefore species-rich areas. Hence, richness should decrease with MRD. To determine the influence of net diversification across different time frames, MRD was computed for subtribe, genus and species levels within the phylogeny, and supplemented with the purely tip-level measure, mean number of species per genus (MS/G). Correlations and regressions, in combination with eigenvector-based spatial filtering, were used to assess the relationship between species richness, the net diversification measures, and potential environmental and geographical drivers.

Results  Species richness increased with all net diversification measures. The regression models showed that richness and the net diversification measures increased with decreasing (absolute) latitude and, less strongly, with increasing energy/temperature and water availability. These patterns therefore reflect net diversification at both deep and shallow levels in the phylogeny. Richness also increased with range in elevation, but this was only reflected in the MS/G pattern and therefore reflects recent diversification.

Main conclusions  The geographical patterns in palm species richness appear to be predominantly the result of elevated net diversification rates towards the equator and in warm, wet climates, sustained throughout most of the Tertiary. Late-Tertiary orogeny has caused localized increases in net diversification rates that have also made a mark on the richness pattern.

Introduction

Species richness exhibits striking geographical variation. The driving mechanisms have been debated since the 19th century and remain a central and controversial issue in ecology and biogeography. The ubiquity and strength of correlations between species richness and climate have led to the conclusion that current climate is a strong driver of species richness at coarse scales of analysis (Hawkins et al., 2003; Field et al., 2005). Nonetheless, relationships between species richness and contemporary environmental variables must also reflect history, as the species richness of a region must ultimately be the product of the evolutionary and biogeographical processes of in situ speciation, migration of species into the region, and extinction (Wiens & Donoghue, 2004; Stevens, 2006).

Two basic evolutionary causes of latitudinal and other broad geographical diversity patterns have been proposed (Fischer, 1960; Cardillo et al., 2005; cf. Mittelbach et al., 2007). First, according to the evolutionary rate hypothesis, species richness reflects the net rate of species formation, the rate of species accumulation taking into account both speciation and extinction (as discussed by Fischer, 1960). Notably, this hypothesis posits that more species-rich areas have experienced higher rates of net diversification owing to higher speciation rates and/or lower extinction rates. Therefore it encompasses both the cradle hypothesis (high speciation rate) and the museum hypothesis (low extinction rate) for high tropical diversity (Chown & Gaston, 2000). The evolutionary rate hypothesis could be considered ahistorical if rates have been steady during the period in which the biota in question have evolved, reflecting factors that have been largely constant over time and continue to affect the biota. However, this hypothesis clearly has a historical component if rates have varied over time, for example if the rate of tropical diversification was higher in the early Tertiary when tropical climates were more extensive than they are today (Wiens & Donoghue, 2004; Fine & Ree, 2006). Second, according to the time hypothesis, species richness reflects the time available for diversification by in situ speciation and immigration: more species-rich areas have been occupied for a longer time or have more ancestral environmental conditions. This is a purely historical explanation.

Many climate-based hypotheses regarding broad patterns of species richness could arguably be considered forms of the evolutionary rate hypothesis (cf. Mittelbach et al., 2007). For example, the energy–diversity hypothesis proposes that species richness increases with net primary productivity due to its effect on the number of individuals per unit area (Currie et al., 2004). Extinction rates are expected to decrease and speciation to increase with the number of individuals (Hubbell, 2001). The speciation rates hypothesis argues that species richness–climate relationships reflect increased speciation rates in warm, or warm and wet, climates due to shorter generation times, faster mutation rates, faster physiological processes, or (although not an ultimate cause) stronger biotic interactions (Rohde 1992; Allen et al., 2002; Currie et al., 2004). In a similar vein, the related water–energy dynamics hypothesis argues that species richness is greatest where liquid water and optimal energy conditions provide the greatest capacity for biotic dynamics (O’Brien, 1998, 2006). Several alternative hypotheses also posit that patterns of species richness result from differences in rates of net diversification, but do so without invoking contemporary climate. For example, the tropics may be richer in species due to their large area and the resulting large species ranges, which leads to increased rates of allopatric speciation and decreased rates of extinction (Rosenzweig, 1992; Chown & Gaston, 2000). Or the tropics may be richer in species because they have experienced less severe Milankovitch climate oscillations, permitting relatively high rates of speciation and low rates of extinction (Dynesius & Jansson, 2000). It has also been proposed that topographic heterogeneity promotes species richness by increasing rates of allopatric speciation, particularly at low latitudes (Fjeldså, 1994; Turner, 2004). The evolutionary rate hypothesis predicts that species of species-rich regions tend to belong to lineages that have experienced high net diversification rates: species richness should increase with mean lineage net diversification.

Time may limit species richness when species capable of occupying an area exist, but have had insufficient time to disperse into an area (Rohde, 1992; for a temperate example see Svenning & Skov, 2005, 2007). However, dispersal limitation can hardly explain the high tropical species richness at the biome level. Therefore, the most important version of the time hypothesis proposes that tropical regions or climates harbour more species because their greater antiquity has provided more time for speciation (Fischer, 1960; Pianka, 1966; Rohde, 1992). For a given clade, the effect of time need not involve climate, but may be purely geographical: there is greater time for diversification in the region longest occupied (Stephens & Wiens, 2003). The time hypothesis is a key component of the tropical conservatism hypothesis (Wiens & Donoghue 2004), which proposes that the latitudinal gradient in diversity is the product of three basic factors: (i) many groups of organisms with high tropical species richness originated in the tropics, leaving less time for diversification in extratropical regions; (ii) tropical climates encompassed a larger geographical area until 30–40 Ma (Willis & McElwain, 2002); and (iii) the integrated time–area effect is maintained or strengthened by evolutionary constraints on ecological adaptation (climatic niche conservatism). An important corollary of the time hypothesis is that basal, relict clades that have experienced low net diversification (high extinction and/or low speciation) rates would occur primarily in long-inhabited and therefore species-rich regions (Wiens & Donoghue, 2004; Stevens, 2006; Mittelbach et al., 2007). Hence, although the time hypothesis does not invoke differences in speciation or extinction rates, it predicts that the mean net diversification rate of the lineages present will appear to be low where species richness is high, in contrast to the evolutionary rate hypothesis. We point out that this is true only when net diversification is estimated for the entire time span since the origin of the common root of all lineages considered, because the time hypothesis does not predict differences in current net diversification rates, estimated for a limited, more recent time span: the time hypothesis predicts that species richness should have either a negative or no relationship with mean lineage net diversification.

Several recent studies provide support for the time hypothesis (Stephens & Wiens, 2003; Wiens et al., 2007; cf. McPeek & Brown, 2007), especially as part of the tropical conservatism hypothesis (Hawkins et al., 2005, 2006; Stevens, 2006; Wiens et al., 2006). However, other studies have provided support for the evolutionary rate hypothesis, alone or in combination with the time hypothesis (Currie et al., 2004; Cardillo et al., 2005; Jablonski et al., 2006; Fine & Ree, 2006; Ricklefs et al., 2006; Ricklefs, 2006b; for a review see Mittelbach et al., 2007). The reasons for these different results are unknown and there is not yet an understanding of the overall relative importance of the two mechanisms. Relatively few previous studies of these hypotheses have focused on plants, notable exceptions being the studies by Davies et al. (2004) and Ricklefs et al. (2006).

The palm family is a dominant constituent of New World tropical rain forests (Kahn & de Granville, 1992) and is also common in a range of other habitats (Henderson et al., 1995). Palms have a largely tropical distribution (Uhl & Dransfield, 1987), reflecting their very limited frost tolerance (Jones, 1995). The latter is probably highly evolutionarily conserved as it reflects key elements of palm architecture, notably a crown composed of large evergreen leaves (Tomlinson, 1990) with very limited frost resistance (Woodward, 1988), and unique stem metabolic properties, with all stem cell types being physiologically active, prohibiting dormancy (Tomlinson, 2006). As a group, palms exhibit a strong latitudinal diversity gradient in the New World (Bjorholm et al., 2005). Geographical variation in palm species richness is strongly related to climatic water availability, but even after accounting for a wide range of contemporary environmental factors, there is a strong latitudinal relationship (Bjorholm et al., 2005, 2006). While palms have been abundant and widespread in the New World since the late Cretaceous (Uhl & Dransfield, 1987; Morley, 2000), one palm subfamily, Coryphoideae, exhibits a strong northern diversity bias in this region, probably reflecting its late Cretaceous to mid-Tertiary abundance across the Northern Hemisphere (Bjorholm et al., 2006). Although the palm family originated in the Cretaceous, dated phylogenies indicate that most major clades in the New World diversified more recently in the Tertiary, including highly species-rich groups such as the Bactridinae with 113 species (diverging from its sister taxon Elaeidinae 42–46 Ma; Gunn, 2004), the genus Geonoma with 51 species (stem age c. 23 Ma; Savolainen et al. 2006), and the Phytelephanteae with nine species (crown group age 25–16 Ma; Trénel et al., 2007).

In this study we attempt to assess whether evolutionary rate or time for diversification has been the primary evolutionary driver of the present-day patterns of palm species richness across the New World. We tested whether there is a negative (or no) relationship between palm species richness and mean lineage net diversification, as predicted by the time hypothesis, or a positive relationship, as predicted by the evolutionary rate hypothesis (cf. Kerr & Currie, 1999; Hawkins et al., 2005, 2006). Acknowledging that environmental and geographical effects on diversification may have varied across time periods, we analysed several alternative measures of mean lineage net diversification, emphasizing different levels in the palm phylogeny. To obtain a deeper understanding of the mechanisms driving species richness patterns, we also investigated if and how mean lineage net diversification covaries with climatic and geographical factors. The evolutionary rate hypothesis predicts that mean lineage net diversification is higher in more productive, warm and wet climates, close to the equator, and in more topographically complex areas. In contrast, the time hypothesis predicts that mean lineage net diversification will appear higher in less productive, cold and dry climates, away from the equator, reflecting an absence of old, relict lineages, and perhaps especially towards the south, reflecting a northern coryphoid immigration route (Bjorholm et al., 2006).

In the following section we provide a condensed version of our methods. A detailed account is provided in Appendix S1 in Supplementary Material.

Methods

Study region and species distribution data

We scanned distribution maps for all 550 American palm species from Henderson et al. (1995) at a 1° × 1° longitude–latitude grid-square resolution. The study region consisted of 1512 1° × 1° cells with at least one palm species (Fig. 1). Suprageneric classification followed that of Dransfield et al. (2005), while species names generally followed the system of Henderson et al. (1995); see Appendix S1 for further details.

Figure 1.

 Geographical patterns of (a) species richness; (b–e) mean lineage net diversification of New World palms; (f) mean annual precipitation (MAP) at 1° × 1° grid-square resolution. Mean lineage net diversification is represented by the mean root distance resolved to (b) subtribe (MRDsubtribe); (c) genus (MRDgenus); (d) species(MRDspecies); (e) mean number of species per genus (MS/G).

Measures of net diversification

The root distance (RD) for a taxon is the number of nodes separating it from the base of a given phylogenetic tree (Kerr & Currie, 1999). Taxa with low RD represent lineages with low rates of net evolutionary diversification (speciation minus extinction; cf. Ricklefs, 2006a), while taxa with high RD belong to lineages with high rates of net evolutionary diversification. The results presented use RD values computed for a phylogenetic tree that is based primarily on the new genus-level phylogeny for the palm family published by Asmussen et al. (2006). This phylogeny is based on four plastid DNA regions sequenced for 161 of 189 genera in the palm family. The major clades of this phylogenetic tree are well supported, and the tree is well resolved and includes representatives of all currently recognized subfamilies, tribes and subtribes. However, it does not include all New World palm genera, and it has several polytomies. The tree was further resolved following a number of detailed phylogenetic studies with a more limited taxonomic scope. The resulting combined tree and the details of these modifications are provided in Appendix S1, which also establishes the limited sensitivity of the RD values to the adjustments made to the original tree.

To determine the influence of net diversification at different levels in the phylogeny, reflecting different time frames, RD values were computed for each subtribe (RDsubtribe) and each genus (RDgenus). As polytomies bias RD towards lower values if counted as single nodes, unresolved nodes were counted as the average RD for the taxa for all possible fully resolved topologies. As the palm phylogeny was resolved only to genus level, each genus can be considered a polytomy of its species. Hence, we also computed the RD for each species (RDspecies) by treating each genus as a polytomous node with one branch per species. Finally, to examine evolutionary diversification within genera, we computed the number of species per genus (S/G). These four different measures of net diversification were transformed into grid-based measures of mean lineage net diversification by computing the arithmetic mean RD (MRDsubtribe, MRDgenus, MRDspecies) or mean number of species per genus (MS/G) for the species present in each 1° × 1° cell (Kerr & Currie, 1999; Hawkins et al., 2005, 2006). We did not normalize the MRD values using the z-score approach of Hawkins et al. (2005, 2006), as changing sample size (here, species richness) does not change the estimate of the mean in random samples (e.g. MRD). Therefore, from a random sampling perspective, the expected MRD is independent of the number of species and the observed MRD is correct and unbiased, while the z-score approach may well generate uninterpretable patterns (B. Hawkins, pers. comm.).

Statistical analyses

We used correlation analyses to test for a relationship between species richness and each measure of mean lineage net diversification. The degrees of freedom were adjusted to the spatial autocorrelation present (Rangel et al., 2006) using the method of Dutilleul (1993).

We investigated 19 models representing potential determinants of mean lineage net diversification (MRDsubtribe, MRDgenus, MRDspecies, MS/G) and, for comparison, species richness of palms across the New World (Table 1). This set of models was selected to encompass a broad range of models relevant to assessment of the evolutionary rate and time hypotheses, but it is not claimed to be exhaustive. The models were based mainly on the climatic plant diversity models developed by O’Brien (1998); O’Brien et al. (2000), and Francis & Currie (2003), and the well-documented association between palms and wet, warm regions (Jones, 1995; Bjorholm et al., 2005, 2006). We also represented additional non-climatic factors (topography, geography and history) in these models. Predictor variable abbreviations are defined in Table 1. The environmental data sources and derivation of the models are explained in detail in Appendix S1.

Table 1.   Models representing potential determinants of mean lineage net diversification and species richness of palms across the Americas.
NumberModel predictors
  1. |LAT|, absolute latitude (°); LAT, latitude (°N); MAP, annual precipitation (mm year−1); MAT, mean annual temperature (°C); MMP, minimum monthly precipitation (mm month−1); MTCO, mean temperature of the coldest month (°C); PET, potential evapotranspiration (mm year−1); TOPO, elevation range (m), log(+ 1)-transformed; TOPO*LAT, TOPO multiplied by |LAT|max − |LAT|, where |LAT|max is the maximum value attained by |LAT| in the data set, included to reflect the greater importance of TOPO at low latitudes; cf. Rahbek & Graves (2001); WD, water deficit (mm year−1).

1|LAT|
2MAP + PET + PET2
3WD + PET + PET2
4WD + MAT + WD × MAT
5MAP + PET + PET2 + |LAT|
6WD + PET + PET2 + |LAT|
7WD + MAT + WD × MAT+ |LAT|
8MAP + PET + PET2 + |LAT| + TOPO
9WD + PET + PET2 + |LAT| + TOPO
10WD + MAT + WD × MAT + |LAT| + TOPO
11MAP + PET + PET2 + |LAT| + TOPO*LAT
12WD + PET + PET2+ |LAT| + TOPO*LAT
13WD + MAT + WD × MAT + |LAT| + TOPO*LAT
14MAP + MAT
15MTCO + MMP
16MAP + MAT + |LAT|
17MTCO + MMP + |LAT|
18MAP + MAT + |LAT| + LAT
19MTCO + MMP + |LAT| + LAT

The models were first fitted as ordinary multiple linear regression models, but as strong positive spatial autocorrelation remained in the residuals, we also refitted the models using spatial filters (Diniz-Filho & Bini, 2005). Spatial filters were generated as the eigenvectors of a principal coordinates analysis of a pairwise matrix of geographical distances between all grid cells, following the approach of Diniz-Filho & Bini (2005). The eigenvectors represent the spatial relationships among the grid cells at different scales. The first eigenvectors represent broad-scale variation, while eigenvectors associated with small eigenvalues represent small-scale variation. Adding spatial filters to the regression models allowed unbiased estimation of regression coefficients and their significance levels (Diniz-Filho et al., 2003; Diniz-Filho & Bini, 2005). However, we note that spatial autocorrelation may not bias ordinary least-squares estimation of regression coefficients (Hawkins et al., 2007). Instead, controlling for spatial autocorrelation shifts the spatial scale of the analyses and emphasizes smaller-scale environmental relationships (Diniz-Filho et al., 2003). We therefore report both the unfiltered and filtered regression results. Further details on the regression modelling and the spatial filtering are provided in Appendix S1.

Results

Correlations between species richness and mean lineage net diversification

Maps of mean lineage net diversification are shown in Fig. 1. Overall species richness increased with all measures of mean lineage net diversification (Table 2): areas with high palm species richness were dominated by species from lineages with particularly high net diversification rates (Fig. 2). Species richness increased strongly with MS/G, which purely represents diversification at the tips of the phylogeny. There was also a clear positive relationship between species richness and subtribe-level MRD, which purely represents diversification at relatively deep levels of the palm phylogeny. Hence, high species richness reflected high net diversification rates sustained since origination of the palm subfamilies. Despite the overall richness–diversification relationship, there are some obvious discrepancies (see Fig. 1 and the residual map in Fig. 2): notably, the large region in south-eastern Brazil (lower Amazon Basin and the Atlantic Forest) is characterized by high mean lineage net diversification, but low species richness. A similar pattern is found for northern Central America. In contrast, species richness of western Amazonia and the Chocó–southern Central American region is higher than expected from the diversification patterns. Figure 2 shows only the richness residual map for the richness–MRDspecies regression, but the residuals maps for the two other MRD measures were similar. The residuals map for the richness–MS/G regression is also similar, but indicates that richness in south-eastern Brazil is now higher than expected, while richness in northern Central America is even more consistently lower than expected (not shown).

Table 2.   Correlations between species richness and mean lineage net diversification.
Measure of mean lineage net diversification Pearson’s r Spearman’s rd.f.
  1. Species richness (ln-transformed) and MS/G were approximately normally distributed with skewness and kurtosis < |1.0|, while the MRD measures exhibited moderate negative skew (skewness −1.25 to −1.39) and MRDgenus had kurtosis 1.43.

  2. P-values based on degrees of freedom (d.f.) corrected for spatial autocorrelation using Dutilleul’s (1993) method. n = 1512 1° × 1° cells.

  3. *< 0.05; **< 0.01; ***< 0.001.

MRDsubtribe0.691**0.398n.s.13.3
MRDgenus0.537***0.234n.s.34.4
MRDspecies0.778**0.618*12.0
MS/G0.751**0.773**8.9
Figure 2.

 Species richness as a function of mean lineage net diversification (MRDspecies). Line indicates linear regression of species richness on MRD, ln(species richness) = 0.242 MRDspecies−2.26, R2 = 0.605, F = 2312, < 0.0001, n = 1512. A map of the residuals is also shown.

Environmental and geographical correlates of mean lineage net diversification and species richness

Among the regression models lacking spatial filters (Table 3), species richness increased strongly towards the equator and more moderately with decreasing water deficit and increasing potential evapotranspiration (PET) and, more weakly, with elevational range. Similar relationships were found for MS/G. The three MRD measures decreased strongly not only away from the equator in general, but especially northwards. They generally also increased with increasing precipitation and temperature.

Table 3.   The most strongly supported regression models without spatial filters for palm species richness and mean lineage net diversification across the Americas.
MeasureModelFR2adj (models†)AICc (w, %)I|max|‡ (lag, km)
  1. Standardized regression coefficients, model F-ratio, model fit (adjusted R2), AICc and Akaike weight (w) are shown. The maximum Moran’s I (I|max|) for the residuals across 21 distance lags (each with c. 54,400 pairs) and the distance lag it is associated with are also shown. Model support was evaluated based on w. However, in every case, the model preferred based on its Akaike weight also had the highest R2adj in the set of 19 candidate models. n = 1512 1° × 1° cells.

  2. *< 0.05; **< 0.01; ***< 0.001; ****< 0.0001.

  3. †Models with an R2adj that is reduced by at most 0.05 compared to the shown optimal model (model numbers as in Table 1).

  4. ‡The residual errors had Moran’s I = −0.225 (lag 4662–5404 km) and 0.332 (lag 6197–7098 km) for ln(Richness) and −0.396 (lag 6197–7098 km) for MS/G; apart from the ≤ 595-km lag, Moran’s I for the residual errors was otherwise < |0.20|.

Ln(richness)−0.336 WD 0.325 PET−0.086 PET2−0.537 |LAT|0.136 TOPO14610.828−21010.377
 ************************(5–12)(100)(≤ 595)
MRDsubtribe 0.044 MMP 0.285 MTCO−0.421 |LAT|−0.435 LAT 6460.6319350.540
 ************** ****(18,19)(100)(≤ 595)
MRDgenus 0.038 MMP 0.261 MTCO−0.235 |LAT|−0.537 LAT 3790.50020330.504
 ************* ****(18,19)(100)(≤ 595)
MRDspecies 0.154 MAP−0.016 MAT−0.584 |LAT|−0.297 LAT 4190.52629630.515
 ****n.s.******** ****(5,6,8–12,18,19)(100)(≤ 595)
MS/G−0.156 WD 0.323 PET−0.043 PET2−0.402 |LAT|0.185 TOPO3200.51361530.387
 *********************(5,6,8–13,16,18,19)(100)(≤ 595)

Strong positive spatial autocorrelation remained in the model residuals at < 600-km distance (Table 3). Considering the spatial patterns in the residuals of the unfiltered models, it is clear that western Amazonia, south-eastern Brazil and Central America are more species-rich than can be explained by simple climate- and latitude-related factors (Fig. 3). Especially, south-eastern Brazil and Central America are also characterized by high MRD and/or MS/G residuals (Fig. 3).

Figure 3.

 Geographical patterns of residuals from the most strongly supported regression models without spatial filters (Table 3) for (a) species richness; (b–d) mean lineage net diversification of New World palms. Mean lineage net diversification is represented by the mean root distance resolved to (b) subtribe (MRDsubtribe) or (c) species(MRDspecies) and (d) mean number of species per genus (MS/G). The pattern for MRDgenus is similar to that for MRDsubtribe.

Residual spatial autocorrelation was largely removed when spatial filters were included, while the environmental and geographical relationships exhibited only slight changes (Table 4). Among the regression models with spatial filters for mean lineage net diversification, the MMP + MTCO + |LAT| + LAT model was strongly supported for all MRD measures, while different models were preferred for MS/G and species richness (Table 4). However, it is important to note that there was little difference in fit among most models (Tables 3 & 4).

Table 4.   The most strongly supported spatial filter regression models for species richness and mean lineage net diversification.
Measure (no. filters)ModelFR2adj (models†)AICc (w, %)I|max|‡ (lag, km)
  1. Standardized regression coefficients, model F-ratio, model fit (R2adj), AICc and Akaike weight (w) are shown. The maximum Moran’s I (I|max|) for the model residuals observed across 21 distance lags (each with c. 54,400 pairs) and the associated distance lag are also shown. Model support was evaluated based on w. However, in every case the model preferred based on its Akaike weight also had the highest R2adj in the set of 19 candidate models. n = 1512 1° × 1° cells.

  2. *< 0.05; **< 0.01; ***< 0.001; ****< 0.0001.

  3. †Models with an R2adj that is reduced by at most 0.05 compared with the shown optimal model (model numbers as in Table 1).

  4. ‡Apart from the ≤ 595-km lag, Moran’s I for the residual errors was otherwise < |0.10|.

Ln(richness)−0.334 WD0.289 PET−0.047 PET2−0.416 |LAT|0.072 TOPO362.20.907−29950.092
(36)************************(2,3,5–14,16–19)(100)(≤ 595)
MRDsubtribe0.041 MMP0.256 MTCO−0.232|LAT|−0.459 LAT 217.70.848−3740.099
(35)************** ****(18,19)(100)(≤ 595)
MRDgenus0.038 MMP0.140 MTCO−0.204 |LAT|−0.326 LAT 157.30.7848000.110
(31)n.s.************ ****(1–19)(100)(≤ 595)
MRDspecies0.063 MMP0.075 MTCO−0.458 |LAT|−0.242 LAT 130.20.78218260.108
(38)************ ****(1–19)(100)(≤ 595)
MS/G0.218 MAP−0.167 MAT−0.485 |LAT|0.224 LAT 96.50.71753730.076
(36)**************** ****(1–3,5–19)(100)(≤ 595)

The relationships observed were generally in agreement with the predictions derived from the evolutionary rate hypothesis and contrary to those of the time hypothesis. Notably, both species richness and mean lineage net diversification increased with decreasing absolute latitude, and generally also with increasing water and energy availability (Tables 3 & 4). These patterns were found for both MRDsubtribe and MS/Gs, and therefore suggest that the species richness pattern to a large extent reflects high net diversification in warm, wet, equatorial regions, sustained across both deep and shallow levels in the palm phylogeny. However, the increased species richness in mountain areas (positive TOPO relationship in the model without filters, negative MAT relationship in filter model; see Table 1 for definitions) was reflected only in the MS/G pattern (Tables 3 & 4). All MRD measures decrease towards the north, while the reverse relationship is observed for MS/G (Tables 3 & 4). Hence, in the northern part of the study area, the palm species generally come from lineages that have experienced low net diversification except within genera. Reflecting these opposite deep and shallow phylogenetic-level latitude effects, species richness exhibited no overall trend with (signed) latitude.

Discussion

Support for the evolutionary rate hypothesis

Our results indicate that net diversification rate, rather than time for diversification, is the most important driver of the geographical patterns of palm species richness in the New World. First, species richness is positively correlated with all measures of mean lineage net diversification (Table 2), as predicted by the evolutionary rate hypothesis, but not by the time hypothesis. We found that species-rich areas are dominated by palms from lineages with high net diversification rates, both deep in the phylogeny (notably MRDsubtribe) and at the tip of the phylogeny (MS/G). Assessment of absolute ages corresponding to these phylogenetic levels is difficult, particularly given the limited taxonomic coverage of dating studies done so far. For those subtribes (or tribes if not further subdivided) where we do have time estimates, stem ages typically fall in the interval of 60–40 Ma (Gunn, 2004; Savolainen et al., 2006; Trénel et al., 2007), in the early Tertiary. Based on the same studies, diversification of modern species within genera appears to have occurred mostly within the last 5–20 Myr, but again a large uncertainty exists, and for many genera no crown age estimates are available. High palm species richness thus appears to be associated with environments that have favoured high net diversification throughout most of the Tertiary. This historical interpretation of our results is consistent with the fact that palms have been a dominant element of the equatorial South American vegetation since the late Cretaceous and early Tertiary (Morley, 2000). Second, the estimated geographical and environmental relationships were generally consistent with the evolutionary rate hypothesis, not the time hypothesis. According to the time hypothesis, species from old, relict lineages should be concentrated in warm and wet areas and at low latitudes, for which reason these areas should be characterized by low MRD. However, the opposite pattern was observed: the dominance of species from lineages with high net diversification rates increased towards the equator and with increasing energy and water availability. Our findings for New World palms are consistent with: (i) the positive relationship between species richness and evolutionary rate in flowering plant families (Barraclough & Savolainen, 2001); (ii) the positive relationship between regional species richness of mangrove trees and shrubs and rate of lineage origin (Ricklefs et al., 2006); and (iii) increasing evidence of higher rates of diversification in tropical clades of many groups of organisms (Mittelbach et al., 2007). Nevertheless, there are some noteworthy discrepancies from the overall richness–diversification relationship (Figs 1 & 2), which may reflect region-specific conditions, notably in south-eastern Brazil (a possible explanation is given later). Although we find that the geographical patterns in palm species richness appear to be driven mainly by differences in net diversification, we cannot assess the relative importance of extinction (the museum hypothesis, Chown & Gaston, 2000) and speciation (the cradle hypothesis, Chown & Gaston, 2000). Since RD measures the net diversification (speciation minus extinction) of a lineage, MRD cannot easily be used to distinguish speciation from extinction as the driver of the geographical richness patterns. As discussed below, both factors are probably involved.

Geographical and environmental drivers of palm diversification

The three main predictors of palm species richness were energy, water and absolute latitude (Tables 3 & 4). The consistent positive relationship between energy/temperature and species richness and mean lineage net diversification indicates that increased net diversification rates under warm conditions have made an important contribution to the geographical variation in palm species richness in the New World. The overall strongly positive relationship between species richness and energy is mirrored in the positive relationships between mean temperature of the coldest month (MTCO) and mean lineage net diversification, especially MRDsubtribe. As there was a negative MS/G–MAT relationship in the filter model, at small scales, high temperatures have not favoured recent diversification. Hence, the increase in palm species richness with increasing energy may reflect mainly relatively ancient (early to mid-Tertiary) high rates of net diversification under warm conditions. Increased net diversification could reflect lower rates of extinction and higher rates of speciation caused by an increase in the number of individuals as well as increased speciation rates generated by water–energy dynamics and biotic interactions (O’Brien, 1998, 2006; Currie et al., 2004; Mittelbach et al., 2007; cf. Hubbell, 2001). However, increased net diversification could also reflect the greater ecological success of palms in warm climates by virtue of their large leaves and stem physiology (Tomlinson, 1990; see Introduction). Or, the relationship between energy/temperature, species richness and mean lineage net diversification could reflect an area effect, because warm environments were more widespread in the early Tertiary period (Burnham & Graham, 1999; Morley, 2000). A time-integrated area effect of that kind (Fine & Ree, 2006) would be maintained as a species richness–environment relationship due to phylogenetic niche conservatism (Wiens & Graham, 2005), and is the only one of these explanations that is potentially consistent with the negative small-scale MS/G–MAT relationship.

The consistent positive relationship between availability of water and species richness and mean lineage net diversification indicates that increased net diversification rates under wet conditions also made an important contribution to the geographical variation in palm species richness in the New World. If energy is not limiting, availability of water is likely to correlate with productivity and water–energy dynamics (O’Brien, 2006). As for energy/temperature, the relationships between species richness and mean lineage net diversification and MMP/MAP/WD might reflect productivity and water–energy dynamics, constraints on the ecological success of palms in dry climates, or a time-integrated area effect (as wet environments were more widespread in the early Tertiary; Burnham & Graham, 1999; Willis & McElwain, 2002).

The positive relationship between species richness and topographic range recurred only in mean lineage net diversification at the species level (MS/G), represented by a negative MAT relationship in the filter model. This lends support to the view that late-Tertiary mountain uplift (Burnham & Graham, 1999) is an important driver of recent diversification in New World palm flora. In fact, the largest New World palm genus, Chamaedorea (77 species), is strongly associated with mountainous regions, as are several other species-rich genera, notably Aiphanes (22 species), Wettinia (21 species), Ceroxylon (11 species) and Prestoea (11 species; Henderson et al., 1995). Orogeny may have increased net diversification rates by providing a larger area featuring cool tropical environments and by creating barriers to dispersal, thereby promoting adaptive radiation and allopatric speciation. Nevertheless, topography is clearly a less important determinant of New World species richness patterns for palms than for birds (Fjeldså, 1994; Rahbek & Graves, 2001), perhaps reflecting the fact that birds have adapted much more readily to cold climatic conditions.

The most consistent pattern found was a linear decrease in palm species richness and all measures of mean lineage net diversification with increasing absolute latitude, away from the equator (Tables 3 & 4). This relationship was strongly supported, no matter what climatic or topographic variables were included as explanatory factors, with or without spatial filters. Standardized regression coefficients indicate that it is nearly always one of the strongest relationships. These findings show that a clear latitudinal diversity gradient exists at both large and small scales, even when key aspects of climate are accounted for. In addition, the latitudinal gradients in mean lineage net diversification at both deep and shallow phylogenetic levels suggests that the relationship between species richness and latitude is driven mainly by a latitudinal gradient in net diversification rates, and that this gradient has been operative throughout the evolutionary history of New World palms.

What drives the relationship between latitude and net diversification?

Our analysis included key measures of climate and productivity. Therefore direct effects of contemporary temperature, water–energy dynamics and productivity can reasonably be ruled out as causes of the persistent diversity–latitude relationships, although spill-over effects may occur. Another potential environmental contributor to the rapid diversification at equatorial latitudes is intense ultraviolet radiation, especially in tropical mountains (Lee & Lowry, 1980). However, this explanation is probably not relevant to the many genera that are largely confined to the heavily shaded forest understorey. Figure 1 clearly shows that a much greater area is available to palms in equatorial South America than away from the equator. Hence an area effect (Rosenzweig, 1992; Chown & Gaston, 2000) seems plausible. An area effect would also explain the consistent negative correlation between absolute latitude and mean lineage net diversification across the range of phylogenetic levels considered, as the equatorial South American region has remained large and inhabited by palms throughout the evolutionary history of the family. Palms already dominated the huge equatorial Palmae palynofloral province (most of South America, Africa, India and perhaps parts of Southeast Asia) during the late Cretaceous and early Tertiary (Morley, 2000), suggesting that conditions, perhaps climate and/or area, were already conducive to greater rates of net diversification in this region when palms began their rapid initial diversification. Therefore our results are consistent with the time-integrated area effect reported for biome-level patterns of tree diversity (Fine & Ree, 2006). Such an explanation could also account for the higher-than-expected species richness, MRD and MS/G in south-eastern Brazil and western Amazonia (Fig. 3). In contrast, the higher-than-expected species richness in Central America seems unlikely to reflect area effects, given the relatively small area of this region, and the fact that it is mirrored mainly in high mean lineage net diversification at shallow phylogenetic levels (Fig. 3).

A time-integrated area effect driven by Tertiary cooling

A time-integrated area effect as an explanation for current patterns in palm species richness is consistent with the Tertiary development in climate and its impact on the overall area habitable by palms. First, it is important to note that most palm species diversity in the New World appears to be of mid- to late-Tertiary origin, as discussed above. In the late Cretaceous and early Tertiary, palms were present from Patagonia in the south to Greenland in the north (Uhl & Dransfield 1987; Wilf et al., 2003), but the subsequent intensification of the latitudinal climatic gradient (Ruddiman, 2001) caused palms to retract progressively from extratropical latitudes (Uhl & Dransfield, 1987) due to their very limited frost tolerance (Jones, 1995). As the extratropical regions became less favourable to palms and their habitable portions decreased in area, extinction rates will have increased and speciation rates decreased. Hence, the extratropical palm assemblages, especially those located at the more geographically isolated northern margin (Florida and the Caribbean islands), are characterized by many relict taxa (lineages with low net diversification rates): notably the coryphoid subfamily (e.g. Sabal), with its extensive Northern Hemisphere boreotropical history (Bjorholm et al., 2006), and a number of other Central American/Caribbean lineages with little net diversification above the species level (e.g. Pseudophoenix of the Ceroxyloideae and Chamaedorea and its allies Roystonea and Reinhardtia of the Arecoideae; Figure S1). These northern non-coryphoid lineages may also have a particularly long history in the Central American–Carribean region. Putative fossils of Pseudophonix and Chamaedorea are known from the early Tertiary period of North America (Berry, 1914; Melchior, 1998). This explanation is consistent with the strong decrease in MRD measures towards the north (Tables 3 & 4). At the same time, the huge equatorial South American region has continuously provided ideal warm and wet conditions for palms since the late Cretaceous (Kahn & de Granville, 1992; Morley, 2000; Willis & McElwain, 2002; Colinvaux, 2005). Hence, the current patterns of palm species richness and mean lineage net diversification are probably the result of low rates of net diversification in extratropical regions due to extinctions, especially in the north, and sustained high net rates of diversification in equatorial South America. Coupled with phylogenetic niche conservatism (Wiens & Graham, 2005), this explanation is also consistent with the observed species richness–climate relationship (as discussed above).

Conclusions

We suggest that the geographical patterns in species richness of palms across the New World may predominantly be the result of elevated net diversification rates towards the equator and in wet, warm climates that have been sustained throughout most of the Tertiary. Late-Tertiary orogeny appears to have caused localized increases in net diversification rates that have made a small but consistent mark on the overall species richness pattern. Overall, a combination of time-integrated area effects and phylogenetic niche conservatism may offer the most consistent explanation for the relationships of species richness and mean lineage net diversification to absolute latitude and modern environment. The particularly low rates of net diversification that characterize the palm assemblages at the polar margins of the palm-inhabited region probably reflect the fact that palms have suffered severe extinction in the extratropical regions due to the Cenozoic cooling trend, and have largely been unable to adapt to the expanding cooler climates.

Acknowledgements

We are grateful for economic support from the Danish Natural Science Research Council (grant 21-04-0346 to J.C.S. and grant 272-06-0476 to H.B.) and the Carlsberg Foundation (grant 04-0445/10 to F.B.).

Biosketches

Jens-Christian Svenning’s research focuses on ecoinformatics, macroecology, biogeography and community ecology of plants and mammals, and the application of these fields in global change biology.

Finn Borchsenius’s main research interests are plant systematics, phylogeny and biogeography, with special focus on palms and Marantaceae.

Stine Bjorholm studies American palm biogeography with special focus on macro-scale distribution patterns and their underlying causes.

Henrik Balslev studies palms, with special interest in the distribution of species richness and how it relates to environmental conditions and historical events, and how palm species richness affects local people.

Editor: Bradford Hawkins

This article arose from a paper presented at the third biennial meeting of the International Biogeography Society, held in Puerto de la Cruz, Tenerife, Canary Islands, 9–13 January 2007.

Ancillary