The end Pleistocene was a time of considerable environmental upheaval (Rule et al., 2012; Malhi et al., 2016), with the extinction of most megafauna (species weighing > 45 kg) (Lyons et al., 2004) occurring on the backdrop of a rapidly changing climate (Severinghaus et al., 1998; Alley, 2003). While the loss of megafauna reduced taxonomic richness more broadly, recent work has explored the megafauna extinction’s role in altering ecosystem processes through the analysis of functional traits withing communities. These studies have found that not only does functional diversity decline (Davis, 2017), but resilience declines with the loss of megafauna that filled unique ecological roles (Hedberg et al., 2022), highlighting the hidden consequences of species diversity loss. This work is timely, as the idea to rewild landscapes with so-called modern “functional equivalents” of Pleistocene megafauna is increasingly presented as a viable solution for restoring degraded ecosystems (Donlan et al., 2006; Svenning et al., 2016).

The search for modern functional equivalents relies on having a clear understanding of the biological roles of extinct species, and therein lies the rub. Much of our understanding of the ecological function of extinct animals has been inferred through comparisons with living analogs (Janis and Ehrhardt, 1988; Janis, 1995; Mendoza et al., 2002) and community-level paleoecological analyses are often conducted using discrete taxon-level traits and characteristics (Gladstone-Gallagher et al., 2019; Hedberg et al., 2022). Trait resolution can affect interpretation of functional structure (Kohli and Jarzyna, 2021) and the use of coarse categorizations, such as dietary group assignments (e.g., “grazer,” “mixed-feeder,” “browser,” “omnivore,” and “carnivore”) also does not account for variation in behavior within communities or across a taxon’s geographic distribution. By ignoring variation, we risk limiting our understanding of the ecology of extinct species and masking community-level differences in functional diversity (Violle et al., 2012), which poses a hindrance to the possibility of rewilding.

Here, we ask how dietary function varies over the distributions of two iconic ice age taxa: mammoth (Mammuthus) and American mastodon (Mammut americanum). There are many reasons that we expect landscape-scale patterns in intraspecific variation in diet. Within the niche of a species, and reflected by its geographic range, there are conditions that are most optimal for persistence and less optimal conditions near boundaries (Hutchinson, 1957). Variation in niche fitness is reflected in patterns of population densities across geographic ranges, which exhibit a pattern of central tendency across many organisms (Brown, 1984). Additionally, there is an interrelatedness between the niche, environmental conditions, and patterns of biological responses such as predation and competition (Macarthur and Levins, 1967; Maguire, 1973). Because the types and abundance of plants are constrained along gradients of temperature and precipitation (Whittaker, 1967) we predict that variation in the dietary behavior of mammoth and mastodon follow climate, and that diets consumed in the most environmentally suitable parts of the distribution, or “core areas,” differ from diets consumed closer to geographic boundaries, or “edges” (Hutchinson, 1957; Maguire, 1973; Brown, 1984).

Individual-level trait data can provide useful information regarding variation in ecological function between individuals, across populations, and across landscapes. Stable isotope analysis (SIA) of δ¹³C from enamel (δ¹³C_enamel) is an individual-level proxy for the relative consumption of C₃- and C₄-based resources (e.g., Cerling et al., 1997; MacFadden et al., 1999; Secord et al., 2012; DeSantis et al., 2019). In environments where C₄ grasses are favored, δ¹³C_enamel can differentiate between the consumption of C₄ grass and C₃ browse to quantify degrees of browsing, mixed-feeding, and grazing behavior in herbivores (Teeri and Stowe, 1976; Cerling et al., 1998) and quantify variation across these feeding strategies within a taxon (Pardi and DeSantis, 2021; DeSantis et al., 2022).

Often as a complement to SIA, dental microwear texture analysis (DMTA) is another tool that can provide information on the diets of individual animals (DeSantis, 2016). DMTA using scale-sensitive fractal analysis measures the following attributes: anisotropy (epLsar), complexity (Asfc), textural fill volume (Tfv), and heterogeneity of complexity compared among surfaces in a 3 × 3 grid (HAsfc₃) and in a 9 × 9 grid (HAsfc₉) (Ungar et al., 2003; Scott et al., 2005, 2006; Scott, 2012; DeSantis, 2016). Through DMTA, microscopic tooth wear is used to characterize diets of differing hardness and toughness (Scott et al., 2006). High anisotropy is characteristic of individuals consuming tough food item such as flesh, in the case of carnivores, or grass, in the case of herbivores. High complexity distinguishes individuals consuming hard and brittle food items, while low complexity is characteristic of consuming soft items. In herbivores, DMTA has been successfully used to differentiate between diets ranging from obligate grazers to browsers to frugivores (Scott, 2012), including diets that are isotopically similar (DeSantis, 2016; DeSantis et al., 2017).

Our aim is to place intraspecific variation from SIA and DMTA into a geographic context to identify landscape-scale patterns in diet of mammoths and mastodons. Species distribution modeling (SDM) has been increasingly used to study the distribution of species over space and time, especially in response to climate change, including among fossil taxa (Martínez-Meyer et al., 2004; Elith and Leathwick, 2009; Maguire and Stigall, 2009; Wang et al., 2021). Such analyses leverage the availability of detailed paleoclimate reconstructions (e.g., Collins et al., 2006; Brown et al., 2018) as well as fossil occurrence data that are now widely accessible through databases (e.g., Williams et al., 2018). Presence-only methods, such as Maxent, are especially useful for analyzing the likely distributions of extinct species for which occurrences can be verified, but absences are uncertain or unknown (Phillips and Dudík, 2008; Elith et al., 2011). By modeling the distribution of species, relationships between climatic variables and likelihood of occurrence can be established, and geographic regions can be assessed as being more or less, suitable to a taxon.

Species distribution modeling and individual-level dietary proxies are useful tools for understanding the ecology of species. Here, we combine these approaches to explore how multiple aspects of the niche contribute to dietary variation. Mammoth and mastodon were selected for this study because they are broadly distributed, are well represented in the late Quaternary fossil record, and have diets that have been well described. Mammoth have morphological adaptations (Maglio, 1972) that permit a broad diet ranging from mixed-feeding to grazing (Smith and DeSantis, 2018, 2020; Pardi and DeSantis, 2021; DeSantis et al., 2022). In contrast, mastodon are browsers with a more narrow breadth in δ¹³C (Green et al., 2017; Smith and DeSantis, 2018, 2020; Pardi and DeSantis, 2021; DeSantis et al., 2022), but the type of browse that is consumed can come from a variety of sources (Lepper et al., 1991; Newsom and Mihlbachler, 2006) and can vary over time and space (Green et al., 2017). Our approach is to model their distributions using Maxent, and then compare variation in dietary proxies (via SIA or DMTA) over space. We ask if the diets of mammoth and mastodon living in the core areas of their distributions (and niches) differ from those living closer to the edges.

Rather than attempt to strictly define geographic ranges of mammoth (Mammuthus) and mastodon (Mammut americanum), which SDMs often over- or under-predict (Mellert et al., 2011; Marcer et al., 2013; Lee-Yaw et al., 2022), our aim was to compare dietary function within Core areas of greatest likelihood to dietary function in Edge areas of lower likelihood (Peterson et al., 2018; Figures 2, 4). We established correlative relationships between climatic variables and likelihood of occurrence to test the hypothesis that mammoth and mastodon consume variable diets at the landscape scale according to relative suitability of the environment using the maximum-entropy approach of species distribution modeling. We found evidence for significant dietary preferences of mammoth (Figure 3 and Table 1), while preference in mastodon was less resolved (Figure 5 and Table 2).

Despite the cosmopolitan distribution of mammoth and their apparent high variation of dietary breadth (Smith and DeSantis, 2018), greater consumption of grass in Core areas across their modeled distribution suggests a preference for grass (Figure 3) as would be expected from both their morphology (Maglio, 1972) and as a requirement for adequate rates of tooth wear (Fagan et al., 1999). These preferences resemble those of Elephas maximus (Asian elephants) which are variable mixed-feeders with a preference for grazing (Sukumar et al., 1987; Baskaran et al., 2010; Koirala et al., 2016).

As a dietary proxy, δ¹³C_enamel can discriminate between browse and graze consumption at low latitudes (below 37°) because most low-latitude grasses use the C₄ photosynthetic pathway and have tissues that are isotopically distinct from those of C₃ trees, forbs, and shrubs (Teeri and Stowe, 1976; Paruelo and Lauenroth, 1996; Cerling et al., 1997, 1998; Macfadden et al., 1999). High δ¹³C_enamel values from individuals sampled at low latitudes in our study indicate greater consumption of C₄ grass by mammoths where they are most likely to occur. Preference is exhibited when a resource is utilized at a higher frequency than it occurs, and these individuals are from landscapes of mixed, but C₃-browse dominant, resources. Local pollen records from Camel Lake (Watts et al., 1992; Wang et al., 2019) and Lake Tulane (Grimm et al., 1993; Wang et al., 2019; Figure 2) confirm the presence of mixed-parkland and deciduous forest environments coincident with SIA samples during Heinrich Stadial 1 through the Younger Dryas (Supplementary Table 6). Where, then, are these mammoth grazing?

Elephants typically maintain small home ranges (<250 km) (Bonhof and Pryor, 2022) but exhibit variable nomadic behavior that is influenced by seasonal change of habitat, the availability of food and water, as well as sex (Sukumar et al., 1987; Baskaran et al., 2010; Koirala et al., 2016). Similar movement patterns in mammoths have been inferred from ⁸⁷Sr/⁸⁶Sr analyses, although longer treks to other geographic areas are feasible (Bonhof and Pryor, 2022). If grasses are distributed heterogeneously in patches within a browse-dominated environment, mammoth could have selectively made use of these resources by traveling between patches. However, enriched δ¹³C_enamel values signaling the use of C₄ resources indicate that grazing was primarily at lower latitudes, as the relative abundance of C₃ to C₄ grasses increases with latitude (Teeri and Stowe, 1976; Paruelo and Lauenroth, 1996; Cerling et al., 1997, 1998; Macfadden et al., 1999).

Mammoth sampled from Edge areas of lower likelihood made greater use of mixed C₃ and C₄ resources (Figure 3). These individuals may be consuming the local vegetation. Pollen from Montezuma Well (Davis and Shafer, 1992) documents predominantly desert vegetation which includes on average ∼5% diversity from grasses during Heinrich Stadial 1 through the Younger Dryas, near SIA samples in the Southwest (Figure 2 and Supplementary Table 6). Another nearby pollen site is Bear Lake (Weng and Jackson, 1999), which documents predominantly spruce parkland with on average <1% grasses and sedges during the Bølling-Allerød and the Younger Dryas. Alternatively, these individuals could be acquiring a mixed signal by consuming C₃ grasses from higher latitudes: future studies of individual movement can help clarify where these animals are foraging.

Dietary preferences are less clearly defined for mastodon. Their isotopic breadth is, overall, relatively low and indicates restricted use of C₃ resources (Figure 1; Pardi and DeSantis, 2021). However, analyses a coprolites demonstrate that they consumed a broad variety of C₃ plants (Lepper et al., 1991; Newsom and Mihlbachler, 2006). We therefore analyzed published DMTA values to explore variance in food texture across their modeled distribution. Mastodon from Core areas of highest likelihood of occurrence have somewhat higher complexity (Asfc) values (median = 1.904, Figure 5 and Table 2) than Edge areas (median = 1.055) which hints at greater consumption of brittle or hard food items (bark, nuts, and seeds) in Core areas, although this difference is non-significant (p = 0.0788). The bulk of what is known about mastodon diets inferred through DMTA comes from a handful of studies comparing regional and temporal variation across sites of differing habitats (Green et al., 2017; Smith and DeSantis, 2018, 2020) and age demographics (Smith and DeSantis, 2018). Sampling for mastodon DMTA does not have nearly the breadth of geographic coverage as SIA for mammoth (Figures 2, 3). Future analyses of DMTA from samples collected outside of the Core areas of highest likelihood will help to clarify dietary preferences.

Although we were unable to identify significant differences in mastodon diets according to their likelihood of occurrence using DMTA, contemporaneous pollen records point to habitat types that may have been more suitable. Core areas of high likelihood of occurrence are consistently present for mastodon in the northern midcontinent during the late Pleistocene (Figure 4). Numerous pollen records from this region (Brewster Lake, Appleman Lake, Stotzel-Leis Site, and Carter Site) indicate the presence of mixed-parkland and prairie throughout the this time (Shane, 1987; Curry et al., 2007; Gill et al., 2009; Figure 4; Supplementary Table 6). This contrasts with pollen records near mastodon in Edge areas in the southeast, such as Camel Lake (Watts et al., 1992; Wang et al., 2019) and Lake Tulane (Grimm et al., 1993; Wang et al., 2019; Figure 4) which confirm the presence of mixed-parkland and deciduous forest.

When considering the scale and temporal grain of different dietary proxies, δ¹³C_enamel and microwear can reasonably be applied to questions of habitat tracking and behavioral plasticity (Davis and Pineda-Munoz, 2016). δ¹³C_enamel integrates the isotopic signal of the resources being used over the weeks to months of life during which the enamel mineralizes, which could also mean an integration of resources used across a geographic area for a highly mobile animal. For this reason, SIA is a reasonable proxy for what an animal is eating at the scale of a landscape (Davis and Pineda-Munoz, 2016) as we have applied it here. In contrast, microwear records properties of the food consumed during the days to weeks leading up to death (Grine, 1986). It therefore has high fidelity to the specific location where an individual is found as a fossil. DMTA has been applied to characterize interspecific dietary variation (Scott, 2012; DeSantis, 2016) as well as intraspecific variation over time and space (Rivals et al., 2007).

The mastodons sampled for DMTA from Friesenhahn Cave warrant further comment. These individuals are juveniles and have low complexity values (Supplementary Table 3). It is unclear if these low values represent ontogenetic niche partitioning or dietary differences that are reflective of the environment, and the question of whether there are ontogenetic shifts in mastodon diet, more generally, is still open (Smith and DeSantis, 2018). The megafaunal remains at Friesenhahn Cave span a dynamic time (15–20 ka) (Graham et al., 2013) where the region experienced a shift from high likelihood of mastodon occurrence to low likelihood (Figure 4). The local vegetation around Friesenhahn Cave is described in other studies as C₄ open grassland with some riparian forests (Hoppe, 2004; Koch et al., 2004; Graham et al., 2013), consistent with our classification of this locality as outside the Core niche of mastodon.

There are refinements that could be made to our models that should be considered before any application beyond the scope of our study. High likelihood of occurrence is not equivalent to presence (Elith and Leathwick, 2009), and we do not account for physical barriers to dispersal. Doing so, however, would not qualitatively affect our results, as we are only concerned with drawing interpretations from dietary proxies from known occurrences. Our models are explicitly limited to relating climate variables with occurrence, which assumes climate is adequately correlated with resources that are required by Proboscideans (e.g., water and vegetation) (Sukumar et al., 1987; Baskaran et al., 2010; Koirala et al., 2016; Bonhof and Pryor, 2022). Realized niches are impacted by variables beyond climate, including biotic interactions (Hutchinson, 1957; Leibold, 1995; Chase and Leibold, 2009), which our models do not account for.

A potentially strong biotic interaction that warrants future investigation is competition. As some of the very largest animals on the landscape, Proboscideans function as ecosystem engineers (Owen-Smith, 1992). Today, African elephants (Loxodonta africana) compete with mesobrowsers and mesomixed feeders, but facilitate mesograzers (Fritz et al., 2002). The presence of megaherbivores, such as modern elephants, impacts the feeding of smaller sympatric herbivores; however, Pleistocene environments supported a much more diverse megaherbivore community. Were sympatric megaherbivores competing with each other, or partitioning resources? DMTA analyses of three middle/late Miocene Proboscideans (Gomphotherium subtapiroideum, Gomphotherium steinheimense, and Deinotherium giganteum) from the Molasse Basin in Southern Germany supports niche partitioning rather than competition for resources (Calandra et al., 2008). In contrast, competitive exclusion imposed by Mammut and Mammuthus is a plausible cause for the eventual extinction of gomphotheres (Cuvieronius) in North America (Smith and DeSantis, 2020).

Mammoth and mastodon sometimes do co-occur, at least regionally. In the Northern, North American mid-continent, they exhibit partial isotopic niche overlap in their use of C₃ resources (Widga et al., 2021), although whether this represents consumption of C₃ browse or graze within these mammoth is not resolved. The influence these taxa have on each other more generally when they share a landscape remains to be clarified. Do they partition resources, and if so, how is this impacted by geographic variation in environmental suitability as we have identified in the present study? Our models identify geographic locations where there is overlap of Core niche areas for mammoth and mastodon, as well as areas where one is more likely to occur over the other. For example, the Core area of mammoth overlaps with the Edge areas for mastodon in Florida. In contrast, there appear to be few regions where the Core area of mastodon overlaps where mammoth are at the edge of their niche. Our current dataset is not extensive enough to directly compare diets within and outside of areas of sympatry, but our models could provide context for future comparisons.

Our study is unique in that it aims to explore dietary preference and spatially explicit variation within the context of the ecological niches of extinct taxa. Dietary proxy data alone can be incredibly useful for quantifying how resources are being use by individuals and populations over time and space (MacFadden and Cerling, 1996; Scott et al., 2005, 2006; DeSantis, 2016). Our approach provides a means for contextualizing proxy data by the suitability of the environment across the geographic distribution of a taxon. This context matters because ecological interactions and factors influencing populations at geographic range margins are not the same as in the center of the range–boundaries exist because of limiting conditions preventing persistence that are not so limiting elsewhere (Hutchinson, 1957; Brown, 1984).

The approach presented here provides a means for applying distributional context to interpretation of paleoecological data, where occurrences are known but boundaries of geographic ranges are poorly defined. This may be particularly useful when considering the potential causes and consequences of extinctions across landscapes. For example, one could reasonably point to climate and environmental change as a possible cause for late Pleistocene declines and extirpation of mastodon and mammoth in the Central and Southern Great Plains, as likelihood of occurrence decreases with climate change over time (Figures 2, 4). However, losses in the Core area of their distributions may have other causes, such as hunting and landscape changes caused by humans, which is supported by synchronous patterns of extinction and the timing and trajectory of human dispersal (Surovell et al., 2005, 2016; Hamilton and Buchanan, 2007).

Analyses of functional diversity can reveal hidden consequences of taxonomic diversity loss. However, applying taxon-level traits in functional diversity analyses may mask variation that arises from plastic behavior. Specifically, dietary behavior can vary due to the non-uniform distribution of resources and biotic interactions over space. Here, we illustrate this point by exploring how diet, measured through proxies, varies over the modeled distributions of mammoths and mastodon. Mammoth dietary behavior varies by context across their geographic distribution, despite having evolutionary adaptations for grazing and exhibiting a preference for grass overall as a taxon. In contrast, specific dietary preferences for mastodon are less resolved and our analyses were unable to identify significant differences in the selection of browse across their distribution as it relates to likelihood of occurrence. The ecological roles of some species may be context specific and need to be critically evaluated when planning for the management of reintroductions or introducing novel species to restore lost ecological function.

The original contributions presented in this study are included in the article/Supplementary material, further inquiries can be directed to the corresponding author.

MP collected data, conducted analyses, and wrote the first draft of the manuscript. LD collected data, aided in the development of the research, and helped with revisions to the manuscript. Both authors contributed to the article and approved the submitted version.