“Explosive” radiations featuring rapid opportunistic morphological and ecological diversification are phenomena previously reported for some islands (e.g.  and references therein). Extreme ancestral bottlenecks, together with on-going hybridization and incomplete lineage sorting, can prevent phylogenetic reconstruction in cases of island radiations if they have been recent and produced many species . However, a good understanding of phylogenetic relationships within radiating groups is key for further evolutionary studies into mechanisms and whether change is adaptive, due to drift in small populations or other phenomena .
For the endemic New Caledonian Diospyros species, previous studies, based on multiple plastid  and low-copy nuclear  markers, showed 21 species to be closely related (Figure 1) and were not able to clearly resolve phylogenetic relationships among them. In the combined data set (plastid and nuclear markers; ) only seven of the 21 species included were found to form highly supported groups of accessions from single species. Individuals belonging to each of the remaining 14 species failed to cluster according to their taxonomic circumscription. Dating analysis based on plastid and low-copy nuclear markers showed that the common ancestor of this clade of endemic New Caledonian Diospyros species has arrived in New Caledonia around nine million years ago . Diospyros vieillardii has been shown to be sister to the rest of this endemic clade and separated from the rest of the species around 7.2 million years ago.
Results of the current study using genome-wide AFLP markers reveal that most species form unique groups paralleling recognised species. Around one-third (eight species, NJ dendrogram, Figure 2A) and one-half (11 species, Bayesian tree, Figure 2B) of the species, are genetically distinct with high support (Figure 2). However, the overall AFLP results prove unable to clearly resolve the backbone of trees, similar to previous results obtained from analyses of DNA sequence data . Intra-specific variation was greater (~80%) than that found at inter-specific level (~20%). This low ratio of among- versus within-species divergence in the context of considerable morphological and ecological divergence is indicative of a recent diversification . Such a process can explain why we were able to get clear species boundaries for most species but were unable to clearly resolve phylogenetic relationships between them.
Two species that did not form well-defined clades (D. minimifolia and D. parviflora) were previously considered by White  to show variability in leaf morphology that may indicate that they are in fact a collection of several species. For D. minimifolia White  mentioned that the type population (close to population 15 of this study) has smaller leaves compared to other populations of this species. In our results this population clusters together with the majority of the D. minimifolia accessions; the population that is separated from the rest (population 16) is from Gaji. According to White D. parviflora is a wide-spread species, showing considerable variability of leaf morphology even within populations, making it impossible to differentiate these into different species. Our results show all accessions of D. parviflora, except those from Plateau de Tango (population 24), to form a group. All included accessions from D. parviflora are from ultramafic localities.
To our surprise, the AFLP results do not show any significant grouping according to ecological (edaphic, climatic, elevational), geographical or morphological factors (Additional file 3). The two weakly differentiated groups revealed by Structure and PCO also do not correspond to any conspicuous phenotypic characteristics. The allele-frequency divergence between the two groups found by Structure is low, which explains why we did not observe the two groups in the Bayesian and NJ tree-building results. Taken together, these results indicate that positive selection has perhaps acted on few genomic regions  and has resulted in phenotypic diversification of New Caledonian Diospyros. Variation in copy number of specific genomic regions may be an additional aspect of molecular variation that, although invisible to AFLP markers, could form the basis of adaptation to different environmental conditions .
The individuals of D. vieillardii, D. umbrosa and D. flavocarpa form a minimally isolated group (squares in the grey group) in the PCO (Figure 3). Previous phylogenetic analyses (Figure 1) showed these three species to be sister to the rest of the taxa. Due to its morphological and ecological features D. sp. Pic N’ga from Île des Pins could be a hybrid between D. calciphila and D. vieillardii, but D. vieillardii is now not known from this island. In PCO, individuals of this putative species are located between individuals of D. calciphila and D. vieillardii (Figure 3). The split between the two groups observed (Figures 3 and 4) could be relatively old, separating two lineages that developed in isolated regions. For instance, dry periods of the Pleistocene caused aridification in many areas, and some vegetation types persisted only in local refugia e.g. [34–36]. After climatic conditions became more favourable, the two groups probably expanded rapidly into newly suitable habitats where they overlapped; the time scale of these fluctuations (ca. 0.02 – 0.1 myr; ) was probably not enough to allow woody species with long generation time such as Diospyros to diverge and become permanently reproductively isolated . There are a few admixed individuals in the Structure analysis (Figure 4), which implies that hybridization might have played a role in evolution of this group.
Accelerated rates of evolution at few genes as a result of positive selection could have resulted in the morphological and ecological diversification apparent today in this group of New Caledonian Diospyros species. Furthermore, in addition to retention of ancestral polymorphisms, frequent gene flow could have acted against genome-wide genetic differentiation between the species. Barriers to gene flow between these species may be highly porous, with only few genes responsible for ecological and morphological adaptations evolving on distinct trajectories under strong selection, which leaves the rest of their genomes open to gene flow . Finding these few genes with AFLP is realistically improbable because they are a miniscule component in comparison the rest of these genomes. In the case of a recent and rapid radiation in plants, it could be argued that the bulk of regions sampled by AFLP have not evolved quickly enough to accumulate substitutions that could indicate species relationships. Our results are similar to those found in various other island genera (e.g. Araucaria in New Caledonia, ; Ourisia in New Zealand, ).
Diospyros vieillardii, which is sister to the rest of the taxa belonging to this group of New Caledonian endemics [12, 13], is confined to ultramafic soils, which supports the hypothesis of this being an exaptation of the progenitor of this New Caledonian Diospyros clade to ultramafic soils when the whole island was still covered by heavy-metal-rich substrates; similar findings have been made in other plant groups in New Caledonia e.g. . Later, erosion reduced the extent of this geological layer to one third of the island , and existing species began to move onto other substrates where they subsequently diverged, forming distinct species. Such observations have been made in various other New Caledonian groups (e.g. Araucaria, ; Spiraeanthemum, ; Codia, ). A few studies have examined the adaptive basis and processes involved in rapid radiations in New Caledonia e.g.  and Hawai’i (e.g. lobeliads, ; silverswords, ). Linking ecological parameters and/or phenotypic traits associated with speciation has to be done with caution because range alterations, subsequent evolution, and species extinctions might have erased initial signals found in only a few genes. Therefore, the associations observed today may be misleading, and the specific conditions/traits that were indeed linked to speciation, if any, may no longer be present .
Further work involving common garden experiments would provide insights into the effect of environmental conditions on morphological traits and therefore plasticity of genomes of the New Caledonian Diospyros species. Unfortunately, such experiments are time and cost intensive. It is difficult to obtain ripe fruits of all Diospyros species, and in addition it is difficult to germinate and grow them, which is a crucial aspect of conducting such experiments. Reciprocal transplantation of seedlings across environments are of course more easily conducted than common garden experiments, but they are still time consuming and costly; in addition species adapted to one soil type often will not survive when transplanted to other soil types.