The Botanical Review

, Volume 74, Issue 1, pp 178–196

The Impact of Ecology and Biogeography on Legume Diversity, Endemism, and Phylogeny in the Caribbean Region: A New Direction in Historical Biogeography


    • Plant Sciences and Plant PathologyMontana State University
  • Angela Beyra Matos
    • Centro de Investigaciones de Medio Ambiente de Camagüey, CITMA

DOI: 10.1007/s12229-008-9006-8

Cite this article as:
Lavin, M. & Matos, A.B. Bot. Rev (2008) 74: 178. doi:10.1007/s12229-008-9006-8


The legume family is so well represented in the Caribbean that if a preserve was needed somewhere on earth to harbor all of the primary lineages in this family, the flora of just Cuba would suffice. Molecular phylogenetic, biogeographic, and evolutionary rates analysis all suggest that legume diversity and endemism in the Caribbean are mostly of recent origin and are likely a function of the abundance of seasonally dry tropical forests (SDTFs) throughout the neotropics. Legumes have a strong ecological affinity for SDTFs, and the Caribbean basin is well covered by this forest type. Rate-variable molecular clock analysis suggests that the majority of worldwide island lineages of legumes have ages of much less than 30 Ma. Singular historical events invoking land bridges or mobile continental plates are thus not needed to explain Caribbean legume diversity and endemism. The Greater Antilles are large islands located close to the American continent. They are therefore expected to fairly represent the diverse continental lineages of legumes. Yet, they are distant enough to be dispersal limited. As such, island lineages can speciate and diversify over evolutionary time unimpeded by high rates of immigration from the mainland. Vicariance and other standard phylogenetic methods of historical biogeography are likely to be replaced by those of ecological and island biogeography. This is because model selection approaches derived from the neutral concept of isolation by distance will be able to quantify patterns of alpha and beta diversity and detect niche assembly and phylogenetic niche conservatism within and among metacommunities that are hypothesized to constrain phylogeny.


La familia leguminosa está tan bien representada en el Caribe, que si fuera necesario preservar algún sitio sobre la tierra, que albergue todos los linajes principales de esta familia, la isla de Cuba, con su flora endémica, podría ser seleccionada entre las áreas que cumplen esta condición. Todos los análisis moleculares, filogeneticos, biogeográficos y de tasas de evolución sugieren que la diversidad y endemismo de las leguminosas en la región del Caribe, es en la mayoría de los casos, de origen reciente y es probablemente una función de la abundancia de los bosques tropicales estacionalmente secos (SDTFs) a lo largo del neotrópico. Las leguminosas tienen una preferencia ecológica fuerte por los bosques tropicales estacionalmente secos (SDTFs), y la cuenca del Caribe está bien cubierta por este tipo de bosque. El análisis molecular de tasa variable sugiere que la mayoría de los linajes de leguminosas de las islas tienen edades mucho menores que 30 millones de años. De este modo, los eventos históricos singulares que invocan puentes terrestres o placas continentales móviles, no necesariamente explican la diversidad y endemismo de Leguminosas del Caribe. Las Antillas Mayores son islas grandes localizadas relativamente cerca del continente americano. Por consiguiente, se espera que en estas islas, estén bien representados los diversos linajes continentales de leguminosas. A pesar de todo, estas islas están bastante distantes de las lagunas oceánicas, lo cual rinde en las islas grandes una dispersión algo limitada. De esta manera, los linajes de estas islas pueden especiar y diversificarse a escala de tiempo evolutivo, sin impedimento por altas tasas de inmigración desde el continente. Así, los métodos de biogeografía de la vicarianza y otros métodos filogenéticos estándar de Biogeografía Histórica tienen la probabilidad de ser sustituidos por los métodos ecológicos y de Biogeografía de las islas. Esto se debe a que los métodos de selección del modelo derivado del concepto neutral de aislamiento por distancia permitirá cuantificar los patrones de alfa y beta diversidad y detectar desviaciones de la relación positiva fuerte entre las distancias geográficas y genéticas (niche assembly) y la conservación de las preferencias ecológicas ancestrales que tienden a heredar las especies (phylogenetic niche conservatism) dentro y entre comunidades que son ensayadas para formular hipótesis sobre el papel de la Biogeografía y la Ecología en la determinación de la filogenia.


Molecular phylogenetic analysis has been revealing of many unsuspected findings, and this has been particularly true of studies in the legume family. For example, phylogenies that exhaustively sample at the species level and below have shown that geography and ecology are often as good or better predictors of relatedness than is morphology. In addition, the estimated age of the divergences of legume clades separated from each other either by an ocean gap or on different continents typically postdates the estimated ages of the putative tectonic explanations for continental vicariance biogeography (e.g., Lavin et al., 2004), which is a general finding (e.g., Queiroz, 2005). If such ecological and geographic structure is commonly found among legume phylogenies, and tectonic history is not responsible for such structure, then ecological processes including island biogeographical ones must explain the phylogenetic structure that has so profoundly influenced recent legume taxonomies (summarized in Lewis et al., 2005).

The genus Strophostyles exemplifies the influence of ecology and biogeography on phylogeny (Riley-Hulting et al., 2004). With three species confined to open woodland sites in southeastern United States, Strophostyles is not related to an Asian lineage, as might be expected (e.g., Tiffney, 1985), but rather is sister to Dolichopsis, which comprises two endemic species to the Chaco region in South America. The chaco is forest vegetation distinguished in part from other adjacent tropical dry forests or tropical woody savannas by having a frost period (Pennington et al., 2000). No morphology predicts the sister relationship of Strophostyles and Dolichopsis, but geography and ecology does. This sister group relationship is confined to New World temperate savanna-type vegetation. Because ecology and geography are often excellent predictors of phylogenetic relatedness, they should get as much attention from a methodological perspective as does morphology.

The important ecological focus in terms of Caribbean legumes is on the seasonally dry tropical forests (SDTFs sensu Prado & Gibbs, 1993; Pennington et al., 2000, 2004). SDTFs are rich in cacti, euphorbs, and other succulent taxa, in addition to families like legumes. SDTFs are poor in grasses and other taxa adapted to regular burning. These two facets of SDTFs strongly suggest that this vegetation type is persistent, or that residents do not suffer routine wholesale death and replacement especially from immigrants. This vegetation occurs as fragments scattered throughout the neotropics including the Caribbean Islands and adjacent North and South American mainland, as well as the African Horn region (Somalia-Masai) and southwestern Africa (the Karoo-Namib; Schrire et al., 2005).

Legume clades are very often confined to this vegetation, which suggests that a species tends to inherit its ancestral ecological predilection (i.e., phylogenetic niche conservatism; Harvey & Pagel, 1991). Legume clades also are very often confined to one patch of STDF (e.g., Zygocarpum and Wajira in the Somalia-Masai; Thulin & Lavin, 2001; Thulin et al., 2004), which suggests that SDTF species tend to inherit their approximate ancestral geographic position (i.e., dispersal limitation; Hubbell, 2001). The genus Chapmannia, for example, is endemic to SDTF-type vegetation and comprises a Somalia–Masai clade sister to a Florida–Mesoamerican clade (Thulin, 2000; Lavin et al., 2001). The estimated age of the trans-Atlantic Chapmannia crown is 14.2 ± 1.7 Ma (Lavin et al., 2004). Stochastic dispersal is therefore the only explanation for this ecologically confined and highly geographically structured Chapmannia phylogeny.

The general points to be drawn from the above discussion and that pertain to island biogeography are first, dispersal is not an ad hoc explanation or an “anything goes” concept, as suggested by Nelson and Platnick (1981). It is a stochastic process operating within potentially definable ecological confines and under specific island biogeographic rules. Quantifying these confines and rules should be part of dispersal biogeography. Second, oceanic and continental islands are no different from the ecological or geographical (e.g., topographical) islands located within continents, from an island biogeographic perspective. Islands surrounded by water may be discrete to the human eye. Openness to visual inspection, however, is not a prerequisite to the study of the processes causing niche conservation and dispersal limitation on ecologically and geographically determined islands lying within continents.

The Recency of Legumes in the Antilles

The legume family is well represented in the Antilles from both a species and phylogenetic diversity perspective. The diversity of this family in the Antilles is not likely a function of its antiquity in the region. First, the diversity of legume species on the Caribbean islands is at a higher level than expected for islands in general (Fig. 1, Table 1). This could well be mostly a function of the relatively large cover of SDTFs in the Caribbean Basin and elsewhere in the neotropics (see maps in Pennington et al., 2000; Schrire et al., 2005). Island biogeographically, both the mainland source and island target SDTF areas are large. Jamaica has a total number of legume species generally expected of the Caribbean Islands even though it was submerged during the middle Tertiary (Buskirk, 1985). Using Akaike Information Criteria (Burnham & Anderson, 2002; Johnson & Omland, 2004), a model comprising the continuous variable of island size and a categorical variable indicating a continuous (or not continuous) Tertiary biota was determined to be as likely or more so than other models comprising additional explanatory variables such as distance from mainland and latitude (Fig. 1). We hypothesize that the reason all of the Caribbean Islands are rich in legume species is because of the affinity of legumes to the SDTFs (McKey, 1994; Pennington et al., 2000). We thus predict that the Caribbean island SDTFs with respect to taxa like legumes differ from American continental SDTFs only as a function of dispersal limitation (i.e., geographical distances will be a good predictor of community phylogenetic distances).
Fig. 1

A model explaining legume species richness on islands throughout the world. The Caribbean islands are indicated with solid circles or triangles (these 16 islands are listed in Table 1). Continental islands with a continuous Cenozoic flora are indicated with triangles. Significance values are reported for the slopes, which in the case of the “old” islands is not significant. This model suggests that island area for the “old” islands is not a good predictor of legume species richness if only because all of the “old” islands are about equally rich in legume species

Table 1

Islands of the Antilles for which Data on Total Number of Legume Species and Infraspecific Taxa (N) and the Number of Endemic Legume Species and Infraspecific Taxa (Nend) Could be Estimated Using Floristic Accounts, which were Updated with More Recent Monographic Treatments of Particular Legume Genera (Lavin et al., 2001)
















Howard et al. (1988)








Gooding et al. (1965)








Correll and Correll (1982)








Sauget and Liogier (1951); A. Barreto Valdés and A. Beyra Matos (unpublished data)








Nicolson (1991)








Howard et al. (1988)








Howard et al. (1988)








Liogier (1985)








Adams (1972)








Howard et al. (1988)








Howard et al. (1988)

Puerto Rico







Britton and Wilson (1924)

St. John







Acevedo-Rodríguez (1996)

St. Kitts







Howard et al. (1988)

St. Lucia







Howard et al. (1988)

St. Vincent







Howard et al. (1988)

The age column refers to two categories where “old” signifies a continuous Cenozoic terrestrial flora and “young” indicates the island has emerged since the middle Tertiary or later because they are either oceanic or reemergent continental islands. Also reported are island area in km2, distance from the nearest continent in km (dist), and northern latitude (lat).

Second, endemic legume species diversity is not just a function of area, but also of whether an island flora has had a continuous Cenozoic history; that is, not submerged during the mid-Tertiary like Jamaica (Buskirk, 1985) or not oceanic like the Lesser Antilles (Fig. 2). The re-emerged Jamaica has had only since the middle Tertiary for endemic species to accumulate. This could explain the slightly fewer than expected endemic species on this island with respect to older island biotas (Fig. 2). Regardless, the diversity of endemics in the Caribbean is at a level expected for islands in general. Using Akaike Information Criteria, a model comprising the continuous variable of island size and a categorical variable indicating a continuous (or not continuous) Tertiary biota was determined to be the most likely compared to other models comprising additional explanatory variables such as distance from mainland and latitude (Fig. 2). A persistent Cenozoic flora as an important explanatory variable for numbers of endemic species was initially thought by Lavin et al. (2001) to be a signature of antiquity, such as those Tertiary vicariant events that have been proposed for the Greater Antilles (e.g., Rosen, 1978; Iturralde-Vinent & MacPhee, 1999). We hypothesize now that in addition to area in explaining endemic legume species richness (cf. anolis lizards; Losos & Schulter, 2000), time, as a proxy for stability of a habitat like the seasonally dry tropical forest, may also be an important explanatory variable.
Fig. 2

A model explaining endemic legume species richness on islands throughout the world. Caribbean islands are indicated with solid circles or triangles (these 16 islands are listed in Table 1). Continental islands with a continuous Cenozoic flora are indicated with triangles. Significance values are reported for the slopes. This model suggests that endemic species richness is a function of a continuous Cenozoic biota

Third, the Antilles harbor much legume phylogenetic diversity. For example, the total tree length of the legume matK chronogram scaled in millions of years (Ma) is 2,204 Ma, while that of the Cuban lineages = 730 Ma, or about one-third the total (Fig. 3). This was calculated using the Paloverde program (Sanderson, 2006) and a penalized likelihood rate smoothed Bayesian consensus phylogeny (Lavin et al., 2005). This illustrates that the legume family in Cuba comes from many of the phylogenetically divergent or deep branching lineages of the family. Both area (cf. Losos & Schluter, 2000) and evolutionary stability of SDTFs (cf. Lavin, 2006; Pennington et al., 2004, 2006) could allow the accumulation of many lineages regardless of the extinction of particular lineages.
Fig. 3

A cladogram of the Leguminosae derived from over 500 matK sequences sampled from across all the diverse lineages of the legume family. The most recent common ancestor (MRCA) of the Caesalpinioideae is labeled C. The MRCA of the Mimosoideae is labeled M, and the MRCA of the Papilionoideae is labeled P. Lineages leading to terminal taxa represented in the Cuban flora are traced in red

The reason for interpreting the above three findings with island biogeographic explanations is that historical ones can be ruled out. First, unequivocally old Antillean lineages (i.e., greater in age than 40 Ma), as exemplified by Arcoa of the Umtiza clade (Fig. 4), are rare among legumes. The minimum age of the entire Umtiza clade and the divergence of Arcoa is over 50 Ma (Lavin et al., 2005). Other candidate legumes endemic to the Antilles and that could be as old as Arocoa are unknown and perhaps rare. Area relationships derived from the taxon-area phylogeny of the Umtiza clade (Fig. 4) via any of the standard cladistic vicariance methods would yield nonsensical results from a continental tectonic perspective (e.g., a close relationship of South Africa with North America and eastern Asia, or Hispaniola with Old World continents). The limited ecological amplitude of the Umtiza clade (i.e., inhabiting mostly SDTF or temperate vegetation) reveals that even over long periods of time the ancestral ecological predilection is phylogenetically constrained. The SDTF ecological setting is unequivocally optimized at the ancestral node using traditional vicariance approaches (Schrire et al., 2005). Such an unequivocal ancestral optimization cannot be attained for geographical setting or for most morphological characters (see acctran optimizations in Fig. 1 of Herendeen et al., 2003), suggesting that the biogeography of the Umtiza clade is more ecologically than historically determined.
Fig. 4

The Umtiza clade (adapted from Herendeen et al., 2003; Lavin et al., 2005). An old Caribbean lineage is exemplified by Arcoa, a monotypic genus endemic to Hispaniola and distributed in seasonally dry tropical forests (SDTF), a vegetation that is rich in succulent taxa but low in grass abundance (mean age and standard deviation are reported in Ma). The monophyly of the Umtiza clade is consistently resolved but weakly supported, suggesting it harbors ancient constituent lineages

Second, endemic Antillean diversifications, such as those represented by Pictetia and Poitea (Fig. 5), that have ages of 20 Ma or less are uncommon among legumes. Lineages like Arcoa, Pictetia, and Poitea were targeted for phylogenetic systematic analysis because traditional taxonomic studies suggested they might represent isolated lines from the early Tertiary that bear a signature of a unique historical event. Such a signature was never forthcoming (Lavin et al., 2001), so island biogeographic explanations were sought (Lavin et al., 2004). Although the ages of the Pictetia and Poitea extant diversifications within the Greater Antilles are less than 10 Ma, corresponding stem clade ages (i.e., the maximum age of these Caribbean diversifications) are less than 20 Ma. This suggests that continental tectonic history could never have left an imprint on these lineages. Molecular data once again revealed how ecology (island biogeography) strongly constrained the phylogenetic history of Pictetia and Poitea, which have a strong affinity for SDTFs.
Fig. 5

Middle-aged transoceanic Caribbean crown clades (i.e., with an estimated age of about 10–30 Ma) are exemplified by the legume genera Pictetia and Poitea (adapted from Beyra & Lavin, 1999; Lavin et al., 2000, 2003; mean ages and standard deviations are reported in Ma)

Third, in contrast to the Arcoa, Pictetia, and Poitea examples, numerous legume lineages like that represented by Vigna ekmanii (Fig. 6) were not targeted for study because the search for a signature of antiquity was the original aim of our historical biogeographic studies. Most of the species of Phaseolinae are lianas that occupy tropical and subtropical forested and savanna-like habitats. Again, ecology is the obvious determinant of the geographical distribution in Vigna and relatives. In the Antilles, widespread or endemic species belonging to genera otherwise best represented outside the Antilles are common. This reveals that most legume species in the Antilles will be shown to have a phylogenetic signature of recency similar to that of Vigna ekmanii.
Fig. 6

A young transoceanic Caribbean crown clade (i.e., with an age less than 1 Ma) is exemplified by Vigna ekmanii and its mainland neotropical sister species (adapted from Riley-Hulting et al., 2004; Thulin et al., 2004; Delgado-Salinas et al., 2006). Even the age of the formation of the geographic phylogenetic structure observed at the largest scale, the African, Asian, and New World Phaseolinae subclades, is less than 12 Ma. This is too young to be explained by continental tectonic history (mean ages and standard deviations are reported in Ma)

Finally, the move from a search for unique historical events to general ecological processes in explaining biogeographic patterns was made particularly compelling because the ages of worldwide trans-oceanic legume crown clades were determined to be mostly less than 30 Ma (Fig. 7). Given that all historical explanations invoke land bridges and similar events with ages that are all over 30 Ma, an alternative to historical biogeography is needed. This implicates transoceanic dispersal rather than tectonic continental history for two reasons. First, the age of the Aves Ridge is estimated at 33–40 Ma (e.g., Iturralde-Vinent & MacPhee, 1999). The age of the Walvis Ridge is estimated at about 35 Ma (e.g., Morley & Dick, 2003). The age of a North Atlantic land bridge is estimated at the youngest to be 33 Ma (e.g., Tiffney, 1985). Even the age of the most recent land bridge potentially connecting Madagascar and Africa is estimated at 26–45 Ma (Yoder et al., 2003). Second, the results depicted in the histogram are extremely biased toward old ages. Further sampling of the additional 18,975 species of legumes would render more recent ages. This is because all the divergent legume lineages are represented with 350 species data set. Additional sampling will only fill in the gaps of the legume phylogeny. This last point is underscored by a potential bias in the sampling of matK sequences. Legumes like Arcoa were specifically targeted because they represent potentially old and isolated lineages. So were legume groups represented by Pictetia and Poitea. What were specifically not sampled were all of the very many young transoceanic crown clades, like that represented by Vigna ekmanii and its mainland sister species of Phaseolinae (Fig. 7). The general point is that the diversity of legumes in the Caribbean Islands will ultimately be shown to have little antiquity if only because few transoceanic legume clades have the level of antiquity that would be consistent with vicariance biogeographical explanations.
Fig. 7

The age distribution of worldwide trans-oceanic crown clades (TOCCs) is derived from Lavin et al. (2004), and estimated from the legume phylogenies originally presented in Wojciechowski et al. (2004) and Lavin et al. (2005). At that time, about 350 species of legumes were sampled to resolve just the relationships among the early branching lineages. Sampling did not occur with respect to any biogeographic question. TOCC ages were recorded by tracing from the tips of the matK chronogram backward in time until a most recent common ancestor (MRCA) was first encountered that spanned two continents separated by an ocean gap. This histogram illustrates that many such ages are less than 20 Ma, and nearly all are less than 30 Ma

An Ecological and Island Biogeographic Alternative for Historical Biogeography

Cladistic vicariance and related approaches to the study of historical biogeography have been preoccupied with detecting the influence of historical events on phylogenetic patterns. An influence is explained by the fit of a taxon-area cladogram to competing continental tectonic hypotheses. In the Caribbean basin, such competing tectonic hypotheses have involved the eastward movement and reconfiguration of the Greater Antillean islands (Rosen, 1978) or the presence of a once-emergent but now submerged Aves Ridge (Iturralde-Vinent & MacPhee, 1999). The fit of the taxon-area cladogram via component, three-area statements, or Brooks Parsimony Analysis has always been a nebulous affair and without definitive results. In part this is because these different approaches each involve multiple or different assumptions, none of which can be objectively evaluated one against the other (e.g., Lavin et al., 2001). Also, choosing among taxon-area cladograms with the same histories (a requirement for vicariance analysis) is difficult given that absolute ages cannot be estimated from a cladogram and relative ages are at best weakly implicit. Finally, clade support is generally weak for area cladograms. For example, a minor change in the area coding for a single wide spread species (e.g., adding an additional area of endemism to its distribution) can result in an entirely different area cladogram (e.g., Pennington et al., 2004).

In contrast, ecological and island biogeographic approaches to the study of biogeography are concerned with detecting enduring processes that influence phylogenetic patterns. Such approaches integrate plant phylogeny with information on plant community composition to ask questions such as: How does island size affect endemic speciation (cf. Losos & Schluter, 2000)? Does the patchy distribution of SDTFs on the Caribbean Islands influence ecological (e.g., beta diversity) and phylogenetic patterns (e.g., geographic structure) in the same manner as the patchy distribution of SDTFs on the mainland? To what degree are Caribbean clades ecologically structured? For example, how common is the phylogenetic niche conservatism exemplified by Poitea, a clade confined to Antillean SDTFs excepting the subclade comprising P. florida, P. sabinea, and P. carinalis that occupies wet forest? Do particular vegetation types impose different phylogenetic histories on constituent clades? That is, do wet forest clades have greater success at over-water dispersal because of the intrinsic attributes of this vegetation (e.g., successful immigration facilitated by high rates of resident death due to drought)? After all, Poitea carinalis from the wet forests of Dominica is the only species of the genus to occur outside the Greater Antilles (Lavin, 1993), whereas SDTF-confined Pictetia has no occurrence outside the Greater Antilles (Beyra-Matos & Lavin, 1999). Do the SDTFs (sensu Prado & Gibbs, 1993; Pennington et al., 2000) in the Caribbean and elsewhere impose a different history on phylogenies compared to other dry tropical vegetation types? Savannas could be much like wet forest in experiencing higher rates of resident mortality because of drought, for example, which facilitates high rates immigration. In contrast, drought-resistant SDTFs show little evidence of fire-resistance, indicating little if any historical disturbance via fire. Thus, SDTFs could be much more dispersal limited than other tropical forest types because of their persistent nature (e.g., Lavin, 2006) and because of the fragmented distribution of this vegetation. If dispersal limited as such, then constituent clades of SDTFs will show phylogenies with a higher degree of geographical structure and constituent SDTF communities will show lower degrees of alpha and higher degrees of beta diversity compared to constituent clades or communities from other types of tropical vegetation types, including wet forests and savannas.

An ecological and island biogeographic approach to ecology that involves phylogeny has been suggested by Hubbell (2001), where speciation rates and the degree of dispersal limitation shape the phylogeny of constituent lineages within a metacommunity. The abundant recent literature on neutral ecological theory is not explicit about how ecology impacts phylogeny, however (e.g., Holyoak & Loreau, 2006; Hu et al., 2007; Munoz et al., 2007; Volkov et al., 2007). An explicit approach to investigating the role of biogeography and ecology in shaping phylogeny can be derived from the population genetics concept of isolation by distance (e.g., Grefen et al., 2004; Jensen et al., 2005). A strong positive relationship between genetic and geographic distance is essentially a neutral assumption. Such a relationship could illuminate relevant community level properties such as dispersal limitation (dispersal assembly) caused by, for example, the patchy but persistent distribution of a metacommunity. Species may not occupy their total potential range because of inherent properties of the metacommunity that render geographically close local communities similar in species or sister species composition. This is akin to the population genetic findings for North American wolves, where geographically close populations are on average genetically similar (Grefen et al., 2004). Deviations from a strong positive relationship between geographical and genetic distances imply a deterministic cause or niche assembly. Wolves from cold temperate habitats were genetically dissimilar from those of warmer southern latitudes regardless of the distance separating populations (Grefen et al., 2004). This implicates an interaction of climate (niche assembly) and geographic distance (dispersal assembly) in structuring wolf populations.

Similarly, a strong positive relationship between community composition distances and community phylogenetic distances to geographic distances is also a neutral assumption, or evidence of dispersal limitation (Hubbell, 2001; Hardy & Senterre, 2007). Deviations from neutrality here would implicate niche assembly (using community composition distances as the response) or phylogenetic niche conservatism (using community phylogenetic distances). As such, a model selection approach can be set up such that geographic (stochastic) and environmental (deterministic) explanatory variables (all pairwise distances) are used to predict community composition distances and community phylogenetic distances (e.g., also all pairwise, such as the net relatedness index of Webb et al., 2002).
  1. (a)

    Species abundance data and phylogeny. Tuomisto et al. (2003), Vormisto et al. (2004), Hardy and Sonke (2004), and particularly Hardy and Senterre (2007) show how biodiversity can be modeled using community composition distances or community phylogenetic distances as response variables. Hardy and co-workers show that the best form of plant community data is relative species abundances because such data can be modeled using derivations of Simpson’s diversity index. Hardy and Senterre (2007) derive community composition distances and community phylogenetic distances from the Fst statistic. In this manner, the response variables (community composition and community phylogenetic distances) are very explicit measures of alfa and beta diversity. The community phylogeny from which community phylogenetic distances are derived, however, must be ultrametric (e.g., chronograms or rate smoothed phylograms) in order to achieve unbiased estimates of community differentiation along a spatial, temporal, or environmental gradient. Chronograms with branch lengths measured in ages (Ma) can be produced from Phylomatic ( or a community phylogeny can be generated from molecular data and then rate smoothed (e.g., via penalized likelihood rate smoothing; Sanderson, 2002) calibrated using absolute or relative time.

  2. (b)

    Species incidence data and phylogeny. Ultrametric community phylogenies may not be readily available for plant communities or confidence in the branch lengths of such phylogenies could be low. Also, relative species abundances may not be easily obtained because, for example, herbs, shrubs, and trees are being censused, or the plots or sampling sites may be highly altered by human activity. In this case, response variables such as the net relatedness index (Webb et al., 2002), a phylogenetic distance generated from Webb’s Phylocom ( using node numbers or branch lengths from non-ultrametric trees, could be utilized as the response variable. Similarly, Jaccard’s distances, or any community composition distance derived from species incidence rather than abundance data, could be modeled as the response variable.

  3. (c)

    Model selection approaches. Either using abundance or incidence data, hypothesis testing can involve whether a particular set of a priori targeted vegetation types, biomes, or habitats are differentially constraining the phylogeny of a taxonomic group or community from an ecological and geographical perspective (i.e., different metacommunities are potentially being identified). The general model includes geographic distance (e.g., Euclidean distances derived from geographical coordinates; Legendre, 1990) and at least a categorical variable indicating in which targeted vegetation type, biome, or habitat (i.e., hypothesized metacommunity) the particular sample sites are located. In this manner, an interaction between ecological setting and geographic distance can be evaluated. Using mantel correlations and model selection approaches (e.g., Burnham and Anderson, 2002; Johnson and Omland, 2004), within versus among metacommunity comparisons can be used to determine the significance of the interaction.

If Jaccard’s or any community composition distance is the response and the interaction between geographic distance and metacommunity type is determined to be important, then the model selection approach is suggesting that two hypothesized metacommunities in which sample sites are located are imposing niche assembly (Fig. 8). If the slope of the among pairwise comparisons is negligibly different from that of the within pairwise comparisons, this would suggest that dispersal assembly adequately explains community composition among all sample sites (i.e., the two hypothesized metacommunities are one and the same).
Fig. 8

A model depicting the differences in community composition among pairwise comparisons of community samples as a function of the interaction of metacommunity type and geographical distance. Jaccard’s distance is the response. Sample sites and species incidence data are imaginary. All community pairwise distances are indicated by a categorical variable as within versus between metacommunity comparisons. Hypothesized metacommunities could include the SDTF versus wet forest throughout the Caribbean basin, or SDTFs on the Caribbean Islands versus SDTFs from the American mainland. If niche assembly is occurring between hypothesized metacommunities, then the slope of the among pairwise comparisons (gray line) will be relatively flat and with a higher intercept compared to that of the within comparisons (black line). This is because among comparisons will be different no matter the geographic distance between sites. In this case, ecology is imposing a constraint on dispersal among hypothesized metacommunities

If community phylogenetic distance is the response and the interaction between geographic distance and metacommunity type is determined to be important, then the model selection approach is suggesting that the two hypothesized metacommunities in which sample sites are located are imposing phylogenetic niche conservatism. If the slope of the among pairwise comparisons is negligibly different from that of the within pairwise comparisons, this would suggest that all sample sites are phylogenetically homogeneous and that the two hypothesized metacommunities are not imposing long term barriers to speciation events (Fig. 9). Even if ecology imposes a short term barrier to immigration (Fig. 8), a long term or phylogenetic barrier may not necessarily follow (Fig. 9). In this case, any metacommunity designation would imply short rather than long-term integrity. Such “short term only” metacommunity distinction may well be the case for Caribbean Island versus American mainland SDTFs. Long-term metacommunity distinction (where the points in Fig. 9 would be distributed more like those in Fig. 8) would likely be the case for Caribbean SDTFs versus other types of dry forests or savannas, or SDTFs and wet forests. This is because these vegetation types tend to be inhabited by different phylogenetic lineages (e.g., Lavin, 2006).
Fig. 9

A model depicting the differences in community phylogenetic distances (e.g., net relatedness index, NRI) among pairwise sample comparisons as a function of the interaction of metacommunity type and geographical distance. Sample sites and species incidence data are imaginary. All community pairwise distances are indicated by a categorical variable as within versus between metacommunity comparisons. Hypothesized metacommunities could include the SDTF versus wet forest throughout the Caribbean basin, or SDTFs from the Caribbean Islands versus SDTFs from the American mainland. If phylogenetic niche conservatism is occurring between hypothesized metacommunities, then the slope of the among pairwise comparisons will be relatively flat and with a higher intercept compared to that of the within pairwise comparisons. This is because among comparisons will be phylogenetically different no matter the geographic distance separating sites. In this imaginary case, however, slopes and intercepts of the within (black line) versus among (gray line) comparisons are nearly identical. This is the sort of evidence that would suggest the hypothesized metacommunities are not imposing an ecological barrier to speciation events, and that the two metacommunities are phylogenetically homogeneous, or nearly so

In either of the above examples, geographic distance no longer has to be considered a proxy for deterministic environmental variables. Specific quantitative environmental variables can be introduced into a model (cf. Vormisto et al., 2004; Hardy & Senterre, 2007) with or without geographic distance, and selection among competing models (cf. Burnham & Anderson, 2002) can settle the issue.

We have not yet undertaken an integrated study of tropical legume communities and phylogenies for the Caribbean, but are beginning to form collaborations among legume systematists and ecologists to conduct such studies on the American mainland (e.g., Oliveira-Filho et al., 2006, 2007) with the intention of expanding into the Caribbean region. The imaginary data set analyses (Figs. 8 and 9) are derived from current studies (e.g., Oliveira-Filho et al., 2007; Synder et al., 2007) and are intended only to illustrate the approach that should replace traditional phylogenetic methods of biogeography. Our ultimate intent will be to use this new approach to test the prediction that the SDTFs form a distinct metacommunity from other tropical vegetation, including seemingly similar savannas that are rich in grass species and prone to regular intervals of fire disturbance, in contrast to SDTFs. We also want to test the hypothesis that the SDTFs from the Caribbean Islands differ from American mainland SDTFs only as a function of dispersal limitation (e.g., Fig. 9). That is, they are neither ecologically distinct nor taxonomically depauperate, and harbor the same phylogenetic lines at similar levels of diversity as do the rest of the scattered patches of neotropical SDTFs. This would counter a traditional view generally applied to the Caribbean Island biota, as summarized in Hedges (2000), which is that it is depauperate at higher taxonomic levels; i.e., “the central problem” identified by Williams (1989).


We wish to thank Alfonso Delgado Salinas, R. Toby Pennington, Martin F. Wojciechowski and for constructive comments and other assistance, which greatly improved the manuscript.

Copyright information

© The New York Botanical Garden 2008