Imbalance and shifts in diversification rate can be used to study evolutionary processes. The radiation of a clade, for instance, results in an imbalanced tree, or the branch leading towards a radiating clade exhibits a higher diversification rate in comparison with a clade that has not radiated (Sanderson & Donoghue, 1994). After the identification of radiating clades, (a)biotic explanations can be sought for the observed patterns in (im)balance (Mooers & Heard, 1997).
Radiations in Annonaceae
Significant imbalance was revealed in net diversification rates among Annonaceae lineages by the topological method, and this imbalance was mainly located in Malmeoideae (Table 1) along the branches leading to Clade A, Clade B and the miliusoid clade (Fig. 1). No significant among-lineage diversification rate variation within Annonoideae was detected, except for the branch at the base of Clade A (Fig. 1). This single rate shift was insufficient to cause rejection of the null hypothesis that Annonoideae as a whole diversified under a stochastic branching model. In general, this is true for most of the speciation in Annonaceae. This supports a recent study showing that the diversification of Annonaceae was not severely affected by the events of the Cretaceous–Palaeogene boundary, and that Malmeoideae and Annonoideae have a constant rate of diversification (Couvreur et al., 2011b). However, that study indicated no major shifts of diversification throughout the evolutionary history of the family, whereas here three such shifts were detected (Fig. 1).
Annonoideae and Malmeoideae together comprise the majority of the genera and species in Annonaceae. Annonoideae and Malmeoideae both consist of approximately 50 genera, but the number of species is different (c. 700 vs. c. 1600 species for Malmeoideae and Annonoideae, respectively; Couvreur et al., 2011b). In addition, Annonoideae includes the 10 largest genera of the family, comprising c. 1000 of its 1600 species. The cause for this Annonoideae–Malmeoideae difference is not understood. The topological method did not indicate any significant imbalance between Annonoideae and Malmeoideae. Furthermore, the temporal method indicated that the difference in species richness was not attributable to an increase in diversification rate along the stem lineages of these clades. However, a recently reported younger age for the crown node of Malmeoideae (based on Bayesian dating) might alter this picture (Erkens et al., 2009; Su & Saunders, 2009), although this should be studied further (Couvreur et al., 2011b; Pirie & Doyle, 2012). The temporal method uncovered a small increase in diversification rate at the base of Annonoideae and a small decrease at the base of Malmeoideae, but neither change was significant. The distribution of significant log N rate shifts between these clades differed. Four of the five largest increases in diversification rate occurred in Annonoideae and only one in Malmeoideae (and the latter in a poorly supported part of the tree, rendering this result questionable). However, these four radiations did not produce the bulk of the species present in Annonoideae, a result similar to that of Couvreur et al. (2011b). Of the 10 largest genera in the family (Annona, Artabotrys R.Br., Duguetia A.St.-Hil., Fissistigma Griff., Friesodielsia Steenis, Guatteria, Goniothalamus, Monanthotaxis Baill., Uvaria L. and Xylopia L.), only Goniothalamus constituted a radiation of species under the methods used for this study. Such a result was also supported by the study of Couvreur et al. (2011b). Furthermore, Guatteria, the largest genus in the family (Erkens et al., 2007a; Erkens, Westra & Maas, 2008) did not (although only marginally nonsignificant) constitute a radiation. In relation to its sister group, Guatteria is smaller (200–250 Guatteria species vs. c. 700 species for Clade A) and the clade might actually be seen as species poor. This lack of species might be the result of the extinction of early lineages of Guatteria in the Miocene (Erkens et al., 2007b, 2009), leading to a long branch in the phylogenetic tree. Taking this extinction into account, these clades might actually have been more balanced.
In Annonaceae, clade size is not a good predictor for the onset of a radiation, and this study illustrates again that the size of a group of organisms is not a priori evidence that the group arose from nonrandom speciation and/or extinction (Slowinsky & Guyer, 1993).
Characters associated with radiations
Invoking key innovations is controversial. A key innovation can be defined as an aspect of the organismal phenotype that promotes diversification (many other definitions exist, however; Hunter, 1998). The rationale is that a shift in diversification rate can be coupled to the evolution of a presumed key character along the same branch (Sanderson & Donoghue, 1994). However, these correlations should be made with great care, as traditionally this process simply entails the identification of whichever feature of a group seems to be most distinctive (Slowinsky & Guyer, 1993). Furthermore, a lack of replication prevents statistical testing of the putative key innovation (Schluter, 2000). Even if replication is achieved, it is possible that a character that is causally involved in increasing diversification rates in one clade might not have the same effect in another clade (Brooks & McLennan, 2002). This is because key innovations by themselves are not a sufficient reason for biological expansion, as evolution always occurs in a context (Hunter, 1998). In this article, we do not provide a comprehensive study of all factors possibly functioning as key innovations, but, rather, a simple survey to pinpoint single (or a combination of simple) factors that might be associated with the detected significant rate shifts in the family. It should be further tested whether these characters are indeed responsible for the detected radiation.
The topological method indicates imbalance at the base of three clades. Clade A has one clear leaf architectural synapomorphy (Johnson, 2003). The whole clade has distichous trunk phyllotaxis (only otherwise found in Anaxagorea and Cleistopholis Pierre ex Engl., the latter belonging to the early diverging subfamily Ambavioideae), whereas the other genera have a spiral arrangement. For Clade B, as a whole, no clear synapomorphies exist, although there is a strong geographical structure, in contrast with Annonoideae. Next to a strongly supported Neotropical clade, it contains the South-East Asian miliusoid clade, at the base of which the third imbalance occurs. The miliusoid clade is found to be separated from the rest of Malmeoideae by several pollen characters. The miliusoid taxa have globose, cerebroid or echinate, disulcate pollen, whereas other Malmeoideae have monosulcate, perforate to reticulate, boat-shaped pollen (Mols et al., 2004). The clade is not completely South-East Asian as it contains a small clade of Central American genera (Richardson et al., 2004). Because of the geographical structure in Malmeoideae, Clade B and the miliusoid clade might be species rich as the result of a radiation after a founder event.
The temporal method indicates four significant rate shifts among genera. The largest is along the stem branch of the South-East Asian genus Goniothalamus (Table 3; Saunders, 2002, 2003; Saunders & Munzinger, 2007; Nakkuntod et al., 2009). Although we detected a significant rate shift, the overall diversification rate calculated for the genus based on the stem age was not amongst the highest found in the family (Couvreur et al., 2011b). However, this result was based on estimations from the stem node (in contrast with the crown node ages used here), which could induce some bias concerning the origin of all extant taxa in the genus. Although none of the flower or fruit characteristics is remarkable in terms of evolutionary innovations in Annonaceae, it is known that fruit and seed structure is extremely diverse in Goniothalamus, presumably reflecting differing frugivores and seed dispersal mechanisms (Nakkuntod et al., 2009). Thus, it might be hypothesized that divergent selection on fruit and seed dispersal mechanisms might be a driving factor for speciation in this genus.
With respect to geography, it should be noted that, although the topography of South-East Asia (and, especially, its island archipelagos) is conducive to allopatric speciation, there is no obvious reason why Goniothalamus should have diversified so much more rapidly than other genera in the same region. Species of Annonaceae, including Goniothalamus spp., often have narrow distribution patterns. There is somewhat equivocal evidence to suggest that a smaller geographical range size in birds is associated with higher rates of diversification (Isaac et al., 2003). This conclusion is opposite the general view that the probability of allopatric speciation increases with range size (Rosenzweig, 1978). The factor range size should be further explored to see whether this ecological variable has any correlation with the patterns found in Annonaceae.
The second largest shift in diversification rate occurs along the stem towards the Central American genus Stenanona (Table 3). This genus is part of a small clade of seven Central American genera that is embedded in the large South-East Asian miliusoid clade (Mols et al., 2004; Couvreur et al., 2011b; Chatrou et al., 2012). Stenanona is found from Mexico (Veracruz) to Colombia (Nariño). In the field, it is easily recognizable by the dramatically long drawn-out, aristate petal apices, a synapomorphy for the genus (Schatz & Maas, 2010). This petal morphology, in combination with the pink to blood red colour of the flower, is suggestive of a fly pollination syndrome (Schatz, 1987). If so, this would be one of the few cases of nonbeetle pollination in the family (Gottsberger, 1999; Saunders, 2012), and perhaps a cause for the radiation of the genus. Stenanona was also identified as having one of the highest diversification rates in Annonaceae in a previous study (Couvreur et al., 2011b).
The shift in diversification rate along the stem branch of Stenanona indicates a radiation of species and, possibly, one of the adaptations as described above. However, the unsupported topology of this part of the tree in Figure 1 warrants caution. In addition, for many terminal taxa in Malmeoideae, the age of their most recent common ancestor is unknown (e.g. the sister group of Stenanona, Desmopsis Saff. p.p and Stelechocarpus Hook.f. & Thomson). An accurate estimate of the rate shifts along these terminal branches cannot therefore be given (Fig. 1). Further conclusions on the putative radiation of Stenanona should therefore be postponed until age estimates for these taxa have been obtained and support values are sufficiently high.
In conclusion, the few shifts in diversification rate that occurred in the evolutionary history of Annonaceae are not easily linked to presumed key innovations. Annonaceae is not the only group for which it has proven to be difficult to correlate biotic (and abiotic) factors to shifts in diversification rate. Because of this difficulty, there has been a recent renewal of interest in the hypothesis that cladogenesis may be random, or nearly random, with respect to the intrinsic biology of the organisms (Ricklefs, 2003; Davies et al., 2004).
Sampling and diversification analyses
The temporal method indicates that nodes in more recent time periods tend to display greater rate shifts than older nodes (Figs 1, 4B; Table 3). Several explanations can be given for this result. For instance, this could be a consequence of frequent shifts in diversification rate, whereby differences between recent clades can be detected, but these tend to average out at deeper levels. Another explanation could be that two sister clades with balanced species numbers are joined by a relatively long stem branch. This would lead to a reconstructed high rate in both sister clades relative to the rate expected for their nesting clade, a situation not recognizable from topology alone (Davies et al., 2004). Incomplete taxon sampling (or extinction) could thus be a confounding factor with respect to the balance of the tree in Figure 1, and has been shown to bias the outcome of other analyses. This is because oversampled clades will tend to have shorter branches than undersampled clades (Savolainen et al., 2002). Recent data collection has shown that the long branches subtending Goniothalamus (Nakkuntod et al., 2009; Couvreur et al., 2011b) and Guatteria (Erkens et al., 2009) are not the result of undersampling. However, for Isolona and Monodora, the effect of undersampling is clear. The branch leading to this clade is relatively long and both genera are reconstructed here as having a significant rate shift (Table 3). Further sampling of the African Long Branch Clade has broken up this branch (Couvreur et al., 2008a, b, 2011b), but reveals no major topological changes. The newly added genera form a monophyletic group with Isolona and Monodora, thus not influencing internal branch lengths at deeper levels. Indeed, adding these unsampled genera to the analysis shows that there is no significant rate shift in this clade (data not shown). In addition, some estimated crown node ages are older (Couvreur et al., 2008a) than others (Pirie & Doyle, 2012). These older estimates allow more time for speciation and the gradual accumulation of species. This example again clearly demonstrates that detected rate shifts should not be taken at face value, and that undersampling (or, more worryingly, extinction) can always influence the analysis.