Introduced herbivores restore Late Pleistocene ecological functions

Significance

Humans have caused extinctions of large-bodied mammalian herbivores over the past ∼100,000 y, leading to cascading changes in ecosystems. Conversely, introductions of herbivores have, in part, numerically compensated for extinction losses. However, the net outcome of the twin anthropogenic forces of extinction and introduction on herbivore assemblages has remained unknown. We found that a primary outcome of introductions has been the reintroduction of key ecological functions, making herbivore assemblages with nonnative species more similar to preextinction ones than native-only assemblages are. Our findings support calls for renewed research on introduced herbivore ecologies in light of paleoecological change and suggest that shifting focus from eradication to landscape and predator protection may have broader biodiversity benefits.

Abstract

Large-bodied mammalian herbivores dominated Earth’s terrestrial ecosystems for several million years before undergoing substantial extinctions and declines during the Late Pleistocene (LP) due to prehistoric human impacts. The decline of large herbivores led to widespread ecological changes due to the loss of their ecological functions, as driven by their unique combinations of traits. However, recently, humans have significantly increased herbivore species richness through introductions in many parts of the world, potentially counteracting LP losses. Here, we assessed the extent to which introduced herbivore species restore lost—or contribute novel—functions relative to preextinction LP assemblages. We constructed multidimensional trait spaces using a trait database for all extant and extinct mammalian herbivores ≥10 kg known from the earliest LP (∼130,000 ybp) to the present day. Extinction-driven contractions of LP trait space have been offset through introductions by ∼39% globally. Analysis of trait space overlap reveals that assemblages with introduced species are overall more similar to those of the LP than native-only assemblages. This is because 64% of introduced species are more similar to extinct rather than extant species within their respective continents. Many introduced herbivores restore trait combinations that have the capacity to influence ecosystem processes, such as wildfire and shrub expansion in drylands. Although introduced species have long been a source of contention, our findings indicate that they may, in part, restore ecological functions reflective of the past several million years before widespread human-driven extinctions.

Global extinctions and range contractions of large-bodied mammalian herbivores have occurred across the world beginning ∼100,000 y ago and peaking toward the end of the Late Pleistocene (LP) (1). Emerging consensus indicates that LP losses were primarily driven by prehistoric human impacts (2, 3) either alone or synergistically with climate change (4). On the other hand, recent introductions of herbivore taxa outside their native ranges has increased species richness across much of the world, in some continents to levels approaching the LP (5).

The prehistoric declines of large-bodied herbivores led to widespread ecosystem changes, including reduced nutrient cycling and dispersal, reduced primary productivity, increased wildfire frequency, and intensity, and altered vegetation structure (6–8). Likewise, introduced herbivores have been found to drive changes in vegetation structure (9), to increase water availability in deserts through grazing and disturbance (10), and to reduce fuel loads and thus wildfire (9, 11).

These effects emerge from the distinct ecological functions of large herbivores. Here, we define “function” as the capacity of organisms to affect their environment as determined by their combinations of traits, such as body mass, fermentation type, and diet (12) (SI Appendix, Fig. S1). For example, large-bodied hindgut grazers have the unique capacity to bulk graze large quantities of low-nutrient grasses (8, 13, 14). However, the downstream ecological effects of this function vary with ecological context (e.g., precipitation, soil type, and predation pressure). For example, bulk grazing can lead to the formation of high productivity grazing lawns, but this process is shaped by interactions among soil nutrients, rainfall, and herbivore densities (15).

Most extant plant and animal species evolved in the context of diverse large-bodied herbivore assemblages from the early Cenozoic (30–40 million ybp) until the LP extinctions (16). However, most research on introduced herbivores has been conducted under the premise that they are ecologically novel and thereby disadvantage resident species (e.g., ref. 17). The possibility that introduced herbivores may, in part, restore the ecological functions that characterized the past several million years until the LP extinctions has been suggested (18–21) but has not been rigorously evaluated.

Here, we analyze how the twin anthropogenic forces of extinction and introduction have shaped herbivore functional diversity and the extent to which introduced herbivores restore lost, or introduce novel, ecological functions relative to preextinction LP assemblages. We do so by comparing native-only and inclusive (native and introduced) herbivore species assemblages of the present day to LP assemblages in terms of trait structure across the continents.

Results

We compiled native-only, inclusive (native and introduced), and LP species lists for continental assemblages for all herbivore species ≥10 kg recorded over the past ∼130,000 y, i.e., since the last interglacial period (Dataset S1). Globally, LP extinctions resulted in a loss of 160 of 427 (35%) herbivore species ≥10 kg, particularly in North America (67% lost), South America (65%), Australia (64%), and Europe (56%) (Fig. 1A). Thirty-three herbivore species were introduced into new continents, replacing lost species richness by 50% in Australia and Europe, 46% in North America, 42% in Africa, 27% in South America, and 11% in Asia (Fig. 1A).

Fig. 1.

Trait space changes resulting from LP extinctions and recent introductions. (A) Species richness per continent. Introductions have numerically replaced lost species richness by between 11% (Asia) and 50% (Australia and Europe). Fill color indicates species fate with the legend shared with C. Inclusive = native + introduced modern assemblages. (B) Global herbivore trait space. Arrows indicate how particular traits shape trait space axes. The first two PCoA axes (∼62% of variation) of trait space are shown (see the SI Appendix, Fig. S1 for PCoA axes 3 and 4). Points indicate species, and the fill density indicates their density distribution with the legend shared with C. (C) Changes in continental trait space (PCoA 1 and 2) from extinctions and introductions. Crosses indicate centroids of the first two PCoA axes. Locally extinct species went extinct within the respective continent but survived elsewhere. Native only = modern native assemblages; inclusive = native + introduced modern assemblages.

To understand how extinctions and introductions affected the capacity of herbivores to influence their environments, we compiled a trait dataset of body mass, fermentation type (simple gut, hindgut, foregut nonruminant, and ruminant), diet (graze and browse), habitat (aquatic, terrestrial, and arboreal), and limb morphology (plantigrade, digitigrade, and unguligrade) (SI Appendix, Table S1). Fermentation type was recorded in terms of fermentation efficiency with ruminants scoring highest. Limb morphology was included as a trait due to its influences on soil disturbance (22), locomotion (e.g., cursoriality and fossoriality), and habitat constraints, which can otherwise be difficult to infer for extinct species (23–25).

To analyze changes in trait combinations, we constructed multidimensional trait spaces using principal coordinates analysis (PCoA) to describe the primary axes by which herbivores differ from each other in terms of their traits and, thus, encapsulate the overall functionality of herbivore assemblages. Collectively, the first four axes of global herbivore trait space accounted for 78% of interspecific trait variation (Fig. 1B and SI Appendix, Fig. S2). PCoA 1 (34% of total variation) was primarily shaped by body mass (r = 0.87) and diet (grazing r = 0.78; browsing r = −0.66), while PCoA 2 (28% of total variation) primarily reflected fermentation efficiency (r = −0.93). Changes along these axes revealed that, as expected, LP extinctions of herbivores led to substantial contractions in overall trait space, shifting the balance toward smaller-bodied species with more efficient fermentation strategies (e.g., ruminants) (Fig. 1B).

To understand how the overall diversity of species trait combinations changed from extinctions and introductions, we compared the volume of occupied multidimensional trait space for LP, native-only, and inclusive assemblages. Trait space volume is a metric of functional diversity also known as functional richness (26). LP extinctions reduced the total volume of occupied multidimensional trait space by 62% globally, particularly, in Australia (99% contraction in volume), North America (83%), and South America (83%) (Fig. 2A). In contrast, introduced herbivores replaced lost trait space volume by an average of 39% globally, particularly, in Australia (100% replaced and 30% expanded over LP levels), South America (47% replaced), Europe (22%), Africa (18%), and North America (17%) (Fig. 2A).

Fig. 2.

Change in trait space volume and functional dissimilarity. (A) Difference between native-only and inclusive trait space volumes from the LP volume for each continent. Trait space volume is the four-dimensional volume of each trait space (also known as functional richness). Contractions in trait space volume following LP extinctions (native-only points) have been offset by introductions in inclusive assemblages. The dashed line indicates no change from LP. Native only = modern native assemblages (blue); inclusive = native + introduced modern assemblages (gray). (B) Total functional dissimilarity to the LP, calculated from the overlap of four-dimensional trait spaces. Functional dissimilarity (measured as Sørensen’s β) is composed of two additive components: nestedness is dissimilarity caused by being a subset of another trait space, while turnover is the degree to which assemblages do not overlap (e.g., novelty).

While introductions substantially replaced lost trait space volume, to understand whether inclusive assemblages with introduced species are more or less similar to the LP than native-only ones, we calculated functional dissimilarity (Sørensen’s β) from the overlap of native-only and inclusive trait spaces with LP ones. Functional dissimilarity, such as other β diversity measures, is composed of two additive components: nestedness, the overlap of assemblages in trait space (e.g., the degree to which one is a subset of the other) and turnover, the degree of nonoverlap in trait space (e.g., novelty) (27). We found that introduced taxa make inclusive continental assemblages more functionally similar to the LP than native-only assemblages by 39% in Australia, 33% in South America, 21% in Europe, and 13% in North America. However, introductions contribute turnover to inclusive assemblages, driven by the unique traits and trait combinations of some introduced taxa (Fig. 2B). Australia had the highest, with turnover composing 88% of the remaining dissimilarity with the LP composed of turnover. This is due to the introduction of ruminants and larger grazers than those present in its LP marsupial-dominated fauna (Fig. 2B).

To understand how introduced species relate to other taxa, we examined whether their nearest neighbors in trait space are extant or extinct. To avoid comparing species with different thermal tolerances (e.g., tropical versus arctic), we restricted comparisons by Köppen-Geiger climate zones (28) and by body mass bins. Body mass bins were calculated using the Sturges algorithm (29), which finds natural break points in continuous data distributions, thus, reducing analytic bias.

Overall, 64% of introduced species are most similar to extinct LP species rather than extant species. This is most apparent in those continents with high LP extinction rates: In Australia 93% of introduced species are most similar to extinct taxa; followed by 86% in South America, 74% in North America, and 50% in Europe (Fig. 3 and SI Appendix, Fig. S3). In Africa and Asia, which experienced few extinctions, 90% and 75% of introduced taxa are most similar to extant species (SI Appendix, Fig. S2).

Fig. 3.

Select introduced herbivores and their extinct nearest neighbors in those continents most impacted by extinctions and introductions. The color of the top bar indicates the number of extant species (per body mass bin and climate zone) that are more similar to the nearest neighbor than the introduced species is, while the lower bar color indicates dietary guild. For a full list of pairs see the SI Appendix, Fig. S3.

The similarity of introduced herbivores to extinct ones indicates that introductions have restored lost trait combinations and, thus, functions. To better understand which functions have been restored, we focused on key “metabolic” functions herbivores contribute in ecosystems by consuming plant biomass and by cycling and redistributing nutrients (8). These functions are primarily influenced by body mass and dietary guild—traits that control the quality, quantity, and type of vegetation consumed with larger herbivores prone to greater dispersal distances and capable of digesting larger quantities of fibrous low-nutrient vegetation (8, 13, 14). To understand how extinctions and introductions have affected these key metabolic functions, we categorized species into functional groups by combinations of body mass bins and dietary guilds.

We found that 42% of introduced herbivores restore extinct functional groups, particularly, in Australia (60% of lost functional groups restored) and in South America (42% restored) (Fig. 4). Three introduced herbivores (5%) contribute novel functional groups: two species introduced to Australia contribute a grazing functional group larger than any present in the LP (by ∼200 kg), while, in Europe, the introduction of a small-bodied mixed feeder (Macropus rufogriseus) contributes one novel functional group.

Fig. 4.

The loss and restoration of key metabolic ecosystem functions. Forty-four percent of introductions restore extinct functional groups, restoring 14 of 51 extinct dietary body mass groups across continents. Body mass groups were determined analytically with the Sturges algorithm, which finds natural breakpoints in continuous distributions. Three species introduce novel groups to Australia and Europe. Points indicate species and are jittered randomly for visualization within each cell.

Discussion

The redistribution of species through human introductions is primarily perceived as an environmental harm. However, our results indicate that one consequence of introductions has been to counteract global patterns of human-caused extinction by replacing lost functional diversity and making modern trait compositions more similar to those of preextinction LP assemblages. In doing so, the majority of introduced species are functional surrogates for extinct species and many restore metabolic functional groups, particularly in those continents most impacted by LP extinctions.

Many of these restored functions have the potential to affect ecosystems. For example, large-bodied browsers were severely impacted by LP extinctions in Australia (Fig. 4). The restoration of these functional groups may reduce shrub cover and promote grasslands with implications for albedo, carbon storage, and wildfire (30). Likewise, the widespread restoration of extinct large-bodied grazing functional groups, which are capable of bulk grazing large quantities of low-nutrient grasses compared to smaller grazers, could reduce wildfire intensity (30).

In some cases, introduced taxa have close phylogenetic relationships with extinct species. For example, in North and South America, introduced equids (Equus africanus asinus, Equus ferus caballus) are related or conspecific to extinct taxa (Equus francisci, E. ferus) and have restored lost trait combinations, such as large-bodied hindgut grazing. The effects of these species are poorly understood (e.g., ref. 31), but some evidence suggests that the restoration of these trait combinations can have facilitative effects on other species. In North American desert wetlands, disturbance and grazing by wild burros (E. a. asinus) of fast-growing emergent vegetation maintains open water habitat to the benefit of endemic fishes (10) and an endangered amphibian (32). Likewise, grazing by wild horses (E. f. caballus) in North American salt marshes favors higher diversity of foraging birds and increased crab density, while reducing fish density and diversity (33).

In other cases, introduced taxa are unrelated to extinct species in their new homes yet show strong functional similarities, although often in novel trait combinations. For example, introduced hippos (Hippopotamus amphibius) in South America present a chimera of multiple extinct species’ trait combinations. While they are most similar to a giant extinct llama (Hemiauchenia paradoxa, Fig. 3) in all assessed traits bar habitat use, our analysis revealed that they are nearly as similar to an extinct semiaquatic notoungulate (Trigonodops lopesi) in all traits but fermentation type. While the ecological effects of hippos in South America remain unknown, their trait combinations suggest that their effects may overlap with extinct species in certain ecosystem components (e.g., grazing and disturbance in riparian zones) and diverge elsewhere (e.g., direction and rate of nutrient transport).

The apparent novelty of some introduced traits is tempered when compared to the trait compositions of the LP. For example, the disturbance-related effects of Australia’s introduced hoofed ungulates on soils and vegetation appear novel (11). Yet the extinct short-faced kangaroos (sthenurines) possessed monodactyl hoofs and an unguligrade morphology similar to “open-plains” horses (34). However, rumination remains a novel trait in Australia where macropods’ nonruminant foregut fermentation is less thorough (35). It is possible that some of Australia’s plants lack seeds capable of surviving ruminant digestive systems, which could lead to shifts in seed dispersal efficacy and possibly to vegetation structure. Australia also received two grazers larger than any LP grazing species (by ∼200 kg). The introduction of large-bodied bulk grazing may have strong effects on Australia’s uniquely fire-driven ecosystems (36). In a long-term experimental system in Northern Australia, seasonal bulk grazing by introduced water buffalo (Bubalus bubalis) of fibrous low-nutrient grasses reduced wildfire and promoted tree establishment and survivorship through wildfire (9). Whether these ecological effects are novel or overlap with those of extinct mixed feeders is unknown.

Many important traits remain unknown for extinct taxa, such as sociality, movement patterns, foraging behavior, and vulnerability to predators. All could be factors affecting similarity with LP species and thereby shaping how introduced herbivores interact with extant native species. Yet, the diversity of LP herbivores makes it likely that many of these introduced traits have precedents. For example, although the seemingly novel rooting behavior of wild boar (Sus scrofa) is thought to disadvantage native species across their extensive introduced range (37), several extinct species likely had similar foraging strategies, including Platygonus compressus (38) in North and South America and Zygomaturus trilobus (39) in Australia. Indeed, rooting by wild boar increases tree growth rates by mixing leaf litter into upper soil layers, enhancing decomposition rates and, thus, nutrient availability in eastern North American forests (40) and increases bird abundance by increasing accessibility of food resources in northern Australia (41).

While the trait combinations of herbivores drive their functional capacity to affect ecosystems, their actual effects emerge in interaction with ecological contexts, such as predation (42) and landscape connectivity (43). Many apex predators continue to face declines (44). Likewise, ongoing landscape fragmentation restricts herbivore movements and can lead to concentrated herbivory (43). Therefore, while introductions make herbivore assemblages more functionally similar to the LP, they do not necessarily restore ecosystems to LP conditions due to ongoing anthropogenic pressures.

Introduced species have been primarily studied in the context of recent historic states under the premise that their ecological functions are novel. However, presumed novelty yields to functional similarity when compared to the LP, a time period largely reflective of the past ∼40–30 million years of terrestrial evolution (16). Compared to these preanthropogenic conditions, introduced herbivores replace lost functional diversity, make modern herbivore assemblages more similar to LP ones, and restore key metabolic functional groups in the Earth's system. Our results indicate that introductions are an inadvertent counterpoint to prehistoric and historic anthropogenic defaunation and that future research on the ecologies of introduced herbivores would benefit from incorporating deep-time perspectives.

Materials and Methods

We included all extant and extinct herbivore species ≥10 kg that lived the past 130,000 y, following taxonomy in the PHYLACINE v1.2 dataset (45). Many analyses of LP extinctions have focused on species ≥44 kg or 100 kg, however, given that Australia lost all but one herbivore ≥44 kg during the LP extinctions, we included herbivores ≥10 kg to ensure that modern native-only assemblages would have sufficient numbers of species for analysis. Herbivores were selected from PHYLACINE as species with >50% plant in their diet thereby including some carnivorans (e.g., bears). Alternative species lists for the LP exist, varying in their resolution of extinct and extant species. To verify that our results were robust to these differences, we conducted all analyses using the species list in Smith et al. (16) (SI Appendix, SI Text, and Figs. S3–S6).

To determine how modern (native-only) and inclusive (native and introduced) herbivore assemblages compare to LP ones, we compiled three datasets of herbivore species ranges. We drew continental LP assemblages from “present-natural” ranges of herbivores from Faurby et al. (45), which primarily used fossil co-occurrence records to reconstruct modern-day herbivore ranges in the absence of anthropogenic extinction and range contraction. Native-only assemblages were drawn from International Union for the Conservation of Nature Red List (IUCN Red List V6.1) range maps (46). Introduced species distributions, compiled from the IUCN Red List and numerous peer-reviewed sources and databases (Dataset S1), were added to native assemblages to establish inclusive assemblages. Only self-sustaining populations that have been moved across continents were included (e.g., intracontinental introductions were excluded). Domestic herbivores were excluded from analysis as these populations are not necessarily ecologically viable without human intervention, and their behavior and interactions within ecosystems are heavily modified by human management. However, wild populations of domestic species were included.

Given different types of error in each of these scenario’s distribution maps (e.g., ranges estimated using different methods and with different degrees of certainty), we used these range maps to create continental species lists for LP, native-only, and inclusive assemblages. We chose to conduct analyses on the continental scale as they reflect the long-term dispersal limits of herbivores and the plants and other animals they interact with, thus, setting the ecoevolutionary context of modern ecosystems. We limited our analyses to large landmasses excluding smaller islands because they have unique evolutionary histories owing to their isolation. Additionally, this avoids the inclusion of small populations of functionally unique and geographically restricted introduced species (e.g., giraffes Giraffa camelopardalis introduced to a small island in Southeast Asia).

Traits.

We collected data for traits that drive herbivores’ interactions with other species and the environment: body mass, diet, fermentation type, habitat type, and limb morphology. Body mass (kg) was collected as a continuous variable from Smith et al. (47) and PHYLACINE v1.2 (45) and was analyzed as log base 10. We collected dietary data as two ordinal variables for graminoid and browse consumption from the peer-reviewed literature, each ranging from 0 to 3, following the method of MammalDiet (48). While coarse, these two dietary categories remain paradigmatic to herbivore ecology in both ancient and modern systems and reflect key evolutionary and ecological differences in herbivore dietary adaptations (49) (SI Appendix, SI Text, and Table S1).

Fermentation type further influences the quality and quantity of vegetation herbivores consume and the nutrient stoichiometry of resulting excreta (14, 35). Fermentation type was collected as a categorical variable for five fermentation types following Hume (35): simple gut, hindgut colon, hindgut cecum, foregut nonruminant, and foregut ruminant. However, this variable was ranked in terms of fermentation efficiency when constructing trait spaces, and hindgut cecum and colon were lumped because that degree of distinction was unavailable for some extinct taxa. Habitat use was collected as three binary and nonexclusive variables (arboreal, terrestrial, and aquatic) (SI Appendix, Table S1).

Limb morphology was included as it influences disturbance-related impacts on soils (22) and is frequently mentioned in relation to the nonanalogy of introduced herbivores in Australia (e.g., refs. 11 and 50). Furthermore, limb morphology is correlated with other ecological attributes, such as cursoriality, fossoriality, and habitat constraints, which can otherwise be difficult to infer for extinct species (23–25), Limb morphology was obtained from a variety of peer-reviewed literature sources (primarily, ref. 23) and was inferred based on family and order for extinct taxa if specific morphological studies (e.g., ref. 51) were unavailable. Limb morphology was treated as three binary variables: digitigrade, plantigrade, and unguligrade (SI Appendix, Table S1).

All trait data were cross-referenced with the primary literature or handbooks (e.g., ref. 52). Diet data were unavailable for 14 extinct species (of 427 total). We imputed their diets based on their phylogenetic relationship to other species with known diets and traits using the R package “Rphylopars” v. 0.2.9 (53) and PHYLACINE phylogeny. Rphylopars was run with all default parameters including a Brownian motion evolutionary model.

Analytic Methods.

We calculated a Gower distance matrix that evaluated pairwise functional similarity between species. Gower distance is a flexible distance-based measure that can incorporate multiple variable types and has been widely used in analyses of community trait composition and structure (54, 55).

We weighted traits when calculating Gower distance with mass weighted by two; diet, fermentation, and habitat weighted by one; and limb morphology weighted half (SI Appendix, Table S1). Mass was weighted by two as it is correlated with many other life history traits, such as reproductive and metabolic rates. Limb morphology was weighted half because, although it captures morphological diversity and disturbance-related effects, the relationship between it and ecosystem effects is less well established. Other weighting methods produced similar results, but trait spaces were less biologically interpretable (e.g., body mass had no visible correspondence to trait space axes, see sensitivity analyses in the SI Appendix, Figs. S8 and S9).

Multidimensional trait spaces were constructed using the function dbFD in the R package “FD” (56).This conducts PCoA on the Gower’s distance matrix to produce synthetic traits which become the axes of the multidimensional trait space. We used the first four axes of the PCoA after reviewing axis quality (SI Appendix, Fig. S10) and because including more than four axes has been shown to distort functional relationships across a variety of real and simulated datasets (57). The relationship between PCoA axes and traits was determined using the R function envfit in the R package “vegan” with 1,000 iterations (58).

A number of functional diversity metrics exist to describe different attributes of the structure of trait spaces. Since we were primarily interested in the range of functions present, we focused on functional richness, which is a measure of the multidimensional volume of trait space (26). To understand if introduced herbivores restore lost LP trait combinations or contribute novel ones by introducing traits without LP analogs, we calculated trait space overlap with the function beta.functional.pair in the R package “betapart” (version v1.5.1) (59), which returns a Sørensen dissimilarity metric decomposed into its primary components of nestedness and turnover.

To understand if introduced species are functionally most similar to native or extinct taxa, we identified the nearest neighbor of each introduced species from the Gower distance matrix. We filtered pairs by Köppen-Geiger (28) climate zones to prevent comparing species with different physiological tolerances (e.g., arctic species with desert species). We further constrained pairs by body mass bins to prevent spurious pairings between species with identical diet and morphological traits but gross differences in body mass (e.g., without body mass filtering, the 800 kg Camelus dromedarius’ nearest neighbor is a 166 kg kangaroo because of convergence in all other traits, see the SI Appendix, Fig. S11). Body mass bins were calculated using the Sturges algorithm (29), which finds natural break points in continuous distributions and reduces analytic bias. To assess the quality of functional analogy between introduced herbivores and their nearest neighbors and to accommodate the intrinsic distinctiveness of certain species, we calculated the number of extant species per body mass bin and climate zone that are more similar than the introduced analog.

To understand whether introduced species restore key metabolic functions in their new homes, we classified herbivores into functional groups as unique combinations of dietary guild and body mass bins (as determined above). Dietary guild (grazer, browser, and mixed feeder) was classified from the twin ordinal diet scores for graze and browse (SI Appendix, Table S1).

Data Availability.

All data are available in the SI Appendix.