Cryptic adaptive radiation in tropical forest trees in New Caledonia

Authors


Author for correspondence:

Yohan Pillon

Tel: +1 808 932 7512

Email: pillon@hawaii.edu

Summary

  • The causes of the species richness of tropical trees are poorly understood, in particular the roles of ecological factors such as soil composition. The nickel(Ni)-hyperaccumulating tree genus Geissois (Cunoniaceae) from the South-west Pacific was chosen as a model of diversification on different substrates.
  • Here, we investigated the leaf element compositions, spatial distributions and phylogeny of all species of Geissois occurring on New Caledonia.
  • We found that New Caledonian Geissois descended from a single colonist and diversified relatively quickly into 13 species. Species on ultramafic and nonultramafic substrates showed contrasting patterns of leaf element composition and range overlap. Those on nonultramafic substrates were largely sympatric but had distinct leaf element compositions. By contrast, species on ultramafic substrates showed similar leaf element composition, but occurred in many cases exclusively in allopatry. Further, earlier work showed that at least three out of these seven species use different molecules to bind Ni.
  • Geissois qualifies as a cryptic adaptive radiation, and may be the first such example in a lineage of tropical forest trees. Variation in biochemical strategies for coping with both typical and adverse soil conditions may help to explain the diversification and coexistence of tropical forest trees on similar soil types.

Introduction

The great diversity of species in tropical rainforests is a widely known fact and enigma, especially the coexistence, side by side, of many species of trees that have the same basic requirements. The coexistence of morphologically similar congeneric species, in particular, is puzzling (Richards, 1969; Leigh et al., 2004). A growing number of studies indicate that substrates have a strong impact on the composition of tree communities and on the spatial distribution of individual plants within communities (John et al., 2007; Anderson et al., 2010; Schreeg et al., 2010; Katabuchi et al., 2012). Furthermore, soils have played a pivotal role in plant speciation in other vegetation types such as the fynbos and the succulent karoo of the Cape region (Schnitzler et al., 2011). Because most studies of rainforest plants have focused on communities, few have addressed niche partitioning among congeneric species. Fine et al. (2005) suggested that soils played a major role in the diversification of Protium in Amazonian forests; however, for sister species pairs that occur over large geographic scales (i.e. continents), it is difficult to exclude the possibility of earlier phases of allopatry in their divergence history. By contrast, when diversification has occurred on small oceanic islands, ecological speciation with gene flow is the most logical explanation (Savolainen et al., 2006; Papadopulos et al., 2011), demonstrating the value of such systems to test this hypothesis.

New Caledonia, an archipelago in the South-west Pacific is renowned for its rich, unique and threatened flora (Morat et al., 2012). Although it is geologically a continental island, having drifted from Australia, New Caledonia was entirely submerged from the end of the Cretaceous until 37 million yr ago (Myr; Cluzel et al., 1998; Pelletier, 2006), and all studies to date support the post-immersion origin of its biota (Grandcolas et al., 2008; Cruaud et al., 2012; Pillon, 2012). Thus, the history of the island's flora is comparable to that of true oceanic islands (e.g. volcanic islands) or ‘Darwinian’ islands (Grandcolas et al., 2008) sensu Gillespie & Roderick (2002). Nonetheless, the rich flora of New Caledonia is short of dramatic examples of adaptive radiation, particularly when compared to the Hawaiian or Macaronesian Islands (see examples in Givnish, 2010). Few radiations display the diversity of growth forms typically associated with ecological divergence. One notable exception is the genus Oxera (Lamiaceae), which contains a variety of life forms: monocaule shrubs, tangled shrubs and vines that range from slightly to very woody (Mabberley & de Kok, 2004). Furthermore, there are no convincing examples of radiations involving divergence in floral morphology associated with different pollination syndromes aside from Oxera, some orchids and genera engaged in specific plant–pollinator mutualisms such as Ficus (Cruaud et al., 2012) and Phyllanthus (including Glochidion, Kawakita & Kato, 2004). This paucity may be linked to the underrepresentation of lineages with zygomorphic flowers such as monocots, Asterales, Fabales and Lamiales in New Caledonia (Pillon et al., 2010).

The primary distinguishing feature of New Caledonia's terrestrial environment is the abundance of ultramafic substrates that cover one-third of the surface of the main island. Soils that develop on ultramafic rocks (including serpentine) have several characteristics that are limiting for plant growth, including: low availability of nitrogen (N), phosphorus (P) and potassium (K); a high magnesium : calcium (Mg : Ca) ratio; high concentrations of heavy metals (e.g. cobalt (Co), chromium (Cr), nickel (Ni)); and variable but often low water-holding capacity; in addition, fire seems to play an important role in vegetation dynamics (Jaffré, 1980; Proctor, 2003). Because of these edaphic characteristics, New Caledonia might be considered an Old Climatically Buffered Infertile Landscape (OCBIL; Hopper, 2009) or an Old Stable Landscape (OSL; Mucina & Wardell-Johnson, 2011), where elevated nutritional specialization is expected. Ultramafic substrates have been implicated as a cause of the disharmonic nature of the New Caledonian flora, for example the underrepresentation of monocots and Lamiales or the overrepresentation of Oxalidales and Malpighiales (Pillon et al., 2010). Clades that are well represented in New Caledonia have a high frequency of Ni-hyperaccumulating species (Jaffré et al., 2013), a trait associated with ultramafic substrates (Reeves & Baker, 2000). The ability to grow on ultramafic substrates, and possibly metal hyperaccumulation, may have been exaptations (Pillon et al., 2010); that is, these traits appeared in non-New Caledonian ancestors growing on normal soils and turned out to be advantageous when colonizing the island, which was probably entirely covered with ultramafic rocks when it first emerged (Pelletier, 2006). Other types of substrate found in New Caledonia include limestone and mostly old metamorphic or sedimentary rocks (including greywacke, schist and shale) that can be hard to characterize. The variety of New Caledonian soils is an important factor underlying the diversity of the flora (e.g. Jaffré, 1993). How plants have adapted to these different edaphic conditions is poorly known, but the few studies that have addressed this problem suggest the involvement of high trait lability, hybrid speciation and cryptic species, among other possibilities (de Kok, 2002; Pillon et al., 2009b,c).

In this study, we focus on the genus Geissois (Cunoniaceae), a lineage of trees that are known to hyperaccumulate Ni (Jaffré et al., 1979). We studied the leaf element composition and geographic distributions of 13 species in New Caledonia to examine associations among leaf element composition, soil type and degree of sympatry. Molecular phylogenetics and molecular clocks were also applied in order to test the single origin of the genus on the island, to estimate the speed of its diversification, and to determine whether New Caledonian Geissois qualifies as an adaptive radiation.

Materials and Methods

The genus Geissois

The genus Geissois Labill. comprises 19 species, all occurring in Melanesia (Hopkins et al., 2013): 13 species in New Caledonia, four in Fiji, one in Vanuatu, and one in the Temotu Province of the Solomon Islands (Fig. 1). Species of Geissois are tall shrubs to large trees primarily found in rainforests and forest edges, but also in secondary vegetation including savanna (generally dominated by Melaleuca quinquenervia) and maquis (scrubby vegetation on ultramafic rocks). The genus is characterized by its palmate, opposite leaves (3–11 leaflets), its conspicuous intrapetiolar stipules, and its relatively uniform flowers and inflorescences (Hopkins, 2008; Fogliani et al., 2009). The flowers are red and show some minor variation in size and pubescence. They are arranged in bottle brush-like racemes that are axillary or borne on branches, and project stiffly from the stem or are sometimes pendulous. The racemes are simple (or trident in G. hirsuta), and may be solitary or in clusters. They attract a wide range of nectarivorous birds (Barré et al., 2010) as well as bats, geckos and insects (Hopkins et al., in press). The fruits are capsules and the small, winged seeds are dispersed by wind. Most species are distinguished by leaflet and stipule characters, including the number, size and shape of the leaflets, the size and shape of the stipules, and type of pubescence, as well as the presence of domatia (Hopkins et al., in press); none of these characters has any obvious adaptive value (cf. Ashton, 1969).

Figure 1.

Distributions of the genus Geissois in the South-west Pacific and the closely related genera Karrabina and Pseudoweinmannia in Australia (both disjunct between northern Queensland and southern Queensland–northern New South Wales). Lamanonia, the fourth genus of the tribe Geissoieae, occurs in South America (not shown).

Leaf element composition

In order to take into account recent advances in Geissois systematics (Hopkins, 2006, 2007; Hopkins & Pillon, 2011), we updated the information in Jaffré et al. (1979) with new measurements. Herbarium specimens have been shown to be a useful source of samples for the analysis of leaf element composition (Brooks et al., 1977). Therefore, we obtained small leaf samples (c. 1 g) at the Herbier de la Nouvelle-Calédonie (NOU) at IRD Nouméa. We sampled up to five specimens per species whenever possible, and more for species found on a range of soil types (Supporting Information Table S1). No samples were taken from herbarium sheets when doing so would have affected the quality of the information provided by that sheet, and this constrained our choice of samples for some species. Some species are naturally rare and represented by only a few specimens, for example the critically endangered G. belema and G. bradfordii, and therefore fewer samples were included for such species. Otherwise, we tried to maximize the geographical coverage of each species' range. The samples were ground to a powder and 0.25–0.3 g quantities were dry-ashed at 500°C for 5.5 h and re-suspended in 25 ml of 1 M HCl before being run on a Varian Vista MPX ICP-OES Spectrometer (Varian Inc., Palo Alto, CA, USA) at the UH Hilo Analytical Lab. We measured the concentrations of seven elements – aluminium (Al), Ca, Cr, iron (Fe), Mg, Ni and P – in each leaf sample. Samples with a value of Fe above 500 ppm were conservatively excluded, because of possible soil contamination (Reeves & Baker, 2000 recommended a threshold of 1000 ppm). Data were then log-transformed and standardized. We performed Non-Metric Multidimensional Scaling (NMDS) in PAST (Hammer et al., 2001) to obtain a graphical representation of the variation within and among species in leaf elemental content. We used Hotelling's pairwise comparison in PAST to test for differences in leaf element composition among species.

Species distributions

Species distributions were established from a list of specimens of Geissois (Hopkins et al., in press) largely based on the collection of the NOU herbarium. A convex polygon was drawn to represent the distribution of each species, and its surface area was calculated; overlap between those polygons and the ocean was nonexistent to negligible. We also calculated the percentage of geographic range overlap for each pair of species by dividing the area of overlap by the area of the species with the smaller range. Using the range of the more restricted species yielded the maximum estimate of sympatry, and thus the most conservative estimate of allopatry, in each case.

Phylogenetic analyses

For phylogenetic reconstruction, we supplemented and re-analysed data from Pillon (2011). Because of the persistence of multiple ITS copies in many Cunoniaceae, and the low level of variation of plastid genes, we sequenced a single plastid locus, trnL-F, and two nuclear genes, ncpGS and PHYC, using the protocols described in Pillon et al. (2009a). Sequences were available for all New Caledonia species known at that time plus Geisssois denhamii from Vanuatu. We added to this dataset trnL and PHYC sequences for the recently described G. belema and the Fijian G. ternata, and a trnL sequence of the Fijian G. superba from Genbank (Table S2). Best-fit substitution models for trnL, PHYC and ncpGS were HKY, HKY + γ and HKY + γ (respectively) according to JModeltest analyses (Posada, 2008). No evidence for recombination was found in either the PHYC or the ncpGS datasets using GARD (Kosakovsky Pond et al., 2006), assuming no site-to-site variation (most sensitive setting). For each dataset, we ran 2 million generations of Markov Chain Monte Carlo (MCMC) in MrBayes (Ronquist & Huelsenbeck, 2003), sampling the tree every 1000 generations, using the appropriate substitution models and a 500 000-generation burnin; convergence was checked in Tracer (Rambaut & Drummond, 2007). In order to combine the PHYC and ncpGS datasets, where some species are represented by multiple accessions or heterozygotes, we used *BEAST (Heled & Drummond, 2010). We twice ran a 500-million-generation MCMC, with appropriate substitution models, a Yule process for tree reconstruction starting with a UPGMA tree, and a strict molecular clock, sampling one tree every 500 000 generations and with a burnin of 100 million generations.

In order to estimate a date for the origin of the radiation of New Caledonian Geissois, we applied a molecular clock analysis on a reduced combined dataset of ncpGS and PHYC. In spite of the availability of an almost complete generic sampling for plastid genes (Bradford & Barnes, 2001; Hopkins et al., 2013) and the rich fossil record of Cunoniaceae in Australia (Barnes et al., 2001), the absence of variation in plastid genes within New Caledonian Geissois (Fig. S1), as in other New Caledonian Cunoniaceae genera (Y. Pillon, pers. obs.), precluded the dating of the crown radiation. To simplify the analyses, we limited the sampling within Geissois by avoiding heterozygote accessions but making sure to sample the deepest split within New Caledonian Geissois, as well as two other genera of New Caledonian Cunoniaceae: Codia and Spiraeanthemum (group brongniartianum, Pillon et al., 2009a). We ran two analyses of 100-million-generation MCMC in BEAST v1.7.5 (Drummond et al., 2012), applying appropriate substitution models, a lognormal clock, and a Yule tree prior using an input ultrametric tree obtained from r8s (Sanderson, 2003), sampling a tree every 100 000 generations and with a burnin of 10 million generations. Prior age at the root (divergence Cunoniaceae–Brunelliaceae) followed a normal distribution with a mean of 86 Myr and a standard deviation of 2 Myr. This age was obtained by Heibl & Renner (2012) for the divergence between Cunoniaceae and its putative sister group (Brunelliaceae, Cephalotaceae, Elaeocarpaceae). A similar age of 83 Myr was obtained independently for the same node by Xi et al. (2012). All BEAST analyses were run on Lifeportal (University of Oslo), and convergence of the runs was checked in Tracer. Speciation rates were calculated using eqn 4 of Magallón & Sanderson (2001).

Results

Low concentrations of Fe were observed in all leaf samples of Geissois (maximum 230 ppm) with the exception of one sample of G. pruinosa var pruinosa (MacKee 45227, 924 ppm), which was excluded from subsequent analyses because of suspicion of contamination with soil. All species of Geissois naturally present on ultramafic substrates were found to be Ni hyperaccumulators (Ni content > 1000 ppm), with the notable exception of the recently described G. belema (Fig. S2). All species absent from ultramafic substrates had low Ni content, suggesting that the high concentrations reported previously in G. montana and G. racemosa by Jaffré et al. (1979) were likely to result from confusion at a time where the taxonomy of the genus was not well understood. In fact, for G. hippocastanifolia, G. polyphylla and G. montana, Ni was undetectable in four out of five samples. Only one individual of G. balansae, from near Ponerihouen (Haudricourt 817, vague locality, no ecological information) had a relatively high Ni concentration (571 ppm). This finding suggests that this plant was growing on or in the vicinity of ultramafic substrate, although the species is not reported from this habitat. High Ni content was observed in all accessions of G. pruinosa var. intermedia, a taxon that is reported only in the north-east, usually on nonultramafic substrate. Lastly, we report here for the first time evidence that G. polyphylla is an Al hyperaccumulator; four of five individuals sampled had values ≥ 1000 ppm, and no other species of Geissois reached this threshold (see Table S1).

Non-Metric Multidimensional Scaling (NMDS) on all elements analysed allowed a relatively clear separation between species found on ultramafic substrates and those found on nonultramafic ones (Fig. 2). Overlaps were due to: G. hirsuta, which seems to be the only species unambiguously present on both soil types, the one sample of G. balansae with an unusually high Ni content, and G. belema, which grows on ultramafic substrates but does not hyperaccumulate Ni. Each of the five species restricted to nonultramafic substrate had a largely distinct elemental profile, with only slight overlap between G. polyphylla and G. hippocastanifolia, and a larger overlap between G. montana and G. balansae. Although all of these species had similar Ni concentrations in their leaves, they varied considerably in the concentrations of other elements such as Al, Ca and Mg. Species from Vanuatu, Fiji and the Solomon Islands – all of which occur on nonultramafic substrates – had elemental profiles that fell within the range of G. balansae (Fig. S3). The geographic ranges of species from nonultramafic substrates in New Caledonia were highly overlapping (Figs 3, S4; Table S3). The only exception was the species pair, G. montana–G. balansae, the first of which occurred northward from Mt Aoupinié and the second southward from Mt Aoupinié. In contrast to the near-unique elemental profiles of nonultramafic species, those restricted to ultramafic substrate showed highly overlapping leaf element compositions; no species had a distinct profile (Fig. 2), even if the widespread and ecologically variable G. pruinosa was excluded. Also in contrast to nonultramafic species, the geographic distributions of taxa on ultramafic substrate tended to be more disjunct (Figs 3, S4; Table S3). For example, G. bradfordii and G. velutina are restricted to the far south, G. magnifica to the east coast, G. lanceolata to the west coast, and G. belema to the northern island of Art.

Figure 2.

Non-Metric Multidimensional Scaling (NMDS) of leaf element composition of New Caledonian Geissois species. ba, Geissois balansae; be, G. belema; br, G. bradfordii; hip, G. hippocastanifolia; hir, G. hirsuta; l, G. lanceolata; ma, G. magnifica; mo, G. montana; po, G. polyphylla; pr, G. pruinosa (all varieties); t, G. trifoliolata; ra, G. racemosa; v, G. velutina.

Figure 3.

Example of distributions of selected Geissois species in New Caledonia, showing broad sympatry among species on nonultramafic substrates (a), and allopatry of species on ultramafic substrates (b). Areas in red are ultramafic rocks.

Little variation was found in the trnL dataset (Fig. S1) with all the New Caledonian species having identical DNA sequences except for G. hippocastanifolia, which differed by a single base; the two Fijian species shared one substitution. The PHYC tree (Fig. S5) was also poorly resolved with only two groups supported within Geissois: G. hippocastanifolia + G. polyphylla and G. denhamii (Vanuatu) + G. ternata (Fiji). In the ncpGS tree (Fig. S6), G. hippocastanifolia and G. polyphylla formed a clade that is sister to the rest of the species of New Caledonia plus G. denhamii (Vanuatu). Several species had multiple distant alleles, and some represented by multiple accessions were not monophyletic. The combined PHYC-npcGS analysis in *BEAST (Fig. 4) also recovered G. hippocastanifolia and G. polyphylla as sister species, forming a clade that was sister to the rest of the genus (G. racemosa group), including all other New Caledonian species and G. denhamii (Vanuatu). Internal supports within the G. racemosa group were low. The molecular clock analysis on a combined PHYC-npcGS dataset (Fig. S7) indicated an age of 7.3 Myr (3.5–12.8 Myr) for the crown of Geissois (Table 1).

Table 1. Crown age for some Cunoniaceae radiations in New Caledonia
 Species numberCrown age median [95% confidence interval] (Myr)Diversification rate (sp. Myr−1)Diversification rate (sp. Myr−1 km−2)
  1. Diversification rate per km2 is calculated for the main island only (16 372 km2); Geissois belema, which is endemic to Art Island, was therefore excluded from the calculation of diversification rate per unit area.

New Caledonian Geissois137.3 [3.5–12.8]0.261.5 × 10−5
Codia 145.3 [2.5–9.4]0.372.2 × 10−5
Spiraeanthemum group brongniartianum66.3 [2.3–12.1]0.171.1 × 10−5
Figure 4.

Half compatible consensus tree of Geissois obtained from the combined PHYC-ncpGS analysis in *BEAST. Numbers at nodes indicate posterior probabilities above 0.5. var. int., var. intermedia. Substrate preferences are indicated: UM, ultramafic; NUM, nonultramafic; UM-NUM, both types. Hyperaccumulation of aluminium or nickel is indicated by Al or Ni, respectively. Geissois belema is not included here because only the PHYC sequence was available.

Discussion

New Caledonia is well known for its diverse flora; however, no dramatic cases of adaptive radiation have been documented thus far. We show here that Geissois satisfies the criteria of an adaptive radiation (Schluter, 2000): common ancestry, rapid speciation, correlation between phenotype and environment, and trait utility. The difference between New Caledonian Geissois and classic examples of adaptive radiation is the lack of apparent niche separation in Geissois, which occurs on just two broad soil types. Radiation of this genus within the small land area of New Caledonia's main island appears to have involved divergence of metal uptake mechanisms on ultramafic substrates and character displacement, or ultra-fine partitioning of edaphic niches, by species on nonultramafic substrates.

Common ancestry

Molecular phylogenetic analyses provide support for the monophyly of Geissois; all New Caledonian species belong to a single clade and are likely to be descended from a single colonist. Subsequently, it seems that a unique dispersal event towards neighbouring islands gave rise to the endemic species of Vanuatu and Fiji. Geissois polyphylla and G. hippocastanifolia are recovered as sister species, and are in turn sister to the Geissois racemosa group (all other New Caledonian species, including the morphologically divergent G. hirsuta and extra-New-Caledonian species). All species growing on ultramafic substrates belong to the G. racemosa group. This clade was poorly resolved, and several species were not recovered as monophyletic in the ncpGS analysis. The lack of cohesion of the individual species in this group suggests either hybridization among species or retention of ancestral polymorphisms, both of which are consistent with recent divergence.

Rapid speciation

The diversification of Geissois occurred at a rate of 0.26 sp. Myr−1 (Table 1). This rate is similar to that of other New Caledonian plant radiations (Pillon, 2012), which may be considered modest among known plant radiations. Nevertheless, when taking into account the small surface area of the island, Geissois had a diversification rate per unit area of 1.5 × 10−5 sp. Myr−1 km−2. This is comparable to rates reported for rapid island radiations such as Hawaiian Bidens and Macaronesian Echium, and higher than the rates reported for large continental radiations such as Andean Lupinus and Eurasian Dianthus (Knope et al., 2012). This figure is especially significant considering that the archipelago is dominated by one large island and is therefore far less fragmented and hence less suited for allopatric speciation than most archipelagos.

Correlation between phenotype and environment

Although the variation in vegetative traits that characterize the different species of Geissois in New Caledonia shows no clear relationship with ecology, the variation in leaf element composition among species is more readily associated with the environment. All Geissois species growing on ultramafic substrates except G. belema are Ni hyperaccumulators; species absent from nonultramafic substrates have a higher Al content, and G. polyphylla qualifies as an Al hyperaccumulator. As a group, species from ultramafic substrates were elementally distinct from those on nonultramafic ones when considering all elements together. The contrast between these groups is not likely to be due solely to plasticity given that some species occurring sympatrically on ultramafic substrates – and thus likely to be experiencing equal Ni availability – displayed different Ni concentrations, for example G. velutina and G. pruinosa var. pruinosa. In addition, previous glasshouse experiments found that the nonultramafic G. montana is sensitive to watering with a Ni solution whereas the ultramafic taxon G. pruinosa var. pruinosa is resistant (L. Richard et al., pers. obs.), indicating a genetic basis for tolerance to this heavy metal. Although Ni tolerance and hyperaccumulation are different processes, the former is a pre-requisite for the latter, in natural settings at least. Understanding the mechanisms that underlie the variation among Geissois species in the ability to hyperaccumulate Ni will nevertheless require testing under controlled conditions.

Trait utility

Little is known about the evolutionary significance of Ni hyperaccumulation in Geissois. In other plant groups, however, Ni hyperaccumulation is suggested to confer resistance to herbivory (Boyd, 2007), pathogens (Hörger et al., 2013) or drought (Bhatia et al., 2005). There is less information on the significance of Al hyperaccumulation (Jansen et al., 2002), although it has been suggested as a mechanism to deter herbivory. This idea has not been well tested, however, and was not supported in a field study in the Philippines including six Al hyperaccumulator and one Ni hyperaccumulators (Proctor et al., 2000). In the Al hyperaccumulator, Melastoma malabathricum, an increase in oxalate concentration in the rhizosphere enabled solubilization of aluminium phosphate, and an increase in plant P was associated with an increase in plant Al (Watanabe & Osaki, 2002). In the current study, the single Al hyperaccumulator species, G. polyphylla, also had the highest P content. It is possible, therefore, that this species has developed a similar mechanism to increase P uptake. Similarly Ni hyperaccumulation could also be a byproduct of a different physiological pathway rather than an end in itself.

Diversification on nonultramafic substrates

Our data suggest character displacement within Geissois, at least on nonultramafic substrates, where largely sympatric species exhibit different elemental signatures (Fig. 2; Table S4). The geology of New Caledonia is fairly complex, and nonultramafic rocks of the main island include old and often poorly delimited geological layers. As such, we cannot exclude the possibility that different species of Geissois with overlapping distributions have preferences for different soils (microhabitats), which could explain the variation in leaf element composition among them. However, in areas such as Col d'Amieu, G. balansae, G. polyphylla and G. racemosa can grow within a few metres of each other. Furthermore, G. hippocastanifolia and G. montana are often sympatric (Mandjélia, Aoupinié) and can hybridize as suggested by the intermediate morphology of one specimen (Pillon 81, Mandjélia). It is possible that these species differentially exploit available soil nutrients (cf. Richards, 1969), and this may explain their coexistence. Testing this hypothesis would require analysis of individuals growing in controlled conditions. Interestingly the only two nonultramafic species that have somewhat similar leaf element compositions, G. montana and G. balansae, are largely allopatric. Furthermore, even though G. hirsuta has a broad ecology (occurring on both ultramafic and nonultramafic substrates), its elemental signature is different from those of at least G. hippocastanifolia, G. montana and G. polyphylla (Fig. 2; Table S5). The fact that the allopatric G. denhamii (Vanuatu), G. ternata (Fiji) and G. pentaphylla (Solomon Islands) have relatively similar elemental signatures, that are in turn similar to the New Caledonian G. balansae, suggests that the variation in elemental signatures observed among species on nonultramafic substrates in New Caledonia is not the result of random processes but rather of character displacement to minimize competition. The phylogenetic position of G. polyphylla indicates that Al hyperaccumulation is a derived character that evolved in New Caledonia, as it is not found in other species of the Geissois nor in the closely related Pseudoweinmannia and Lamanonia (Nogueira & Haridasan, 1997). The character is, however, found in several less closely related genera of Cunoniaceae in Australia (Jansen et al., 2002).

Diversification on ultramafic substrates

Species growing on ultramafic substrates had largely overlapping elemental signatures, including high Ni content; however, the mechanisms by which they attain these signatures may be different. In a study of the Ni-binding ligand in several hyperaccumulating taxa, Callahan et al. (2012) found that G. bradfordii, G. hirsuta and G. pruinosa var. pruinosa had distinct metabolite profiles. Although in all species Ni was bound to nicotianamine, this metal was also bound to serine in G. bradfordii, to citric acid in G. pruinosa var. pruinosa, and to citric acid and galacturonic acid in G. hirsuta. Therefore, despite having similar concentrations of Ni in their tissues, these species appear to possess different mechanisms for Ni storage. Surprisingly, Geissois belema does not seem to accumulate Ni; this might be explained by the fact that it grows on highly weathered ferralitic soils where Ni is not available in significant amounts (Jaffré & L'Huillier, 2010).

In contrast to nonultramafic substrates that are almost continuous on the main island, ultramafic substrates represent a highly fragmented environment. Their patchy distribution across the island is likely to have favored allopatric speciation and may explain the weaker ecological differentiation observed in some cases, through reduced interspecific competition. Of all 13 species in New Caledonia, G. bradfordii is the only one that seems to have an obviously narrow ecology, being restricted to the bank of a river in an ultramafic area. This rupicolous species has narrow leaflets as expected in plants with such ecology (van Steenis, 1981). Interestingly, it occurs in the south of the island, as do some other rupicolous Cunoniaceae in the genera Spiraeanthemum, Cunonia and Pancheria (Hopkins et al., in press).

Widespread species

Only two species of Geissois are broadly distributed and documented on both ultramafic and nonultramafic substrates. Geissois pruinosa is a species complex most commonly found on ultramafic substrates. Although several records suggested the presence of var. intermedia on nonultramafic rocks in the north-east, all of the samples analysed in the current study had a high Ni content, suggesting that they were collected on ultramafic substrates. Serpentine rocks are patchily distributed in the north-east, and it may be that G. pruinosa always has access to Ni and may be more strictly associated with ultramafic substrates than previously thought. Further fieldwork is required to re-evaluate the ecology of this species. Geissois hirsuta is therefore the only species unambiguously present on both types of soil, and it is both morphologically distinct from other species and uniform throughout its range. Thus, the presence of cryptic species is not indicated (cf. Spiraeanthemum, Pillon et al., 2009b). Rather, it could represent an interesting model for studies of the genetic basis of adaptation to ultramafic conditions. Aside from these two widespread species, each species of Geissois possesses a unique combination of ecology, elemental signature and geographic distribution (Table 2). Therefore, only the mechanism by which these two species remain distinct remains unexplained by this study, although pollinators may play a role (see later).

Table 2. Pairwise ecological, elemental and geographical differences between species of New Caledonia Geissois
  balansae belema bradfordii hippocastanifolia hirsuta lanceolata magnifica montana polyphylla pruinosa racemosa trifoliolata velutina
  1. S, no substrate overlap (ultramafic vs nonultramafic substrate); E, no elemental overlap (no overlaps between polygons in NMDS analysis of leaf element composition); e, minor elemental overlap (for the species pair G. hippocastanifolia/G. polyphylla); G, no geographical overlap of species distributions; g, geographical overlap of < 5%. pruinosa includes both varieties (pruinosa, intermedia).

balansae              
belema S/–/G            
bradfordii S/E/G–/E/G           
hippocastanifolia –/E/–S/E/GS/E/G          
hirsuta –/–/––/–/G–/E/––/E/–         
lanceolata S/E/––/E/G–/E/GS/E/G–/–/–        
magnifica S/E/G–/E/G–/E/GS/E/g–/–/––/–/G       
montana –/–/GS/E/GS/E/G–/E/––/E/–S/E/G–/E/G      
polyphylla –/E/–S/E/GS/E/G–/e/––/E/–S/E/gS/E/G–/E/–     
pruinosa –/–/––/E/G–/–/––/E/––/–/––/–/––/–/––/E/––/E/–    
racemosa –/E/–S/E/GS/E/G–/E/––/–/––/E/–S/E/––/E/––/E/––/E/   
trifoliolata –/E/––/E/G–/E/G–/E/G–/–/––/E/G–/–/G–/E/G–/E/––/–/––/E/–  
velutina S/–/g–/E/G–/E/–S/E/G–/–/––/E/G–/–/GS/E/GS/E/G–/–/–S/E/G–/E/G 

Reproductive isolation

It is not clear how reproductive barriers evolved within Geissois to isolate so many species within such a small geographic area, but the variation observed in soils, leaf element composition and mechanism of metal accumulation may be associated with reduced fitness of interspecific hybrids (i.e. postzygotic barriers). Postzygotic barriers may then promote the evolution of prezygotic barriers through reinforcement (Widmer et al., 2009), which would be consistent with the differential flowering peaks among some Geissois species observed in the field but undetected in the herbarium (Hopkins et al., in press). Variation in flowering time may also be linked directly to differences in substrate (Savolainen et al., 2006), particularly if associated with water availability (Borchert, 1994). With a single possible exception, there is no evidence of diversification in Geissois through adaptation to different pollinators. The uniquely widespread nonedaphic specialist G. hirsuta has distinct pendant, trident-shaped inflorescences that may attract a distinct suite of pollinators. Pollinator isolation might perhaps maintain the distinctiveness of this species in spite of its ecological and geographical overlap with so many other species of the genus. Further work is needed to test these ideas.

In conclusion, the radiation of Geissois on New Caledonia was the result of the rapid diversification from a single progenitor. Speciation was facilitated by the sharp contrast between nonultramafic and ultramafic substrates where Ni-hyperaccumulating species evolved. On nonultramafic substrates, species display broad geographic overlap but generally have distinct elemental signatures, suggesting specialization to different microhabitats within this variable soil type or, more likely, different preferences for soil nutrients. By contrast, on ultramafic substrates, allopatry may have been more important for diversification. However, even though species displayed similar elemental signatures, preliminary biochemical work suggests that different mechanisms evolved within Geissois to deal with ultramafic conditions. New Caledonia Geissois may be the first confirmed case of adaptive radiation in trees. There are only a modest number of confirmed cases of adaptive radiation in plants, and most involve herbs or plants that were ancestrally herbs (Givnish, 2010). Few studies have been able to determine how congeneric tree species partition the ecological space within a given geographical area (Cavender-Bares et al., 2004). Our study suggests that a large part of the diversity within a lineage of tropical forest trees may be hidden in their less-visible interactions with different soils and their different mechanisms of nutrient and metal uptake. These results need to be confirmed in other plant groups and in other geographical areas where contrasts in soil conditions may not be as sharp as in New Caledonia.

Acknowledgements

We wish to thank Laure Barrabé for her assistance at many stages of this work. We thank Jason Bradford, Eve Lucas, Jérôme Munzinger and the DNA banks at Royal Botanic Gardens, Kew and Missouri Botanical Garden for providing leaf material/DNA samples. Material from Vanuatu was collected during the joint IRD/MNHN/Pro-Natura expedition Santo 2006. We thank Jacqueline Fambart-Tinel and the staff of the herbarium at NOU for assistance in the field and for allowing sampling from specimens. We thank Tomoko Sakishima and Jodie Schulten for assistance in the lab, as well as Anne Veillet (Genetics Core Facility, UH Hilo) and Lucas Mead (Analytical Laboratory, UH Hilo) for processing sequencing and chemical analyses. Finally, we thank Becky Ostertag for helpful discussions and three anonymous reviewers for their comments on an earlier version of this manuscript. Funding for chemical analyses was provided by NSF DEB 0954274.

Ancillary