The availability of increasingly comprehensive phylogenies has provided unprecedented opportunities to assess macroevolutionary patterns, yet studies on invertebrate diversification are few. In particular, despite the ecological and evolutionary importance of ants, little is known about their tempo and mode of diversification. Recent advances in ant phylogenetics can now provide a basis for rigorous analyses of the diversification of ant lineages. The goals of the present study are threefold. First, we demonstrate that a hypothesized disproportionate increase in ant diversification during the angiosperm radiation is largely artifactual. Rather, current evidence points to a fairly constant rate of lineage growth during its history. Moreover, an analysis of diversification patterns across the ant phylogeny indicates considerable rate heterogeneity among lineages. Indeed, and contrary to the expectation if lineages had experienced a single rate of lineage increase, we found no correspondence between genus age and diversity. Finally, we demonstrate a statistically significant phylogenetic signal in ant diversification: closely related genera have diversities that are more similar to one another than one would expect by chance. This suggests that the capacity for diversification may be itself a biological trait that evolved during the radiation of the family Formicidae.
Ants are a remarkable case of ecological and evolutionary success (Wilson 1976; Hölldobler and Wilson 1990; Wilson and Hölldobler 2005). With more than 12,000 species worldwide, ants are nearly three times more diverse than all other eusocial groups combined (Grimaldi and Agosti 2000). In addition, although ants account for less than 2% of the known insect fauna of the world, they compose at least one-third of its biomass, with a global population of approximately 1015 individuals (Wilson 1990). An estimate of the animal fauna in the Amazon region indicated that three-fourths of the animal biomass in the terra firme forest is composed of ants and termites (Beck 1971; Fittkau and Klinge 1973). Such ecological success has resulted in the presence of ants in nearly all terrestrial environments outside Antarctica (Wilson 1987; Hölldobler and Wilson 1990).
The past seven years have witnessed a strong interest in ant molecular phylogenetics (Brady 2003; Ward and Brady 2003; Astruc et al. 2004; Saux et al. 2004; Brady et al. 2006; Moreau et al. 2006; Ouellete et al. 2006), building upon the classical work using morphological characters (Brown 1954; Baroni Urbani et al. 1992). These studies have provided important insight into the phylogenetic relationships among the main ant lineages and the extent to which different subfamilies are monophyletic, although the relationships among the most basal lineages are still a matter of dispute (Brady et al. 2006; Moreau et al. 2006). One aspect of ant diversification that has received recent attention is the suggestion that the radiation of the family was associated with the origin of flowering plants and the ensuing increased habitat complexity provided by soil and ground litter of angiosperm-dominated forests, particularly in the tropics (e.g., Wilson and Hölldobler 2005; Moreau et al. 2006). In support for this hypothesis, Moreau et al. (2006) used maximum-likelihood methods developed by Paradis (1997) to test three alternative ant diversification scenarios: (1) a constant diversification rate; (2) a variable diversification rate; and (3) two diversification rates, either before or after 100 million years (Myr). The third scenario provided the best fit to the data, suggesting a possible tracking of the rise of angiosperm-dominated forests between the Early Paleocene and the Late Cretaceous. However, the method of Paradis (1997) was developed to analyze complete phylogenies. Thus, although the study by Moreau et al. (2006) uses one of the largest ant phylogenies to date, it includes less than 2% of all known ant species. The effects of such limited taxon sampling on the inferred pattern of ant diversification are currently unknown.
The goals of the present study are threefold. First, we assess whether the hypothesized disproportionate increase in ant diversification during the Angiosperm radiation is largely an artifact of an incomplete taxon sampling. Second, we investigate the extent of variation in diversification rates among different lineages across the ant phylogeny. Third, we use maximum-likelihood methods to assess the level of phylogenetic signal in ant diversification, i.e., whether closely related genera have diversities that are more similar to one another than one would expect by chance. The implications of these results for the evolution of ants are discussed.
Materials and Methods
The analyses carried out in the present study were based on the comprehensive phylogeny of ant relationships provided by Moreau et al. (2006), which included 19 of the 20 currently recognized subfamilies. There is a disagreement over the best rooting of the ant tree (see Brady et al. 2006), but the relationships within the family seem well-supported. Although the phylogeny by Moreau et al. (2006) does not include all ant genera, those included encompass nearly 90% of all known ant species. Therefore, an analysis of this reduced dataset should nevertheless be representative of the entire family.
First, we used simulations to test whether the observed concentration of cladogenesis events observed by Moreau et al. (2006) is artifactual given the incomplete lineage sampling of ant lineages of that study. Diversification was simulated using a constant diversification, in which each lineage has the same instantaneous probability of speciating or going extinct. We used the method proposed by Paradis (2003) to combine taxonomic and phylogenetic data and estimated the speciation and extinction rates during ant diversification, assuming that those rates remained constant, using the function “bd.ext” in APE (Paradis et al. 2004). Given that the extinction rate was estimated to be zero (see below), each single lineage was allowed to evolve at a speciation rate of 0.01 (without extinction) in each iteration until reaching 12,000 species (the approximate number of known ant species to date). The complete phylogeny was then randomly pruned of the following fractions of its lineages: 0, 50, 75, 90, and 98.8%. The last fraction was chosen to represent the approximate proportion of all ant lineages included in the phylogeny by Moreau et al. (2006) and was repeated 100 times to provide a Monte Carlo approach to determine the confidence intervals of the diversification model. Lineages-through-time (LTT) plots (cumulative counts of the number of lineages over time) were used to represent the increase in ant diversity over time in each of those conditions. Although it is known that extinction did occur during ant history, a model without extinction was nevertheless chosen in the present study, both because it provided the best fit according to the method of Paradis (2003), and also because it would be the most conducive scenario to generate the pattern detected by Moreau et al. (2006), given that adding a constant extinction would result in an apparent increase in diversification near the present in the reconstructed LTT (Nee et al. 1994). However, even when a constant birth–death process was used to generate the phylogenies, the level of taxon sampling in the studied phylogeny is so low that even this difference nearly disappeared after the trees were pruned and the results were qualitatively similar (not shown). All of those simulations were done using the software Phyl-O-Gen (Rambaut 2002).
Variation in diversification rate within the ant clade was investigated using three complementary methods. First, we calculated the degree of imbalance in each node of the tree using the method of Slowinski and Guyer (1993). The number of species in each genus (all of which were assumed to be monophyletic) was based on the respective number of described species, as indicated in the ANTBASE database (Agosti and Johnson 2005; accessed on May 31, 2007). One of the issues with this type of analysis is the possibility of false-positive results due to multiple statistical tests. One proposed solution to this problem is to compare the number of nodes showing significant imbalance to the total number of tested nodes using the binomial test (Katzourakis et al. 2001). However, simulations have indicated that the probability of false-positives using the Slowinsky–Guyer test under a constant diversification rate is very small (M. R. Pie, unpubl. data), and therefore we simply provided the statistically significant nodes without any correction in their associated probabilities. Second, we tested for shifts in diversification rates using a maximum-likelihood method recently developed by Rabosky et al. (2007). This method splits the tree along an edge, fitting one rate to the subtending clade and another to the rest of the (paraphyletic) tree. This “two-rate” model is compared to a single-rate model using a likelihood-ratio test. The test is repeated on all the edges of the tree to identify when the shift in diversification was most likely. The use of the entire tree in each comparison provides a crucial difference between the method of Rabosky et al. (2007) and the more commonly used Slowinsky–Guyer test, which only compares diversification rate differences between sister clades (see Sanderson and Donogue  for a third, hybrid, method). Given that there are several branches for which the shift in diversification is most likely, we identified all the branches for which the likelihood-ratio test was significant. However, to minimize false-positives, we adjusted the critical value as 0.05 divided by the number of branches in the phylogeny. Finally, we computed a linear regression analysis of the relationship between genus age and its species richness. If ant evolution experienced a single nearly constant underlying diversification rate, there should be a statistically significant relationship between the age of different genera and their respective number of species.
We investigated the evolution of the diversification rates themselves by testing whether there was statistically significant phylogenetic autocorrelation across the phylogeny for the size of a genus. This was done by using the model developed by Pagel (1999) based on an extension of a constant-variance random-walk model (sometimes called Brownian motion). Under those conditions, the degree of similarity in a given trait between two lineages is proportional to the extent of their shared history, as indicated by the phylogeny, such that traits evolve at each instant of “time”dt with a mean character change of zero and an unknown but constant variance σ2. Pagel introduced another parameter, λ, to estimate the extent to which the phylogeny correctly predicts patterns of similarity among species. This parameter can range from 1 (as predicted by the Brownian motion model) to 0 (trait similarity among species is independent of phylogeny). Hypothesis testing using this approach is based on the likelihood-ratio statistic, which compares the goodness of fit of a model to the data with that of a simpler model that lacks one or more of the parameters. Analyses using Pagel's method were implemented using CONTINUOUS (Pagel 1999). A similar approach in the use of Pagel's method to study diversification was presented by Phillimore et al. (2006, 2007).
The inference of speciation and extinction rates based on the method of Paradis (2003) provided little evidence for extensive species turnover during ant history, given that the maximum-likelihood estimate of the extinction rate was zero (d/b= 0, SE = 0, b − d= 0.079, SE = 0.0011, where b: speciation rate and d: extinction rate). Therefore, the subsequent simulations were based on a pure-birth (Yule) process.
The effects of an incomplete taxon sampling on the LTT plots are shown in Figure 1. Although the underlying “true tree” in the simulation diversified at a constant rate (resulting in a straight line in the semi-log plot), randomly removing varying fractions of the lineages generated an apparent deceleration of lineage diversification, such that the effect was stronger as the proportion of removed lineages increased. In particular, a reconstructed phylogeny using only 1.2% of the extant lineages generated a curve very similar to that observed by Moreau et al. (2006).
There was considerable heterogeneity in diversification rates among ant lineages, despite differences among methods in terms of when shifts in diversification took place during ant history (Fig. 2). Based on the method of Slowinsky and Guyer (1993), the largest asymmetries involved the contrast between the “living fossil”Aneuretus simoni and the remaining dolichoderines (1 and 616 species, respectively) and between Tranopelta and Pheidole (2 and 969 species, respectively) (Table 1). On the other hand, the method of Rabosky et al. (2007) identified the strongest shifts in the lineages leading to Camponotus and Polyrhachis. Interestingly, it also detected significant shifts in several species-poor lineages, particularly those leading to Onychomyrmex and Concoctio, which suggests that those shifts might in fact have lead to decreases in diversification rate. However, it is important to underscore that the method of Rabosky et al. (2007) was not designed to identify multiple shifts in diversification—it involves simply the comparison of a single-rate against a two-rate model in several alternative branches. The inferred list of nodes/branches where a significant shift in diversification rate was detected is probably not exhaustive, and more refined phylogenetic information, particularly within the most diverse genera, might provide even more instances. It is important to note that several of the nodes/branches identified above are nested within each other. The interpretation of these cases should be done with caution, given that it is often unclear how to discriminate between diverse subclades causing apparent significant diversification shifts in its more inclusive nodes, or whether a more complex series of radiations took place sequentially. In fact, one should not expect that a shift in diversification place should necessarily occur on single nodes/branches. The incremental evolution of key innovations could potentially take place over a longer period of time, so that nested shifts should not simply be dismissed as something “wrong” with the analysis. Interestingly, the heterogeneity in diversification rates is also mirrored by the lack of association between the age of a genus and its species richness (Fig. 3, r=−0.1267; P= 0.1371), suggesting that heterogeneity in diversification rates also took place within genera during ant evolution.
Table 1. Number of species in each of the nodes where a significant shift in diversification rate was detected based on the Slowinsky–Guyer test.
Finally, there was a statistically significant phylogenetic signal in the genus richness during ant evolution. The fit of a null model in which diversity was independent of the phylogenetic relationships among the genera (ln lik =−264.846) was significantly poorer than a model in which phylogenetic autocorrelation was allowed (ln lik =−259.083) according to a likelihood-ratio test (P= 0.00068), with the maximum-likelihood estimate of λ being estimated as 0.267.
The simulations of the effects of an incomplete taxon sampling on the reconstructed LTT plot clearly indicate that the suggested increase in ant diversification associated with the rise of flowering plants proposed by Moreau et al. (2006) is most likely artifactual. Indeed, when a sample of the above-mentioned phylogenies was subject to the same analysis as Moreau et al. (2006), model A (constant diversification) was always erroneously rejected in deference to models B (changing diversification through time based on a Weibull distribution) and C (two diversification rates divided by a breakpoint) (<χ2A,C> = 99.7; <χ2A,B> = 157.9; N= 20 simulations). Similar effects of an incomplete taxon sampling on LTT plots have been shown previously (e.g., Nee et al. 1994), and might in fact point to the opposite conclusion, i.e., a relatively constant rate of diversification throughout ant evolutionary history. This proposition is consistent with the known fossil record in ants. Although ant representation in the oldest Lagerstätte such as the Upper Cretaceous amber of the Kheta Formation is miniscule (0.001% of all insects), that proportion increased steadily over time, reaching 40% of all insects in the Ranzano Formation in the Oligocene (see Grimaldi and Agosti 2000 for a recent review). In fact, when the relative proportion of ants in the different Lagerstätten reviewed by Grimaldi and Agosti (2000) is compared, there is a statistically significant linear increase over time (N= 10, r2= 0.8, P= 0.0004). This result indicates that the bulk of ant diversification took place in the Eocene onward, a result that is also consistent with the molecular phylogeny of Brady et al. (2006).
Forty years ago, Wilson (1976) singled out three ant genera—Pheidole, Crematogaster, and Camponotus—as the most prevalent in the world based on four criteria: species diversity, extent of geographical range, diversity of adaptation, and local abundance. The reason for their prevalence, however, would not be based on any physical or behavioral trait, but rather on their “ability to avoid competitive replacement by other prevailing groups” (Wilson 1976). The results of the present study corroborate Wilson's suggestion with respect to the unusual diversification of Crematogaster and Pheidole (Table 1). However, the Camponotus radiation appears to be more complex, with its onset predating its divergence with Polyrhachis. Moreover, several other cases of radiation within the family have been identified (Fig. 2, Table 1). The evolutionary causes underlying these events clearly deserve to be addressed in more detail by future studies.
Although the phylogeny used here does not represent a random sample of ant species, but rather an overdispersed one, the differences between the two at this level of sampling are likely very minor, at least for simple models of diversification. In both cases, most of older nodes are reconstructed. This point is clear in Nee and May's (1997) study of the effect of Raup's (1973)“field of bullets” extinction scenario on tree shape; one needs only to substitute sampling for extinction.
There is an apparent incongruence between some of our main results. The LTT simulations and the analysis of the fossil record seem consistent with a fairly constant diversification rate over time, yet the SG tests and the ML method of Rabosky et al. (2007) showed statistically significant evidence for heterogeneity in diversification rates among clades, thus raising the question: was the diversification rate constant or variable during ant evolution? We provide two hypotheses to explain this discrepancy. First, one should recognize that there are two axes on a tree, allowing for different patterns when one looks at variation over time or across clades, such that the existence of a considerable variation in diversification among clades can be reconstructed as a constant overall rate through time (Mooers and Heard 1997). Moreover, there is an important fact that is frequently overlooked in this kind of analysis, namely that a statistically significant shift in diversification rate does not necessarily imply that the shift was positive. In fact, even though there are several terms associated with positive shifts in diversification rates (e.g., adaptive radiation, key innovations), no corresponding well-established terms for negative shifts are present in the literature. A negative shift in diversification rate should have an effect similar to background extinction, given that, over time, those lineages would tend to be eliminated from the reconstructed phylogeny. Negative shifts seem widespread during ant evolution and might be associated with limited ecological flexibility or geographical range. This area clearly requires more extensive methodological and conceptual developments before one can recognize the relative importance of positive versus negative diversification rates.
It has been recently suggested that differences in species numbers among clades result predominantly from variation in their age rather than from their diversification rates (Mayhew 2007; McPeek and Brown 2007; but see Magallón and Sanderson 2001; Mayhew 2002) or from external forces, such as lithospheric complexity and the frequency of vicariance events (Cracraft 1982, 1985). The patterns inferred in the present study are inconsistent with either of the above-mentioned suggestions. First, there is no association between genus age and its species richness (Fig. 3). Moreover, there is strong evidence for a phylogenetic signal in diversification patterns, suggesting the existence of heritability of diversification potential over evolutionary time, either in terms of speciation rate or extinction risk. Such heritability is often implied in macroevolutionary studies (e.g., Heard 1996), but little direct evidence has been available to date to support it (e.g., Lovette et al. 2001; Moyle et al. 2009).
Why should there be a phylogenetic signal in diversification patterns?Savolainen et al. (2002) indicated two general classes of possible explanation. First, there is a tendency for the habitat in which a clade diversifies to be “heritable” over evolutionary time, amplifying or reducing the diversification potential of that clade. Second, diversification rates might depend on organismal traits (e.g., body size, mutation rates) that are themselves heritable over evolutionary time. The former hypothesis does not appear to play a role at the phylogenetic level of our study, given that species-rich and species-poor are largely sympatric throughout the world. The second hypothesis seems more plausible in the case of ants, yet the biological traits favoring diversification, either by increasing speciation or reducing extinction, are unclear. The phylogenetic distribution of obvious traits such as foraging strategy, colony size, and caste polymorphism do not seem to correlate directly with species richness. A large-scale comparison of the degree of evolutionary heritability of different organismal traits across ant evolution is fundamental to clarify this issue.
Associate Editor: A. Mooers
We thank E. A. B. Almeida, M. V. Domingues, R. P. Freckleton, and W. A. Boeger for helpful comments on previous versions of this manuscript.