We here explore the use of a Bayesian approach to island biogeography for disentangling the evolutionary origins of a continental-scale floristic pattern, the enigmatic ‘Rand Flora’. The existence of disjunct distributions across many plant lineages between Macaronesia–northwest Africa, Horn of Africa–southern Arabia and east–south Africa has long intrigued botanists, but only now can we start analysing it within a statistical framework.
Phylogenetic and distributional data from 13 plant lineages exhibiting this disjunct distribution were analysed to estimate area carrying capacities and historical rates of biotic exchange between areas. The results indicate that there has been little exchange between southern Africa and the northern African region, and that this exchange occurred via east Africa. Northwest Africa–Macaronesia shows the smallest carrying capacity but highest dispersal rate with other regions, suggesting that its flora was built up by immigration of lineages, probably from the Mediterranean region and western Asia. In contrast, southern Africa shows the highest carrying capacity and lowest dispersal rate, suggesting a flora formed by in situ diversification.
We discuss further improvements of the method for addressing more complex hypotheses, such as asymmetric dispersal between regions or repeated cyclical events.
One of the best known examples of continental floristic disjunctions is the so-called ‘Rand Flora’ pattern: an Afro-Mediterranean phytogeographic pattern that evolutionarily relates floras of disjunct regions such as Macaronesia (the Azores, Madeira, Canary and Cape Verde archipelagos), northwest Africa, southern Arabia and western Asia, with the floras of east and south Africa (Le Brun 1971; Bramwell 1985). The most-cited example of this enigmatic floristic pattern is the Canary Islands ‘dragon tree’ (Dracaena draco) that presumably has its closest relatives in northeast Africa, the Socotra islands and southern Arabia, but many other plant groups have also been suggested as having a similar distribution pattern (Bramwell 1985; Andrus et al. 2004).
Two main hypotheses about the evolutionary origin for the Rand Flora pattern have been suggested: (i) the ‘vicariance hypothesis’: extant species are the remnants of an ancient widespread African flora that went partly extinct as a result of increasing aridification from the Miocene onwards, leaving relict biotas in climatic refuges at the east and west continental margins of Africa (Axelrod & Raven 1978; Bramwell 1985; Quezel 1979; Andrus et al. 2004; Thiv et al. 2010), or (ii) the ‘dispersal hypothesis’: the species' present disjunct distributions are the result of more recent long-distance dispersal events between geographically isolated areas, followed by in situ diversification (figure 1). For the dispersal hypothesis, two main directional routes have been proposed: (i) southwards, either from the Mediterranean region or from west-Asia via the Arabian Plate (e.g. Levyns 1964), or (ii) northwards from the southern African region via eastern Africa (e.g. Galley et al. 2007).
So far, biogeographic studies on the Rand Flora have been mainly descriptive, focusing on reconstructing biogeographic patterns in one or several groups (Bramwell 1985; Andrus et al. 2004). However, in the last years many new phylogenetic studies of plant taxa exhibiting this disjunction have been published, allowing us for the first time to conduct a meta-analysis to test general hypotheses within a statistical framework. At the same time, the development in the last years of new model-based methods of biogeographic inference (Ree & Sanmartín 2009) has made it possible to reconstruct the spatio-temporal evolution of individual lineages and regional biotas with increasing detail. One such method is the Bayesian approach to island biogeography (BIB) proposed by Sanmartín et al. (2008). It uses a Bayesian statistical approach that integrates phylogenetic and biogeographic uncertainty to estimate carrying capacities (equilibrium frequencies of species diversity) and rates of dispersal/biotic migration between geographically isolated areas (i.e. ‘islands’), using DNA sequence data and species distributions. An advantage of the method is that estimates of biogeographic parameters are integrated over the phylogeny of each group (phylogenetic relationships and divergence times), so the method can be used across multiple taxonomic groups differing in their age, evolutionary rate and/or dispersal capabilities. So far, it has only been used in an island context (i.e. areas separated by oceanic barriers), but Sanmartín et al. (2008) argued that the method could also be useful for scenarios in which areas are isolated by ecological barriers, such as high-mountain biotas separated by lowlands.
Here, we explore the usefulness of the BIB method for disentangling the origin of a complex continental floristic pattern such as the Rand Flora, where subtropical floras in the margins of Africa are now separated by tropical lowlands (west–east Africa) or by arid and semi-arid terrains such as the Saharan desert. We combine phylogenetic molecular data of multiple taxonomic groups exhibiting this disjunction into a large meta-analysis for inferring the rate of dispersal/historical migration between these now-isolated floras.
2. Material and methods
DNA sequence data from well-sampled molecular phylogenies including a broad representation of species within each lineage and area of distribution (either previously published or from our own ongoing research) were gathered for 13 groups exhibiting the disjunct Rand Flora distribution: Aeonium (Crassulaceae), Adenocarpus (Fabaceae), Androcymbium (Colchicum) (Colchicaceae), Convolvulus (Convolvulaceae), Monsonia (Geraniaceae), Moraea (Iridaceae), Sideroxylon (Sapotaceae), Ceropegieae (Apocynaceae), Geranium (subg. Robertium) (Geraniaceae) Solanum (subg. Leptostemonum p.p.) (Solanaceae), two subgroups of the large genera Euphorbia (subg. Rhizanthium and subg. Esula p.p) (Euphorbiaceae), as well as the family Campanulaceae. See the electronic supplementary material for details on distributions, phylogenetic studies and references.
Carrying capacities of the study areas and dispersal/migration rates between areas were estimated using the Bayesian island biogeography method of Sanmartín et al. (2008) implemented in MrBayes 4.0 (Ronquist et al. 2008, beta version, http://mrbayes.cvs.sourceforge.net/). As input for the analysis, we used a matrix of aligned sequences for each separate plant group, together with distributions for all included species (see the electronic supplementary material). Five distributional areas were defined: (i) Macaronesia–northwest Africa, (ii) eastern Africa and southern Arabia, (iii) southern Africa, and the ‘outside’ areas, (iv) Mediterranean region and (v) western Asia (see figure 1 and the electronic supplementary material for area definitions). We set up an individual general time reversible (GTR) molecular model for each group and a common GTR biogeographic model for the entire dataset, so that phylogeny and molecular parameters were estimated (independently) per group while biogeographic parameters were estimated across all groups. Differences in age, molecular evolutionary rate and dispersal rate among groups were accounted for by using group-specific molecular and dispersal rate scalers. For further details and specific settings, see the electronic supplementary material.
Posterior probabilities of parameters were surprisingly narrow (table 1, figure S3 in the electronic supplementary material), indicating good behaviour of the model despite the low number of potential dispersal events. Among regions, the lowest carrying capacity was estimated for the northwest Africa–Macaronesian region, whereas southern Africa showed the highest carrying capacity. The opposite pattern was found for dispersal rates, which were lowest for south Africa and highest for the northwest Africa–Macaronesian region, both with east Africa–southern Arabia and the Mediterranean (table 1 and figure 2). The results showed very little or no exchange between southern Africa and northwest Africa–Macaronesia. If any, this biotic exchange took place via the eastern African region, the only region that exhibits significant exchange with southern Africa (table 1 and figure 2).
The high dispersal rate observed between northwest Africa–Macaronesia and the eastern Africa–southern Arabia (table 1 and figure 2) gives some support to the vicariance hypothesis (figure 1a), which postulates the fragmentation of a continuous, subtropical flora in northern Africa that was divided by climatic events into eastern and western refuges (Axelrod & Raven 1978). However, rather than one vicariant event across multiple groups, this rate is likely to reflect repeated events of dispersal and vicariance between east and west Africa following the alternation of cycles of arid and humid periods in the Saharan Desert since the Late Miocene/Pliocene (Thiv et al. 2010).
The high dispersal rate between northern Africa and the eastern region stands in constrast with the low dispersal rate between this region and southern Africa (table 1 and figure 2). This can be interpreted as either a historical low rate of biotic exchange between southern and northern Africa, or, alternatively, that this exchange is too old to leave a signal in our phylogenetic data, that is, the south–north African disjunctions may be older than the northwest-eastern African disjunctions and therefore more likely to have been obscured or wiped-out by later extinction events. Indeed, there is some support for this hypothesis. The fossil record suggests that forests covering the whole of tropical/central Africa appeared in the Oligocene–Miocene, and geological data give the same time frame for mountain formation and the establishment of drier areas in northern Africa. In contrast, direct contact between the African continent and Eurasia was not established until the Miocene, when the Arabian Plate collided with the Eurasian Plate 16 Ma ago. If southern Africa was a part of a continuous Rand Flora, the region is likely to have been separated by climatological barriers from the northern areas earlier than the start of major biotic exchange between northern Africa and Eurasia.
Interestingly, northwest Africa plus Macaronesia show the smallest carrying capacity but present the highest dispersal rate (table 1), suggesting that the flora of this region was built up by immigration of new lineages. Migration from the Mediterranean region was probably the dominant route (figure 2), but dispersal from west Asia via the Arabian Plate and across the Saharan desert is another possible route supported by our data.
Conversely, the high carrying capacity of southern Africa (table 1) gives some support to the hypothesis that the highly diverse South African Cape flora has diverged in situ, facilitated by the relative climatological stability of the area from the Miocene onwards (Linder 2005). Similarly, the low dispersal rate with all other regions but eastern Africa agrees well with the idea that, following the formation of the eastern African mountains in the Pliocene, some south African lineages migrated to the north (via the Grand Rift and the Drakensberg mountains) and gave rise to the highly endemic eastern African mountain flora (Linder 2005; Galley et al. 2007).
The results from this analysis should be considered preliminary for several reasons: the dataset only represents a sample of groups showing this disjunction (Andrus et al. 2004); all phylogenies are not complete and some of the relevant disjunct taxa are missing (e.g. Monsonia, see the electronic supplementary material); finally, there is a potential underestimation of groups with southern African distributions (e.g. Euphorbia). Despite these drawbacks, our results suggest that the Bayesian island model may be useful in a continental setting, where the number of inferred dispersal/migration events between areas is low.
Although originally designed to study dispersal patterns in islands, the BIB method may also be used to detect possible vicariance events, by incorporating estimates of absolute divergence times. Vicariance predicts biogeographical and temporal congruence in disjunct distribution patterns across different plant groups, and that the disjunct distribution is at least as old as the geological barrier that caused it. By plotting the rate of dispersal over time, the appearance of a vicariance event can be detected: a decrease in dispersal frequency between two areas would suggest the formation of a new geographical barrier between them. BIB can even be used to estimate the timing of the paleogeographic barrier from the dated phylogenies, by comparing a one-rate model, in which the rate of biotic exchange is constant before and after the barrier, with a two-rate model in which there are two parameters, the rate of exchange before and after the barrier.
Future work should focus on testing asymmetric dispersal patterns, e.g. northward versus southward migration between southern Africa and northern–eastern Africa, as well as on incorporating absolute times to the inference, either directly as molecular divergence time estimates or indirectly through the use of fossil and paleogeographical information (Ree & Sanmartín 2009).
Two anonymous reviewers provided insightful comments. This research was funded by a postdoctoral fellowship from the Swedish Research Council to C.L.A., and grants from the Spanish Ministry of Technology and Science (project ‘Rand Flora’ CGL2009-13322-C03-01) and the CSIC to I.S.
↵† These authors contributed equally to the study.
- Received January 28, 2010.
- Accepted March 3, 2010.
- © 2010 The Royal Society