Introduction

Evolutionary divergence in feeding ecology is a fundamental response to both ecological opportunity and interspecific competition, often involving coordinated change in prey preferences, foraging habitat, and trophic morphology [1,2]. The origin of new feeding modes is a defining characteristic of many adaptive radiations, including such well-known examples as cichlid fishes and Hawaiian honeycreepers [3]. Ecological release from antagonistic effects of competition and predation leads to the expectation that lineages will quickly diverge in response to ecological opportunity, resulting in “burst-like” dynamics, whereby niche divergence evolves rapidly early in the history of a diversifying clade and gradually slows as lineages evolve into new ecological modalities and saturate accessible ecospace [4–6]. This process may repeat itself as new opportunities arise in the form of biotic turnover after extinction, dispersal to new biogeographic theaters, or the origin of novel phenotypes that alter how an organism interacts with its environment. As a result, the history of many clades can be described by a series of transformational shifts on an ever-changing adaptive landscape [7,8].

The end-Cretaceous extinction event marked the beginning of a dramatic period of ecological opportunity in Earth history. The extinction of non-avian dinosaurs and the resulting availability of uncontested ecospace set the stage for spectacular inter- and intra-ordinal diversification of birds and mammals in the early Cenozoic [9–15]. Continental tectonics and long-term climate cooling throughout the Cenozoic created further ecological opportunity in the form of new biogeographic theaters (e.g., the separation of Australia from Antarctica) and new habitats (e.g., the spread of grasslands) [16–18]. The massive ecological diversification of birds and mammals in response to these opportunities reshaped ecological communities of both land and sea, and the origin of new trophic modalities was a key part of this process [19–21]. So impressive was the diversification of mammals that the Cenozoic is commonly referred to as the “Age of Mammals.”

With nearly as many species of snakes as there are mammals, however, the Cenozoic might just as well be called the “Age of Snakes” [22]. Numbering almost 4,000 species—the vast majority of which diversified in the wake of the K-Pg extinction (S1 Fig) [23,24]—snakes comprise a global radiation that accounts for over 10% of terrestrial vertebrate diversity. Snake evolution has given rise to an enormous variety of feeding habits, many of which are highly specialized and substantially different from the diets of other squamate reptiles (lizards). Numerous functional innovations facilitated the evolutionary expansion of snake diets, including the origin of novel prey subjugation behaviors [25,26], highly kinetic skulls with complex musculature [27,28], and sophisticated venom delivery systems [29–32]. The Cretaceous ancestors of modern-day snakes were already ecologically diverse [33–36], but the massive ecological shifts during the period of snake diversification following the K-Pg extinction are poorly characterized. It was during this time that snake communities familiar to present-day observers were forming, and a better understanding of the trophic transformations that took place will help inform hypotheses regarding links between snake dietary adaptations and lineage diversification [28,30,37,38].

The dietary specialization observed in most snakes combined with their high diversity further suggests that knowledge of tempo and mode in snake diet evolution may also yield more general insights into the mechanisms by which ecological and morphological novelty arises in adaptively radiating clades. The preponderance of diverse clades of dietary specialists among snakes [39] suggests, for example, that ecological specialists are no less evolutionarily versatile (sensu [40]) than ecological generalists, perhaps pointing to the importance of behavioral flexibility or a labile feeding apparatus in facilitating ecological shifts [41–43]. Snakes also display a complex mixture of generalized and specialized morphologies related to diet, and understanding the pattern and timing of ecological shifts in relation to phenotype may help answer questions about the roles of ecological opportunity and developmental constraints as controls on adaptive radiation [44,45].

In this study, we describe the dynamics of trophic niche evolution across extant snakes, combining multivariate natural history observations and a new modeling framework to investigate evolutionary tempo and mode of the multidimensional trophic niche. We assembled a dataset on snake diets (34,060 primary natural history observations of prey acquisition for 882 species) and used these data together with a new stochastic model–based comparative method that we previously developed to reconstruct evolutionary histories of dietary change from primary natural history data [46]. Our use of primary data means that the analysis method explicitly accounts not only for the multidimensional nature of the trophic niche but also for differences in the amount of data per species, so that uncertainty due to lack of knowledge is integrated into the method (Fig 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1.

Inference model (left) and empirical dataset (right) used for the reconstruction of multivariate ecological phenotypes on phylogenetic trees. Left: Hypothetical sample data illustrating the use of primary natural history data for estimating dietary niches. Observed data are the sets of counts of different prey items recorded in the sampled diets of 5 hypothetical species. Subcircles (green, yellow, and blue) denote the number of observations of different food resources within each of the 5 hypothetical species (open circles). These data represent “primary natural history data” as they originate from direct observations of organisms in nature or from examination of museum specimens. The model we developed assumes these data are generated from a set of latent (unobserved) niche states that correspond to distinct multinomial distributions over a set of prey categories (inset bar plots). The inference framework uses the observed data and phylogeny to infer the set of latent niche states and their phylogenetic distribution. Here, an ancestral niche state (thin branches) underwent a trophic shift to niche state “B” (thick branches), resulting in a paraphyletic assemblage of 3 species that share the ancestral niche state (“A”) and a set of 2 species characterized by the derived niche state (“B”). The derived state in this example is associated with trophic expansion, adding a novel resource (shown in teal) to the proportional prey utilization spectrum. Note that sampled diets for individual species vary, even for those sharing a common niche state, because the “true” niche state is assumed to be a probability distribution and is therefore characterized by intraspecific sampling variability. Right: Comprehensive species-level phylogeny of snakes from [47] highlighting major clades, evolutionary timescale, and sample size distribution for number of prey use observations. Rank abundance curve below the phylogeny and segments along the outer semicircle depict the sample size distribution for all snakes with diet observations. Gaps along the outer semicircle occur for species with no diet observations, and these species were pruned from the phylogeny prior to analysis. A total of 34,060 primary natural history observations of prey acquisition by 882 species of snakes were collated for analysis. The data underlying this figure may be found in doi: 10.5281/zenodo.4446064.

https://doi.org/10.1371/journal.pbio.3001414.g001

We find that after an initial shift away from eating invertebrates, the diversity of snake feeding habits increased rapidly after the K-Pg boundary, with substantial increases in the rate of trophic innovation associated with colonization of the Nearctic and Neotropical realms. Our results demonstrate the potential of primary natural history data for broadscale inference in macroevolution and underscore the role of repeated transformational shifts in dietary ecology in driving snake adaptive radiation.

Results and discussion

The merged phylogenetic and diet dataset contains 882 species representing 356 genera from nearly all snake families (the only exceptions being Anomochilidae and Gerrhopilidae). Per-species sample sizes (number of observed prey items) range from 1 to 746 with a mean of 38 and a median of 12, while per-genus sample sizes range from 1 to 2,753 with a mean of 95 and a median of 25, for a total of 34,060 observations (Fig 1). Most observations in the database are from direct encounters with snakes in the field or from dissections of preserved museum specimens. Combining these 2 sources of data results in a more complete picture of the prey spectrum consumed by any given species, as field and museum specimens sometimes differ in relative frequencies of recorded prey types [48]. Snake diets can vary within species, driven by sexual and ontogenetic differences in body size and by geographic variation in available prey types [49,50]. Our compilation records these details when possible, but the dataset used for analysis in the present study aggregates all records available for a given species, thereby creating a composite picture of the prey spectrum sampled by individual species.

Observational natural history data, especially with regard to snake diets, present analytical challenges because sampling is typically highly uneven across species and because the data are strongly zero inflated. We developed a Bayesian phylogenetic comparative method that models dietary niche states as unobserved multinomial distributions from which observed diet data are sampled [46]. The new method uses phylogeny and the observed counts of sampled prey items to jointly infer continuous dietary niche states for each species and their unsampled ancestors (Methods). The resulting trophic network structure is informed by both the observed diet data and the phylogenetic relationships of sampled species, and these 2 sources of information allow us to incorporate observations from species with highly variable sample sizes because the model can use information from well-sampled phylogenetic relatives to estimate dietary niches for species with poorly characterized diets.

Our analyses reveal striking among-clade variation in rates of diet evolution (Fig 2A), and the inferred trophic network structure shows substantial variation in connectivity among different categories of prey (Fig 2B). Nearly all prey groups have an associated set of specialist predators, but more generalized predators occur almost exclusively among snakes that feed on terrestrial vertebrates. The relative absence of generalized diets that include invertebrates and fishes may stem from the unique adaptations required to subdue and consume these prey and the constraints imposed by small body size and specific macrohabitat associations [51]. Even among more generalized species, however, sampled diets rarely include more than 2 or 3 distinct kinds of prey, and there are clear tendencies for some combinations of prey items to co-occur more commonly in sampled diets than others, reinforcing prior concepts of snake feeding guilds [52–54]. Proposed mechanisms for how these associations arise include the correlated co-occurrence of prey items in the environment as well as chemical and functional similarity of exploited prey [38,55–57].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2.

Evolutionary dynamics of diet evolution across the radiation of extant snakes (top) and model inferred trophic network structure (bottom) estimated from 34,060 primary natural history observations of prey acquisition. Top: Reconstructions of ancestral snake diets (branch colors) and evolutionary rates of prey switching (outer semicircle) were inferred using a Dirichlet-multinomial Markov model for multivariate ecological trait evolution. Branch colors denote reconstructed patterns of resource use and are colored according to the same scheme used in the bottom panel. Outer semicircle denotes average rates of diet evolution for each taxon expressed relative to the average for all snakes (see text for details). Evolutionary rates are higher for the colubroid mega-radiation, which accounts for the majority of global snake diversity. Despite generally lower evolutionary rates, however, non-colubroid snakes display a similar breadth of feeding modalities as colubroids. Time-calibrated phylogeny for the 882 species for which diet observations were available was taken from [47]. Bottom: Graphical network illustrating connections between prey types (numbered circles) and individual snake taxa (filled circles). Lines connect snake taxa to prey items in their diets, and line widths are proportional to the model-estimated relative importance of each prey item to a given taxon’s diet. Each prey item is represented by a color (shown by the borders of numbered circles), and the color assigned to an individual snake species is an additive mixture of the colors of the prey items it feeds on. Prey items that commonly co-occur in snake diets are positioned near one another, as are snakes with similar diets. Several rare prey categories (crocodilians, turtles, amphibian eggs, amphibian larvae, and caecilians) are not included here but do not qualitatively alter the appearance of the overall trophic network. The data underlying this figure may be found in doi: 10.5281/zenodo.4446064.

https://doi.org/10.1371/journal.pbio.3001414.g002

The inferred trophic structure suggests that vertebrate prey used by snakes can be loosely arranged along a primary axis with terrestrial endotherms (birds and mammals) on one end and aquatic ectotherms (fishes) on the other (Fig 2). Along this axis, terrestrial ectotherms (mainly frogs and squamates) occupy intermediate positions, with amphibians closer to fishes and squamates closer to birds and mammals. At the broadest scale, these associations are likely to be driven, in part, by effects of body size and macrohabitat. Aquatic snakes that prey on fishes regularly encounter frogs that rely on water for reproduction and larval development, for example, and only larger snakes can safely subdue and consume birds and mammals. Many of the commonly recorded invertebrate prey items in sampled snake diets are themselves dangerous predators (centipedes, spiders, and scorpions) that are important sources of mortality in squamate reptiles [58,59], requiring large body size and venom to subdue [60]. Likewise, several groups that are disproportionately used by snakes are heavily defended by shells (snails) and defensive mucosal production (slugs and annelids) and require specialized behaviors, dentition, and oral gland secretions to surmount [61–64]. Interestingly, prior studies indicate that these invertebrate prey groups appear rarely in sampled lizard diets or form minor components of diets rich in other invertebrate groups, contrasting with the extreme level of specialization observed in snakes [65,66].

Our analysis reconstructs the most recent common ancestor of living snakes as feeding exclusively on invertebrates (mainly insects) with high probability, followed by an early shift to a vertebrate diet (Figs 2 and 3). An insect-feeding ancestor is consistent with phylogenetic relationships among snakes as currently understood: The earliest diverging snake lineages, scolecophidians (blind snakes), comprise a paraphyletic assemblage of species that feed almost exclusively on eusocial insects. How morphologically and ecologically representative blind snakes are of early ancestral snakes is contentious. Some have considered them phenotypically similar to the earliest snakes [67,68], but others have argued that they are highly derived and cannot be considered morphological or ecological analogs of snake ancestors [69,70]. The situation is further obscured by conflicting placements of blind snakes with respect to fossil snakes and other crown group snakes across different datasets and analysis techniques [71]. Regardless, it is clear that the shift to vertebrate feeding happened early in snake evolution, maybe even facilitated by the consequent increase in gut volume resulting from adaptive morphological changes in response to fossorial habits [72].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Expansion of trophic diversity during the global radiation of extant snakes.

Left: Diet-through-time profile showing proportional representation of diet states among lineages at consecutive time slices from the root to the present, with each color representing a distinct multivariate trophic resource state. Colors for each diet state are a proportional mixture of the colors assigned to each prey group under the inferred multinomial distribution (Fig 2). The dominant prey group for a few select diet states is illustrated by the inset silhouettes (e.g., yellow, lizards dominant; purple, insects dominant). Right: The full set of diet states and species assigned to each state, showing interspecific variation in sampled prey items assigned to a particular diet state (proportional utilizations of specific prey classes is outlined in Fig 1). Prey groups are colored following the same scheme used in Fig 2. The diversity of snake dietary niches expanded markedly during the Eocene, when reconstructed cladogenetic events mark the origin of many higher taxonomic snake lineages. Figure illustrates the sample from the posterior with the highest probability for K = 1,000; additional samples are shown in the Supporting information section (S2, S3, S5, and S6 Figs). There is considerable uncertainty in the precise sequence of overall trophic expansion across snakes, but the qualitative pattern showing an early shift to a specialized vertebrate diet followed by an Eocene expansion in among-lineage diet breadth remains largely unchanged (S6 Fig). Each diet state (left) is plotted such that its age of origin corresponds to the crown clade age of the reconstructed ancestor where it first appears. Inset images from PhyloPic are available under public domain. The data underlying this figure may be found in doi: 10.5281/zenodo.4446064.

https://doi.org/10.1371/journal.pbio.3001414.g003

Modifications in skull morphology associated with the origin of large-gaped snakes (Macrostomata) led to the elaboration of vertebrate feeding strategies [28], and snakes subsequently diversified into many distinct feeding modalities after the K-Pg boundary during the Eocene, a time when squamate communities were beginning to recover from end-Cretaceous extinctions [24] (Fig 3). The tempo of trophic expansion during this time is substantially more rapid than the pattern expected under a null model of ecophenotypic diversification (P < 0.05; S2 Fig). Per-lineage dietary niche breadths remained narrow and relatively constant over the same time period (S3 Fig), indicating that the rapid expansion in diet ecospace occupied by snakes was due to repeated transformational shifts in prey preferences and suggesting a possible role for ecological opportunity due to reduced competition in the wake of the K-Pg extinction event.

Our finding that ancestral snake diets were narrowly specialized rejects the idea that many specialized feeding modalities originated from more generalized ancestors (S3 Fig). A pattern of generalists giving rise to specialists was a widely held expectation in early conceptualizations of adaptive radiation [73], but our results suggest that specialists are no less evolutionarily versatile than generalists [74]. This is not to say that no dietary shifts toward highly specialized feeding modalities were preceded by generalized ancestors. For example, in our reconstructions, many egg-eating snakes arose from ancestors inferred to occasionally eat eggs as part of a broader diet, consistent with previous findings [56]. However, in other cases, such stepping-stone–like patterns appear unlikely. For example, 13 of the 15 recorded prey items for the Neotropical dipsadine Rhachidelus brazili are bird eggs, but no bird eggs are recorded in 139 prey items from the diets of its 5 closest relatives, which consist largely of lizards, snakes, and mammals. These results suggest the potential for occasional dramatic (rapid) dietary shifts, an inference that is supported by some observations from present-day snake populations. At least one population of Galapagos snakes (Pseudalsophis), for instance, has taken to intertidal foraging on coastal fishes [75], a behavior unknown from any other populations or close relatives [76] that lends support to the claim that niche shifts are frequently initiated by changes in behavior [77]. These results are consistent with adaptive landscape models—which predict that “peak shifts” toward new phenotypic optima entail explosive change away from current optima [78]—and suggest that ecological trait divergence is, in some cases, consistent with theoretical expectations developed for morphological data.

Our results imply that a striking diversity of trophic modalities are inferred to have originated from a lizard-eating ancestor in a relatively brief period of time (Fig 4, S4 Fig). The evolutionary dynamics of prey switching are quantified using evolutionary flux, a metric that measures gains and losses of different prey groups while accounting for the continuous nature of the dietary niche (Methods). Lizards are abundant in the same terrestrial environments as snakes, and their generally small body size compared with most snakes makes them suitable prey for a broad range of snake body sizes and gape widths. Indeed, many snakes that feed on birds and mammals as adults have juvenile diets comprised of lizards [79], and lizards may have been the target of early selection during the shift to a vertebrate diet. However, there remains considerable uncertainty in the precise sequence of overall trophic expansion across snakes (S5 and S6 Figs).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4.

Evolutionary flux between major trophic resources (top) and rate of trophic evolution (bottom) during the ecological radiation of snakes. Top: Each subpanel illustrates the average number of transitions between a given resource category (e.g., lizards: upper left, yellow) and all resources (numbered circles; see Fig 2); line thickness is proportional to the number of transitions. Reconstructions of ancestral snake diets were inferred using a Dirichlet-multinomial Markov model, and gains and losses between ancestors and descendants were computed under an optimal transport model (see text for details). Colors and numbers follow the same scheme used in Fig 2. Colored lines depict inferred evolutionary gains of different prey categories from the ancestral prey category highlighted in color, and line widths are proportional to the total number of inferred gains. Transitions from only 4 ancestral prey categories are shown. Numerous independent origins of similar feeding strategies occur across the snake tree of life, often from a lizard-eating ancestor. Gains and losses are unequally distributed among prey categories, and some feeding strategies show much greater turnover than others, suggesting that feeding strategies differ in evolutionary accessibility and versatility. Bottom: Reconstructed rates of trophic evolution across 4 major snake radiations indicate that neartic and neotropical (NW) clades exhibit greater net rates of diet evolution than their OW relatives, suggesting that colonization of new biogeographic theaters has been an important source of ecological opportunity in the adaptive radiation of snake diets. In panel (iii), OW relatives include Stichophanes (Dipsadinae) and Pseudoxenodon (Pseudoxenodontinae). Histograms depict the posterior distribution of average clade rates and are derived from evolutionary flux between different trophic resources (see Methods). The data underlying this figure may be found in doi: 10.5281/zenodo.4446064. NW, New World; OW, Old World.

https://doi.org/10.1371/journal.pbio.3001414.g004

Our analysis recovers numerous independent origins of similar feeding strategies across the global snake radiation. Notably, independent origins of specialized mammal eaters first appear unambiguously in ancestral states with the most recent common ancestors of vipers, boas, and pythons during the Eocene, a time when rodents (the predominant mammals recorded in snake diets) were spreading and diversifying around the world [80] and consistent with prior suggestions that the rise of mammals, particularly rodents, provided ecological opportunity for the diversification of some snake clades [30,37,81,82]. Perhaps, most remarkably, vermivory (earthworm feeding) has arisen independently in nearly all major snake lineages, including typhlopids (Acutotyphlops subocularis [83]), uropeltids, xenodermids (Achalinus [84]), pareids (Xylophis [85]), viperids (Atheris barbouri [86]), homalopsids (Brachyorrhos [87]), elapids (Toxicocalamus [88] and Ogmodon [89]), lamprophiids (Oxyrhabdium [90]), natricids, pseudoxenodontids (Plagiopholis [91]), dipsadids, and colubrids. Phylogenetic autocorrelation in the proportion of annelids in sampled snake diets is the lowest of all prey categories: Despite a similar number of reconstructed gains, annelids in sampled snake diets show considerably lower phylogenetic clustering than vertebrate prey items like mammals and fishes (S7 Fig). Such differences hint at the possibility that feeding strategies differ in evolutionary accessibility and versatility, and earthworm feeding may be among the most evolutionary and ecologically accessible feeding strategies available to snakes. Alternatively, low phylogenetic autocorrelation may also suggest that vermivory is a so-called “self-destructive” trait and that there are limited opportunities for species diversification for lineages that switch to an earthworm diet [92].

Reconstructed ancestor-descendant diet sequences reveal evidence of elevated rates of change among colubroid snakes, a cosmopolitan clade comprising most of living snake diversity (Fig 2). Rates of change measure the tempo of prey switching by snake lineages and are calculated by dividing the total evolutionary flux among prey groups by the span of time over which it occurred (Methods). A number of key innovations are hypothesized to have facilitated the spectacular diversification of colubroids and their wide range of dietary adaptations, including the decoupling of locomotory and feeding behaviors and the freeing of the mandible from its role in intraoral prey transport [28,30,32]. Within colubroids, some of the fastest rates of dietary change are associated with colonization of the Nearctic and Neotropical regions by the colubrid subfamilies Natricinae, Dipsadinae, and Colubrinae, consistent with observations from other snake clades that show that new biogeographic opportunities spur evolutionary innovation (S8 Fig) [93]. Within natricines, for example, colonization of the New World resulted in a roughly 200% increase in the rate of trophic niche evolution (posterior mean rate) relative to putatively ancestral background rates for the clade (Fig 4, S8 Fig). Similar increases were observed for dipsadines (90%), colubrines (64%), and viperids (15%).

Colubrids show systematically higher net rates of change than non-colubrids (S9 Fig), suggesting that clade-level differences in dietary lability rather than timescaling effects [94–97] play a role in driving trophic rate variation and hinting at the possibility of a general coupling between rates of lineage diversification and rates of trophic evolution. In spite of generally higher rates of trophic innovation in colubrids, however, nearly all feeding modalities observed in the current dataset also occur in other colubroid lineages that diverged prior to the origin of colubrids, indicating that the colubrid mega-radiation has been facilitated more by an ability to exploit existing ecological opportunities rather than by invasion of previously inaccessible trophic niches.

Conclusions

We demonstrated how primary natural history observations can be integrated with stochastic model–based comparative methods to describe the evolution of complex ecological phenotypes. Although our focus was the ecological diversification of snakes, the methodological and visualization framework we describe can be applied to many multivariate ecological or behavioral data without requiring researchers to first define and circumscribe ecological states among their study taxa. Prior comparative analyses have emphasized the deep historical roots of dietary differences observed in lizard [65] and snake [98] communities. Our results extend this perspective by explicitly quantifying reconstructed evolutionary dynamics of historical ecological transitions, revealing a dramatic ecological expansion in occupied diet space beginning in the mid-Eocene and elevated rates of trophic innovation following colonization of the Nearctic and Neotropical realms by major snake lineages.

By using fundamental observations of organisms in nature to make quantitative inferences about the macroevolution of a complex ecological trait, our study suggests new ways of integrating natural history data into comparative biology. A renewed emphasis on causal modeling in macroevolution [99] together with a concerted effort to collect and analyze primary natural history observations at broad phylogenetic scales promises to open the door to new datasets and novel lines of inquiry into the macroevolution of complex phenotypes like diet, habitat, movement patterns, and demography. Because primary natural history observations capture significantly more of the complexity in natural communities than categorical descriptors, their integration with phylogenetic comparative methods may hold the key to long-standing questions about the evolution of specialization, how patterns of resource use affect the evolution of phenotypes, or how and why species evolve new ways of life.

The data included in the present study represent the combined effort and observations of numerous field workers over many decades. Even with thousands of prey observations, however, there is still a substantial fraction of snake diversity that is poorly known ecologically. The 882 species included in our study cover less than a quarter of living snake diversity, and fully half the species in the dataset are represented by only 12 or fewer observations. Our results call attention to this deficit and highlight a critical need to gather more natural history data to help advance our understanding of how complex ecological traits evolve.

Methods

Supporting information

Acknowledgments

We thank Greg Schneider for facilitating access to the University of Michigan Museum of Zoology biodiversity collections, which served as the source of some of the diet data analyzed here. We thank Alison Davis Rabosky, Stephen Smith, and Ed Ionides for comments on the manuscript and discussion of ideas and methods that improved the research presented here.