Tempo of trophic evolution and its impact on mammalian diversification

We compiled detailed dietary data on primary observations of mammalian diets from the scientific literature and then, using uniform criteria, categorized each species as either a herbivore (a specialist on primary producers), a carnivore (a specialist on consumers), or an omnivore (a generalist, eating both plant and animal material). Using these data and a virtually complete species-level time-calibrated phylogeny of living mammals (ref. 1, based on ref. 18) we can confirm that mammalian trophic strategies are not evenly distributed among mammalian taxa (Fig. 1). Some clades are almost exclusively carnivorous (e.g., aquatic Cetartiodactyla, such as whales and dolphins; aquatic Carnivora, such as seals and walruses) or herbivorous (e.g., terrestrial Cetartiodactyla, such as deer, cows, and antelope) and others appear to switch frequently between omnivory and herbivory (e.g., primates) or omnivory and carnivory (such terrestrial Carnivora as bears, dogs, and foxes). We used this variability to analyze the tempo of trophic evolution across living mammals and the impact of diet on speciation and extinction.

We analyzed the data mapped upon the phylogeny by comparing the fit of eight different evolutionary models using a maximum-likelihood approach (20, 21). These models allowed rates of speciation, extinction, and transition between trophic strategies to be either independent of trophic strategy (rates fixed to be equal across all strategies), constrained by trophic strategy (separate rates for each strategy), or constrained according to status as either a specialist on a single food type (herbivore or carnivore) or generalist (omnivore), with separate rates for each category. Although this method required speciation and extinction to be estimated separately, we discuss only net diversification rate, which is speciation minus extinction, in recognition of the uncertainty of estimating extinction from extant phylogenies (22) (estimated speciation and extinction rates reported in Table S1). Diversification rates and transition rates differ substantially between trophic strategies (Table 1). The best-fitting model according to the Akaike Information Criterion (AIC), which takes into account the number of parameters used in the model as well as the goodness of fit to the data, is the most complex model. This model allows speciation and extinction rates to vary depending on trophic strategy and all six transition rates to be estimated independently. According to the Akaike weights, which are the weight of evidence for each model from the set of different models used (23), there is substantial support for the best-fitting model (0.98 where the weights of all models sum to 1). The model that fixed transition rates to be equal across trophic strategies has the worst fit to the data (ΔAIC = 234 and an Akaike weight of 9.7e⁻⁵²), confirming that transitions between trophic strategies are important constraints on the uneven patterns of lineage diversity among mammals.

In the best-fitting model, analyzed using Bayesian Markov Chain Monte Carlo (MCMC) methods, herbivores have the highest net diversification rate, carnivores have a considerably lower rate, and omnivores have the lowest rate (Fig. 2A). Moreover, the tempo of transitions between the trophic strategies is highly biased. Surprisingly, the highest rates occur into omnivory from herbivory and carnivory and the lowest rates with almost zero transitions occur between herbivory and carnivory (Fig. 2B). These results are supported by the maximum-likelihood parameter estimates from the top four best-fitting models which range in ΔAIC from 19.9 to 48.9 (Table 1); net diversification rate is the highest in herbivores, intermediate in carnivores, and lowest in omnivores; the highest transition rate is into omnivory. Thus, omnivore diversity evolves primarily through transitions into that strategy and rarely by diversification within omnivorous lineages, whereas herbivore and carnivore diversity is chiefly produced through diversification.

These results could have been produced by more than one scenario of evolutionary dynamics. Many discussions of diversification have made the distinction between explanations and models that assume differing intrinsic rates of extinction or speciation, on one hand, and differing ecological limits on the number of species the environment can accommodate (highest “carrying capacity”; according to ref. 24) on the other (24–28). If there is a global limit on the number of species that can be supported within a particular dietary strategy, net diversification may be diversity-dependent whereby rates slow through time as niche space fills up until diversity is at equilibrium (e.g., ref. 29). Indeed, fossil and molecular data frequently identify a slow-down in diversification rate (e.g., refs. 30 and 31), although there are other explanations for an apparent slow-down that do not involve niche-filling (e.g., ref. 32). Under this scenario the trophic group with a lower limit on its diversity will appear over time to have a lower average rate of net diversification than a group that continues on to reach a higher limit, even though they have grown with identical diversification rates (24). The constant-rate models we have used do not allow us to distinguish this situation from one in which the two groups have had consistently different rates. Nonetheless, whether the trophic strategies inferred to have higher rates have diversified at a consistently more rapid pace, or have simply spent more of their evolutionary history diversifying at a high rate, similar biological interpretations can be applied to these two alternative mechanisms.

There may be good biological reasons to expect herbivorous, carnivorous, and omnivorous niches to vary in the diversity they can support. In any ecosystem mammalian primary consumers have access to the largest resource base because primary producers account for the largest proportion of biomass. It is therefore possible that, along with a larger number of individuals, a greater number of herbivore species can be supported globally. Additionally, the stability of plant-based food resources may facilitate diversification, whereas species dependent on food at higher trophic levels may be less able to subdivide niches in a stable fashion. There may also be strong limits on the number of omnivores that can be sustained because some food-web models suggest omnivory should be rare because it destabilizes food webs (e.g., refs. 33 and 34).

However, it is unlikely that the low net diversification rate in omnivores is entirely explained by the omnivorous niche supporting fewer species. In our dataset omnivores are not rare: species richness of carnivores and omnivores is similar (463 and 413 respectively), a pattern that is also evident in natural communities of mammals (35). Low diversification rates within omnivores may be caused by lower speciation rates if dietary specialists are more likely to undergo allopatric speciation than omnivores because specialists’ narrower food tolerances allow their geographic distributions to be more readily fragmented (36–38). Additionally, specialization can open opportunities for further specialization (39–41), thereby generating diversity through niche subdivision. Extreme illustrations of this process have been presented in several studies on phytophagous insects, which have documented higher rates of diversification in these plant-feeding groups (15). Niche subdivision, by definition, tends to increase specialization, so for omnivores (as generalists) the opportunities for producing descendant lineages that retain their dietary breadth might be limited. Accordingly, omnivores should diversify more slowly than carnivores or herbivores. The difference between herbivores and carnivores may be explained by higher rates of extinction in carnivores because carnivores are more vulnerable to disturbance of species lower down the food chain (9, 42). Indeed, higher trophic level has been shown to be correlated with higher extinction risk in extant mammals (10).

Ecologically, omnivory can be a strategy for surviving variability in resource availability (34) and, if lineages shift to omnivory during times of environmental perturbation, this may explain why diversification is low (43) and transition rates are highest into, not out of, omnivory. Taken together, these results suggest that omnivory acts as an evolutionary “sink,” which at the broadest scale contradicts the expectation that ecological generalists are sources of future diversity and ecological specialists are evolutionary dead-ends (16, 44). Nonetheless, omnivory does appear to play an important macroevolutionary role because virtually all trophic transitions within mammals involve omnivory: the estimated transition rate between herbivory and carnivory is approximately zero. This finding is perhaps not surprising: In mammals, herbivory and carnivory often entail different and antagonistic physiological and morphological adaptations. Such changes would likely require large amounts of evolutionary time, increasing the probability of a branching event and the evolution of new lineages with intermediate ecology during the transition. Many omnivorous mammals exhibit intermediate phenotypes: the teeth of omnivores, like carnivores, generally have distinct cusps and low crowns; yet like herbivores, their enamel is thickened or crenulate (15, 17). Omnivores’ intestines are similarly intermediate, with gut lengths intermediate between herbivores and carnivores (45).

Our results imply that diet is a key factor in determining the rates at which different mammalian groups have diversified; however, the link between trophic strategy and diversification could be indirect and involve other important biological or ecological traits. Among the many factors that can drive patterns of diversification, body size and population density are often proposed as strong candidates (46). For body size, the typical expectation is for speciation rates to be higher in smaller species (e.g., refs. 47 and 48) and extinction higher in larger species (49). However, recent analysis of fossils found higher rates of both origination and extinction in larger mammals (10, 50); net diversification rates were either the same in small and large mammals or higher in small species, depending on the data subset. Regardless, size is unlikely to be driving our patterns, as herbivorous mammals tend to be larger than carnivorous and omnivorous species (51) and on this basis should exhibit the lowest rates of diversification. Predictions based on population density are similarly problematic: species relying on resources at higher trophic levels usually live at lower densities (47) and should therefore experience much higher extinction rates (52); omnivores should have intermediate diversification rates. In our results this pattern is borne out in the difference between herbivores and carnivores but omnivores have the lowest diversification rates, contrary to the population density predictions.

Although the determinants of mammalian diversity are undoubtedly complex (53, 54), diet has had profound consequences for mammalian macroevolution. Through trophic strategy the major features that characterize mammals, such as homoeothermy and specialized dentition, are connected to the controls on diversification: herbivores diversify fastest, carnivores are intermediate, and omnivores are slowest. The estimated transition-rate bias between the trophic strategies contradicts some common expectations: omnivory appears to be an evolutionary sink rather than a source of diversity, and herbivory and carnivory are the basis of future diversity, not evolutionary dead-ends. The uneven distribution of lineages within mammals is thereby linked to differences in trophic strategy, yielding a conclusion that has broad implications for understanding mammalian evolution. The model we provide can be integrated with data from the fossil record and tested further using the increasingly sophisticated methods currently being developed for estimating speciation and extinction rates from molecular phylogenies.

We thank R. FitzJohn for assistance with software and for providing a prerelease version of his diversitree package during the analysis process; A. Mooers and T. Kuhn for kindly providing the randomly resolved trees used in this analysis; and C. Jones for helping with early stages of data collection. Helpful suggestions on the manuscript at various stages were provided by two anonymous reviewers, as well as S. Otto, P. Wainwright, the P. Wainwright and S.S.B.H. laboratory groups, and the Mammal Reading Group at the National Evolutionary Synthesis Center. This paper was initiated during the tenure of all four authors at the National Evolutionary Synthesis Center, and partly funded by postdoctoral fellowships and short-term visitor awards (to S.S.B.H. and S.P.; NSF 0717009).