Horses are a classic example of macroevolution in three major traits—large body size, tall-crowned teeth (hypsodonty), and a single toe (monodactyly)—but how and why monodactyly evolved is still poorly understood. Existing hypotheses usually connect digit reduction in horses to the spread and eventual dominance of open-habitat grasslands, which took over from forests during the Cenozoic; digit reduction has been argued to be an adaptation for speed, locomotor economy, stability, and/or increased body size. In this review, we assess the evidence for these (not necessarily mutually exclusive) hypotheses from a variety of related fields, including paleoecology, phylogenetic comparative methods, and biomechanics. Convergent evolution of digit reduction, including in litopterns and artiodactyls, is also considered. We find it unlikely that a single evolutionary driver was responsible for the evolution of monodactyly, because changes in body size, foot posture, habitat, and substrate are frequently found to influence one another (and to connect to broader potential drivers, such as changing climate). We conclude with suggestions for future research to help untangle the complex dynamics of this remarkable morphological change in extinct horses. A path forward should combine regional paleoecology studies, quantitative biomechanical work, and make use of convergence and modern analogs to estimate the relative contributions of potential evolutionary drivers for digit reduction.

Horses are the only living members of the family Equidae, which today comprises just six species in the genus Equus (including zebras, asses, and caballine horses, the group to which domestic horses belong). In contrast to today’s paucity of species, the equid fossil record includes nearly 50 genera and hundreds of species over the last 58 million years (MacFadden 1994). The earliest equids were only dog-sized, with four toes on the foreleg and three on the hind leg (MacFadden 1994). Today’s horses are large, long-legged grazers with a single toe on each leg, which is enclosed in a hard hoof. An enlarged third digit makes up the bulk of the distal limb, with considerably reduced metapodials II and IV present as splint bones fused to the center metapodial (Fig. 1). Recent work has shown that vestiges of digits I and V may still be present as ridges and wings in the proximal metapodial and distal phalanx (Solounias et al. 2018). Despite their large size and long, slender limbs, horses are considerably athletic, reaching a recorded top racing speed of 70 km/h (“Fastest Speed for a Race Horse” 2019); the highest jump recorded by a domestic horse and rider is 2.47 m (“Highest Jump by a Horse” 2019). That horses can accomplish such feats on a single toe, which evolved millions of years prior to human influence, is remarkable.

The anatomy of modern Equus metapodials; proximal articular views are of the metacarpal (A) and metatarsal (B), and the metacarpal and phalanges are shown in anterior (C) and posterior (D) views. Abbreviations: digits II, III, and IV are shown for the metacarpal and metatarsal; PP is proximal phalanx; MP is medial phalanx; DP is distal phalanx. Sesamoids are indicated with lines.

Fossil horses played a critical role in both supporting Darwin’s theory of evolution and, later, the Modern Synthesis (Simpson 1951). In the 1870s, O.C. Marsh had made a considerable collection of fossil horses, which he then arranged into a series of small to large, three-toed to one toe, low-crowned teeth to high-crowned teeth (Marsh 1874). This proposed evolutionary series was so striking for its time that after seeing it, T.H. Huxley, “Darwin’s Bulldog,” rewrote an address to be given at the New York Academy of Sciences to include these fossil horses as evidence of evolution (Schuchert 1940). At the time, orthogenesis—an evolutionary “progression” in a straight line toward some ideal form—was a popular conception of evolution, and this arrangement of horses supported that view (Fig. 2). Thus the classic story of horse evolution was formed: as grasslands took over from forests, the horse gradually evolved larger body size (perhaps to better defend against predators), taller-crowned teeth to handle abrasive grasses, and long, monodactyl limbs to race away from predators in their newly open habitat (Fig. 2; Matthew 1926).

The linear progression of horses (small to large, many toes to one toe, low-crowned teeth to high-crowned teeth), a view that dominated early narratives about equid evolution. Modified from Matthew (1926).

Despite subsequent recognition that equid evolution was in fact more like a bush than a straight line (Simpson 1951; MacFadden 1994), it is still portrayed in a linear fashion in many museums and textbooks (MacFadden et al. 2012). Some trends in equid evolution do appear to exist by gestalt—today’s horses are indeed much larger, hypsodont, and have reduced digits relative to the earliest horses. Monodactyly had two separate evolutions, one in the Dinohippus/Equus lineage and one in the Pliohippus/Astrohippus lineage, strongly suggesting at least some adaptive utility and selection for this condition. It therefore requires careful attention to discuss the evolution of horses without slipping into verbal orthogenesis by drawing a straight line between the earliest horse and the lone surviving genus today, particularly given that trends of digit reduction and increasing hypsodonty do exist in at least some parts of the horse tree (Janis 2007). But evidence from diet, habitat, tooth morphology, and digit state do not match the orthogenetic pattern: decreasing body size was common in lineages such as Archaeohippus or Nannippus; not all tridactyl horses browsed; and not all hypsodont, monodactyl horses grazed (MacFadden 1994; MacFadden et al. 2012).

Beyond the pattern itself, the classic explanations for why horses evolved the way they did is tremendously “sticky” (Schimel 2012). Long after the complexity of the equid tree and the nonlinearity of trait evolution was acknowledged, the initial explanations for each horse trait still held the weight of established fact rather than reasonable hypothesis. The evolutionary story of horses has seen several advances in understanding over the last few decades, particularly as powerful quantitative methods emerge and we accumulate more available specimens through fieldwork, museum cataloguing, and especially digitization (Marshall et al. 2018). With these new data and methods, untested explanations for horse trait evolution have been challenged one by one.

The simple causal relationship between abrasive grass and hypsodonty has been shown to be complicated, with grasslands predating hypsodonty by at least 4 million years in horses, rodents, and lagomorphs (Strömberg 2002, 2006; Jardine et al. 2012). Tooth mesowear, a macroscopic measure of tooth wear that can record information about diet, is highly variable within fossil horse populations and does not always match up directly with grasslands and hypsodonty (Mihlbachler et al. 2011); furthermore, fresh grazing can cause mesowear similar to browsing (Winkler et al. 2019). In extant taxa, hypsodonty correlates more with habitat openness (Mendoza and Palmqvist 2008) than with the proportion of grass consumed, and feeding height (which relates to the amount of soil grit consumed) drives microwear more than diet (Mainland 2003). The study of tooth enamel isotopes has also complicated the relationship between hypsodonty and diet; for example, in one locality, the hypsodont horse species were likely browsers while the species with low-crowned teeth were consuming more grasses (MacFadden et al. 1999), and individual variation in isotope values can be high in large herbivores (Green et al. 2018). Hypsodonty seems to be driven mostly by grit (via phytolith, dirt, or volcanic ash), not grass alone, and evolved under much less of a straightforward evolutionary arms race than initially thought.

Increasing body size, which was initially explained as a defense mechanism against predation (Matthew 1926), has also been suggested as an example of Cope’s rule (that lineages tend to increase in maximum body size through time) in equids (Martin 2018), perhaps as a result of ecological specialization (Raia et al. 2012). Others have argued that because grass is generally less nutritious than browse (but see Codron et al. 2007), larger body size was beneficial because it increased total digestive capacity, thus allowing the animal to process larger quantities of low-quality food (Demment and Van Soest 1985; Lovegrove and Mowoe 2013). A recent study showed strong evidence for transitions to an unguligrade foot posture being associated with rapid increases in body size, and rate of body size evolution, across mammals (Kubo et al. 2019), pointing to the possibility that unguligrady supports larger body sizes. Kubo et al. (2019) suggested that larger body sizes may provide a release from higher levels of predation, again connecting predator pressures to horse evolution. However, many equid lineages retained similar body sizes through evolutionary time or even became smaller (MacFadden 1994), irrespective of expanding grasslands. Recent work suggested that more than 90% of changes in horse body size can be explained simply by diffusion, a random walk of evolution, rather than competition for niches (Shoemaker et al. 2014), so the pattern of increasing body mass may not be a trend at all.

Like hypsodonty and body size, the story of digit reduction is complicated. The classic explanation was that high speeds were necessary for predator escape in open grasslands, with reduced digits on elongated limbs providing this speed (Matthew 1926; Simpson 1951). But high-speed pursuit predators such as wolves did not evolve until approximately 20 million years after many ungulates, including equids, evolved lengthened limbs and reduced digits (Janis and Wilhelm 1993), providing evidence against this hypothesis. Subsequent years have seen three other major hypotheses about the proximate driving force behind digit reduction in horses:

The locomotor economy hypothesis: Open, arid grasslands required longer travel distances to access patchy resources, such as water, and elongated limbs decreased the energetic cost of locomotion by increasing stride length (Janis and Wilhelm 1993). While reduced digits were not explicitly discussed in the referenced paper, reducing mass in the distal limb would decrease moment of inertia (MOI) and thus the energetic cost of swinging the leg (Hildebrand 1960; Myers and Steudel 1985; Browning et al. 2007; Kilbourne et al. 2016), and has been argued to be a driver of digit reduction in archosaurs (de Bakker et al. 2013).

The stability hypothesis: Forests required lateral dodging movements on soft ground, whereas grasslands required high-speed, straight-line movements on hard ground; a tridactyl and monodactyl foot were, respectively, better suited to stability in those environments (Shotwell 1961). Effectively, Shotwell proposed two separate hypotheses: (2a) the soft substrate stability hypothesis, that the tridactyl foot is adapted for stability in lateral dodges on soft ground, and (2b) the hard substrate stability/speed hypothesis, that the monodactyl foot is adapted for stability and speed in straight lines on hard ground. The superior speed of the monodactyl foot on hard ground was also proposed by Matthew (1926), who focused on the rigid hoof and spring-like tendon–ligament system of extant equids.

The body size hypothesis: Evolutionary increases in body mass produced greater bending forces on the limbs, and a single digit resists bending forces better than several smaller digits of the same total size (Thomason 1986).

None of these hypotheses are mutually exclusive. The first two assume that grasslands act as the ultimate driver of digit reduction (via the proximate causes of increased locomotor demands and changing substrate conditions, respectively). The third hypothesis suggests body mass as a proximate cause. As discussed previously, proposed drivers of body mass itself in horses range from evolutionary diffusion (Shoemaker et al. 2014) to grasslands (Illius and Gordon 1992), consistent with the first two hypotheses. Body mass increase in response to the evolution of grasslands might also partially be related to cooling climate (Lovegrove and Mowoe 2013), but climate cooling should also be considered as one of many potential drivers of grasslands themselves (Strömberg 2011). Therefore, if the body size hypothesis is correct, the ultimate driver of digit reduction could be any or a combination of other changes in forage and climate.

In this review, we survey what evidence exists to support or refute these digit reduction hypotheses. First, we give a brief overview of horse evolution with a focus on digits. Next, we discuss an analytical method for quantifying the degree of digit reduction to provide a continuous metric for evolutionary analyses. We follow this with a review of research from biomechanics, macroevolution, and other subdisciplines that has brought new insights into digit reduction in recent years. Finally, we conclude with a discussion of critical gaps in our understanding and make suggestions for avenues of future study.

The phylogenetic relationships of the earliest equids have seen considerable changes since Hyracotherium/Eohippus was universally considered the first horse (Fig. 3). Recent phylogenetic work has split Hyracotherium into H. leporinum, now considered a basal paleothere (outside of Equidae), and an array of new genera for basal true equids, including Sifrhippus and Arenahippus (Froehlich 2002). Another recent phylogenetic analysis has placed Ghazijhippus (found in Pakistan) and Cymbalophus and Pliolophus (found in Europe) at the base of the equid tree, basal to the North American Sifrhippus and Arenahippus (Bai et al. 2018). Regardless which genus is most basal, the earliest equids were small, dog-sized creatures that had four digits on the forefoot and three on the hind foot (semi-tetradactyl). They had low-crowned (brachydont) teeth indicative of a diet of fruits and soft leaves, and although they seem “primitive” relative to extant horses, a remarkably complete skeleton of the basal horse Arenahippus grangeri (previously H. grangeri; considered a junior synonym of Sifrhippus by Secord et al. [2012], but left separate here) shows that it was fairly derived for the time in its foot posture (subunguligrade) and somewhat elongate metapodials (Kitts 1956; Wood et al. 2011). Although the shoulder and hip joints had considerable range of movement, the distal limb in A. grangeri was already primarily restricted to parasagittal motion, as in later equids (Wood et al. 2011).

A simplified cladogram of horse genera with tapir as an outgroup. Topology after Froehlich (2002), Fraser et al. (2015), Jones (2016), and Bai et al. (2018). Subclades are after Famoso and Davis (2014) and Cantalapiedra et al. (2017), but are frequently paraphyletic (e.g., the Merychippus-Grade Equinae). Note that “Merychippus” is a known polyphyletic group, and here we include only one of several phylogenetic positions for taxa called Merychippus; see Fraser et al. (2015). Digit state is shown by lines (solid black for semi-tetradactyl, thin gray for tridactyl, dotted black for monodactyl). Size is indicated by a circle, scaled based on the base-10 logarithm of body mass (where available). Extant taxa are marked with an asterisk.

Coeval with this Arenahippus was another, similarly small equid, Orohippus, which had even more restricted movements in the distal limb due to more stable carpal and tarsal articulations. Wood et al. (2011) hypothesized that Orohippus may have occupied more open terrain than A. grangeri, which was found in tropical forests—reminiscent of Shotwell’s (1961) stability vs. substrate hypotheses. Wood et al. (2011) suggest the following chain of events: as climate changed in the Paleogene (66–12 Ma), equid diets shifted from high-quality fruits and leaves to lower-quality browse and graze; such a dietary shift drove increases in body size (consistent with results from Secord et al. [2012]) to allow processing greater amounts of food; and finally, increased body size drove a need for more centrally-located, upright limbs (reminiscent of Camp and Smith’s [1942] argument for lineage-scale digit reduction).

Like earlier equids, the three-toed (tridactyl) species Mesohippus, Miohippus, and Anchitherium were also likely subunguligrade, with all distal phalanges contacting the ground and supported by a foot pad (Camp and Smith 1942; Sondaar 1968; Thomason 1986). However, relative to earlier equids, their limbs became more restricted to a pendulum-like motion in the parasagittal plane via limb bone fusion (radius-ulna, tibia-fibula) and changes in joint articulations (Sondaar 1968). Along with increasingly parasagittal motion came the lengthening of the limb, particularly distally. In later lineages, beginning with Parahippus at the base of the grazing radiation, limb elongation continued, and the lateral digits were reduced; the side toes likely did not touch the ground at rest (Sondaar 1968). In the monodactyl or nearly-monodactyl lineages, such as Pliohippus and Equus, a tendon-and-ligament suspensory apparatus and a “springing” foot evolved, with markedly elongate phalanges and considerable elastic energy storage (Biewener 1998) that may have benefited them on hard ground in open grasslands (Matthew 1926; Sondaar 1968; Janis and Wilhelm 1993).

An illustration of the measurements used to calculate TRI. First the lengths of the proximal side phalanges (PPlengthIV and PPlengthII) are averaged, then this value is divided by the length of the proximal center phalanx (PPlengthIII). Illustration modified from Matthew (1926).

With the TRI, we can quantitatively represent the real morphological variation present in equid digits (Fig. 5). Whereas previously all tridactyl horses would be coded the same in categorical data, TRI values range from nearly 1 (all three digits of equal size) to less than 0.3 (side digits one-third the size of the center digit), illuminating variation that was previously unavailable to quantitative analyses. These new data make it possible to address questions such as whether digit reduction correlates with changes in other traits, e.g., hypsodonty or body mass (Parker et al. 2018). Furthermore, TRI can be used in the future to explore digit reduction in a variety of other groups, including artiodactyls, with appropriate modification to account for paraxonic symmetry vs. mesaxonic symmetry (i.e., artiodactyls have symmetrical enlarged digits III and IV with the axis of symmetry running between them, whereas TRI was developed for a single, symmetrical center digit III). A TRI dataset expanded to other taxa would open up the possibility of both more quantitative taxon-specific studies and more broadly comparative studies.

A phylogenetic tree of some horse genera showing discrete categories of digit state (left) vs. TRI, a continuous measure of digit reduction (right). TRI captures considerable variation within tridactyl horses that is missed by discrete categories. Modified from Parker et al. (2018).

As terrestrial quadrupeds, horses primarily use their limbs to interact with the environment through locomotion. The forces that act on bones during an animal’s life can be a powerful source of selection; the geometry of bones can often indicate, for example, locomotor style or other functional uses for that part of the body (Swartz et al. 1992; Anyonge 1996; Doube et al. 2018). In domestic horses, many biomechanical studies have linked skeletal morphology to performance in competition, connecting form to function (Barrey et al. 2002; Gnagey et al. 2006; Weller et al. 2006; Hobbs et al. 2010; Kristjansson et al. 2016). An extra load of just 2.4 kg on a horse’s distal limb has been shown to increase cost of transport by nearly 7% (Wickler et al. 2004), providing a direct connection to the locomotor economy hypothesis of digit reduction.

Studies examining the biomechanical and physiological consequences of limb morphology in extant horses rarely connect to the fossil record and to equid evolution, but a combination of these disciplines offers considerable insight into outstanding questions like the driver of equid digit reduction. Although biomechanical performance data cannot be obtained for extinct animals, musculoskeletal modeling is a powerful tool to reconstruct soft-tissue dynamics, forces, and ultimately provide insight into performance in extinct species (Hutchinson and Garcia 2002; Pierce et al. 2012, 2013; Nyakatura et al. 2019). Such studies usually make use of detailed skeletal data from extinct species, sometimes combining it with experimental biomechanical data on extant taxa. Biomechanical studies can therefore help fill in the relationship between morphology, performance, and the environment, helping to connect patterns evident at the macro-level (morphological and taxonomic change over millions of years) to the individual level (where morphology and performance determine fitness; Fig. 6).

Morphology interacts with the environment to create a given performance, which then (modulated by competition) determines fitness in that environment. Selection acts according to fitness, driving evolutionary change in morphology.

Beam bending is a mechanical engineering approach used to calculate stresses on structural beams. When applied to skeletons, beam bending analyses determine the stresses experienced by a bone using the bone’s own internal geometry, the forces it experiences from muscle contractions and external sources (such as a food item being bitten or the ground contacting a foot), and the angles and moment arms at which those forces act. While frequently used to explore the effects of bite forces in the skull (e.g., Van Valkenburgh and Ruff 1987; Busbey 1995; Therrien 2005), beam bending has also been used to estimate locomotor forces in the limbs of extant and extinct animals (Biewener 1983; Alexander 1985; Blob and Biewener 2001), including in horses (Biewener et al. 1983). In extinct equids, beam bending was first applied to the center metapodial to explore locomotor stresses in Mesohippus (subunguligrade), Merychippus (unguligrade), and modern Equus (unguligrade), using in vivo strain gauge data recorded from metapodials of living horses to ground-truth the method (Thomason 1985). Using broken metacarpals to assess internal geometry and reducing forces in Mesohippus by 50% to account for load-bearing side digits, Thomason (1985) found midshaft metacarpal stresses to be similar in the extinct and extant horse. Stresses were highest in the unguligrade grazer Merychippus, suggesting that size increase alone did not drive the transition from the subunguligrade to the fully unguligrade foot; Thomason (1985) suggested that habitat could have been the major other factor driving this morphological change in the distal limb.

Our results supported and expanded on Thomason’s (1985) work, showing that when the side digits bear load proportional to their size, bone safety factors (ratio of failure stress to peak locomotor stress) were in the range of those found for extant mammals from mice to elephants (2–4; Biewener 1991). In contrast, when the side digits did not reduce the load on the center metapodial, stresses close to or surpassing the tensile fracture stress of bone were reached in taxa as late as Parahippus, indicating that side digits were mechanically critical for resisting stress until at least the beginning of the grazing radiation (Figs. 2 and 4). Furthermore, the center metapodial is positively allometric relative to body mass in its cross-sectional geometry (i.e., resistance to compressive and bending forces), meaning that the center digit compensated for reduced digits through evolutionary time (McHorse et al. 2017). This positive allometry lends some support to the body size hypothesis, that a single large digit is a response to increasing body sizes and can better resist the increased loads. The evolution of unguligrady, which has been connected to increased rates of body size evolution (Kubo et al. 2019), could have spurred these changes indirectly. The allometry results of this study (and the timing of side digits becoming unnecessary for load-bearing) are also consistent with the locomotor economy hypothesis; it is possible that as longer strides and thus longer limbs were favored by selection, the inertial costs of maintaining side toes began to outweigh any remaining stabilizing or load-bearing benefit.

Selection for digit reduction on an evolutionary scale requires morphological changes to influence fitness by changing how the animal performs in its environment (Fig. 6; Arnold 1983). In other instances of digit reduction, new selective pressures frequently come from new ecologies or locomotor modes, such as the cetacean transition into water (Shapiro et al. 2007) or the evolution of ricochetal locomotion coupled with out-in-the-open foraging in jerboas (Moore et al. 2015). In horses, various proximate causes of digit reduction have been suggested, as illustrated by the hypotheses set forth earlier in this paper. Yet the most generally accepted ultimate cause of digit reduction in this group, the one that makes for a new relationship between morphology and fitness, is the evolution of grasslands.

Virtually no macroevolutionary work has explicitly addressed digit reduction in horses, but many studies focus on horse macroevolution more generally. Of particular interest is the Miocene grazing radiation of horses in North America (18–15 Ma), which began with Parahippus—the same genus found to be among the last in which side toes were critical for mechanical support (Figs. 2 and 4; McHorse et al. 2017). Diversification rates were high and at least 19 new species originated quite rapidly, although rates of morphological evolution were not elevated (MacFadden and Hulbert 1988; Cantalapiedra et al. 2017). Although the radiation was suggested to be in response to grasslands, rapid diversification lagged grasslands by several million years (Strömberg 2006; Cantalapiedra et al. 2017). Most speciation events during the Miocene radiation were in fact via dispersal into new regions (Maguire and Stigall 2008). Because dispersal was the main driver of speciation, factors that facilitated movement—such as habitat fragmentation due to tectonic and climatic events—promoted speciation (Stigall 2013). If digit reduction promoted greater economy of locomotion, it could therefore have indirectly supported speciation.

There is no denying the scope of environmental change that accompanied the approximately 58-million-year history of horse evolution. In North America, temperatures swung from the warmth of the Paleocene–Eocene thermal maximum, through a gradual, bumpy cooling spanning the Eocene, Oligocene, and Miocene (periodically interrupted by warmer peaks lasting a few million years), and finally dropped into the cyclical chill of Ice Ages (Fig. 7; Zachos et al. 2001). These thermal changes accompanied precipitation changes, from the wet tropical forest of warmer periods through increasing aridity as the climate cooled (Janis 1993). It is against this climatic backdrop that grasslands evolved in North America, becoming regionally dominant ecosystems approximately 22 million years ago and dominant across North America by 7–11 million years ago (Strömberg 2005, 2006).

Global temperature through time, with significant biotic and abiotic events highlighted. Temperature, climatic event, and ice sheet data from Zachos et al. (2001); equid data from Bai et al. (2018), MacFadden and Hulbert (1988), and Janis (2007), with equid and litoptern data from MacFadden (1994); grassland and hypsodonty data from Strömberg (2005, 2011); pursuit predator data from Janis and Wilhelm (1993). Ages indicated by annotations are approximate, and in many cases (e.g., the spread of grasslands) are ±several million years. Note that oxygen isotope to degrees Celsius relationships are calculated for an ice-free ocean, so temperature estimates are only valid until approximately 35 Ma (Zachos et al. 2001).

As the environmental landscape changed, varied habitats appeared. This variety is the basis of Shotwell’s (1961) hypothesis that tridactyl horses and monodactyl horses were better suited to different habitats—woodland-savanna and grasslands, respectively—and thus partitioned habitats accordingly. Shotwell (1961) tracked the biogeography of the genera Hipparion (tridactyl) and Pliohippus (monodactyl) in the Pliocene, connecting it to patterns of faunal change in the Northern Great Basin of North America. In most faunas, including the Southern Great Basin, the monodactyl grazer Pliohippus is found first in the middle Miocene (Clarendonian, approximately 13.6–10.3 million years ago), but in the Northern Great Basin, it is not found until its immigration in the late Miocene and early Pliocene (Hemphillian, approximately 10.3–4.9 million years ago). Shotwell connected this late appearance of Pliohippus to the coeval spread of semi-arid plains and prairie grasslands from the Southern into the Northern Great Basin, arguing that Pliohippus migration tracks this habitat. Similarly, he suggested that Hipparion tracked woodland-savanna habitat, going locally extinct at the end-Hemphillian as savanna habitats were reduced or eliminated but persisting longer where such habitats remained a major feature of the landscape for a longer time (Shotwell 1961). To further support his claim of partitioning by habitat, Shotwell noted that the relative abundance of hipparionines was not different in regions with or without Pliohippus present, supporting the idea that the genera were not in direct competition.

In a recent study that aimed to investigate Shotwell’s (1961) hypothesis more quantitatively, we tested for niche partitioning among different groups—in this case, tridactyl and monodactyl equids—using site occupancy (Parker et al. 2018). With approximately 3500 fossil horse occurrences that could be assigned to a North American Land Mammal Age and to a paleohabitat (forest/swamp, forest, woodland, woodland-savanna, savanna, grassland-savanna, or grassland), we tested whether tridactyl and monodactyl genera were found in the same habitat type significantly less often than by random chance (which would support habitat partitioning). In fact, overlap between these groups was higher than expected by chance in all North American Land Mammal Ages except the Blancan, where it was indistinguishable from random (Parker et al. 2018). Rather than partitioning by habitat, tridactyl and monodactyl horses were found together more often than by chance. Consequently, at the spatial and temporal scale examined, horses were more similar by shared ancestry than different by digit state, and digit state did not correlate with hypsodonty or body size (Parker et al. 2018). These results suggest that the three classic equid traits did not coevolve under a single, grassland-specific selective regime, but rather were the product of multiple selective pressures that varied across the diverse habitats available to equids in the Miocene and Pliocene. This lack of correlation or habitat partitioning points to the conclusion that whatever drove digit reduction was not identical to the driver of other important equid traits, so the cause of digit reduction was probably multifaceted (because, e.g., grasslands as the driving factor of all three traits would likely lead to correlated evolution among them).

Digit reduction is widespread in tetrapods, including theropod dinosaurs, marsupials, rodents, squamates, and ungulates. Although some forms of digit reduction are arguably related to very different drivers than in equids (e.g., in hopping, bipedal rodents), others may be more closely related. Some of the most ecologically and morphologically convergent examples to horses are in the Artiodactyla, the second major clade of North American ungulates (along with the Perissodactyla, in which Equidae is a family). Many artiodactyl taxa evolved elongated limbs and reduced digits throughout the Cenozoic (Janis and Wilhelm 1993), and they have been characterized as a competitor group to perissodactyls, perhaps partially responsible for the decline of the perissodactyl order by outcompeting them (Illius and Gordon 1992), although a qualitative comparison has found that competition leading to replacement was not at play (Cifelli 1981). As with hypsodonty and limb evolution in horses, parallel arguments have been made in artiodactyls for the evolutionary benefit of hypsodonty (DeMiguel et al. 2014) and the unguligrade distal limb (Clifford 2010). However, recent work has shown that the small herbivorous “condylarths,” which were replaced by artiodactyls and perissodactyls, also had cursorial species, suggesting that cursoriality was not the only driver of success in ungulates (Gould 2017).

Developmental studies have also shown that digit reduction is accomplished via different mechanisms in these two ungulate clades. For example, horses (Perissodactyla), camels (Artiodactyla), and jerboas (Rodentia) form the beginnings of all five digits in early embryonic limb patterning but then show a spike in cell death around the reduced digits in later post-patterning stages (Cooper et al. 2014). In contrast, pigs (Artiodactyla) and cattle (Artiodactyla) developmentally reduce digits via restricted expression of Ptch1, with no noticeable increase in cell death (Sears et al. 2011; Cooper et al. 2014; Lopez-Rios et al. 2014). These results suggest that, developmentally at least, there may be more than one way to evolve reduced digits, encouraging a comparative approach to further determine what similarities and differences exist among convergent evolutions of digit reduction.

In convergence with the foot structure of early horses, two genera of caviomorph rodents, Hydrochaeris and Cavia, have three toes on the hind foot (digits II, III, and IV) and in the forefoot have eliminated digit I, reduced digit V to nonfunctionality, and evolved a digit-III-dominant foot (Rocha-Barbosa et al. 2007). Ground reaction forces and effective mechanical advantage of the limbs have been characterized in capybara (Biewener 2005); they were found to have a less erect posture than goats, and so may be a suitable comparison for basal equids, whose posture was not so erect as extant horses.

Even more striking convergence can be found in the South American Litopterna, an order recently found to be sister to Perissodactyla (Buckley 2015; Welker et al. 2015; Westbury et al. 2017). The litoptern genus Thoatherium evolved a monodactyl condition even more extreme than equids by eliminating even the remnant “splint” metapodials II and IV. However, litopterns and horses show an opposite relationship between digit reduction and body mass. Equids with the most reduced digits are the largest species and, in general, less reduced digits are found on smaller species (McHorse et al. 2017). Conversely, the monodactyl litoptern species are the smallest and those with extremely robust side digits are the largest (Janis 2007). This difference suggests that even if the body size hypothesis remains plausible as a driver of digit reduction in horses, body size is likely not driving digit reduction in litopterns. Despite these divergent body mass patterns, tracing the convergent evolution of digit reduction in litopterns, together with studies of habitat and climate in South America (e.g., Strömberg et al. 2013), has the potential to reveal whether other potential drivers are parallel in the two groups.

The future of studying digit reduction in equids is promising. Here we lay out the steps we believe are necessary to support, or reject, existing hypotheses, as well as new ideas for what may have driven the evolution of such a remarkable trait as monodactyly. It is important to recognize that most of the working hypotheses are not mutually exclusive, and as has become clear from previous work, we should expect interrelationships between the potential causes of digit reduction (e.g., cooler climate may affect body size directly, which is a hypothesized driver of digit reduction, but cooler climate also leads to more open habitats, which is a different hypothesized driver). However, we argue that the way that these selective drivers interact has a significant effect on how we conceptualize the “why” behind digit reduction, and therefore it is a valuable endeavor to uncover the primary driver or drivers of monodactyly and digit reduction as a whole.

If the primary driver of digit reduction was the need for better economy while covering long distances, as hypothesized by Janis and Wilhelm (1993), then digit reduction should virtually always go hand-in-hand with limb elongation. Research manipulating MOI in limbs has shown that decreasing MOI does reduce cost of transport (Martin 1985; Myers and Steudel 1985; Wickler et al. 2004), and added distal limb mass increases cost of transport in extant horses (Wickler et al. 2004), so some degree of energetic savings is almost certainly a consequence of digit reduction. The logical next step is a theoretical exploration of the magnitude of energetic savings from (1) reduction of MOI at the distal limb due to loss of digits and (2) elongation of the limb over (3) a range of body sizes and taxa. The results from these calculations could be used to create a theoretical cost of transport morphospace that connects changes in limb length, relative differences in segment MOI, estimated limb swing frequency, and body mass to a resulting cost of transport. That morphospace could be used to calculate, e.g., how much locomotor economy was improved between different taxa and whether the difference gained by digit reduction constitutes a significant energetic savings. To evaluate significance of energetic savings, it would be necessary to relate the results of the calculations to known costs associated with swinging legs in living animals (e.g., Fedak et al. 1982). Tracking these changes through time would further allow testing of whether quantitative improvements in locomotor economy over evolutionary time tracked (or slightly lagged) aridification and the spread of grasslands. This analysis is ideally suited to include artiodactyls as a comparative group, because they also evolved longer limbs, reduced digits, and would have experienced the same pressures at the same times where they overlapped spatially with horses.

The idea that a tridactyl foot is more stable for lateral dodging on soft substrate, whereas a more rigid single hoof is more stable for and provides faster straight-line locomotion on hard substrate (Shotwell 1961), remains untested. The soft substrate hypothesis is concerned primarily with stability, whereas the hard substrate may have considerably more to do with elastic energy storage and energy dissipation. Whether theoretical or practical (as in biorobotics), this hypothesis requires a test of whether tridactyl feet are indeed more stable on softer substrates. Examination of locomotor performance in extant taxa with reduced digits would provide an interesting first step, which could be complemented by studies using modeling and simulation to explore the effect of digit number on stability in the equid distal limb; a combination of simulation and biorobotics would offer considerable flexibility (e.g., Nyakatura et al. 2019). Biomechanical modeling work such as this can be more powerful when ground-truthed with locomotion data, such as speed, joint kinematics, and forces from extant animals—a challenge when the majority of equid diversity is extinct. Monodactyl equids are straightforward to study in that domestic horses are anatomically extremely similar, particularly in the distal limb, to both wild Equus and to other extinct monodactyl taxa.

Several taxa are potential modern analogs for tridactyl or semi-tetradactyl horses, and would therefore be suitable for study of locomotor biomechanics (directly testing dodging stability on softer substrates), for ground-truthing the proposed simulation and biorobotics work, and for comparative studies of distal limb anatomy (including internal geometry, i.e., resistance to bending, torsion, and compressive forces). Tapirs (Tapiridae) are one of three extant families in the perissodactyl order, the others being equids and rhinoceroses, and are similar to basal equids in their digit state: four digits on the front leg and three in the back. Tapirs may therefore offer a convenient semi-tetradactyl species for locomotion studies of substrate-based stability; they have been used for comparative anatomical and biomechanical studies (McHorse et al. 2017), and we have collected kinetics and kinematics data from Baird’s tapir (Tapirus bairdii) that will provide locomotor forces to scale for future finite element modeling work. A recent study has argued that the semiaquatic tapir benefits from lateral splaying in phalanges II and IV, which could allow for greater stability on soft, muddy surfaces beneath water (Endo et al. 2019), echoing the substrate stability hypothesis itself.

While tapirs provide the closest phylogenetic match for a study of digits and stability, they may not provide the best biomechanical one. In contrast to equids, tapirs are semiaquatic (Nowak and Paradiso 1999), and although some tapir species may provide morphological analogs for other basal perissodactyls such as palaeotheres (MacLaren and Nauwelaerts 2019), the range of body sizes differs considerably in the two families. Most basal equids had a body mass between 5 and 20 kg (MacFadden 1986; Secord et al. 2012), but even the smallest extant tapir species have body masses over 130 kg (Nowak and Paradiso 1999). Therefore, caviomorph rodents and small artiodactyls may provide a closer biomechanical analog in terms of the forces generated during locomotion. The two genera of caviomorph rodents with remarkable convergence toward the equid foot condition are much closer in size: species tend to average 30–60 kg and 1 kg in Hydrochaeris and Cavia, respectively (Biewener 2005; Ferraz et al. 2005). Similarly, some species of artiodactyl are quite small (e.g., tragulids and moschids are generally 1000 specimens, Carrasco et al. 2005). The phytoliths (grass species indicators) and paleosols (C3 vs. C4 grass indicators) are also well-characterized in the area (Fig. 8; Fox and Koch 2003; Strömberg 2005, 2011). Other potential candidate regions include the John Day region of Oregon and the state of Florida, both with remarkable fossil records of horse evolution and extensive research into climatic and habitat change through time (Stock 1946; Macfadden and Cerling 1996; MacFadden et al. 1999; Retallack 2004; Maguire and Stigall 2008; Maguire 2015). However, there are tradeoffs to choosing regional-scale studies: in exchange for better-controlled data and more power to detect trends locally, one gives up some amount of power to explain global trends. In other words, trends at one scale cannot necessarily be extrapolated to others—a critical challenge of macroevolutionary studies in general (Jablonski 2008).

Equid occurrences (orange, Paleocene, Eocene, and Oligocene; yellow, Miocene, Pliocene, and Pleistocene) across a section of North America. Size of the circle is scaled to number of occurrences. The densely sampled Great Plains region is highlighted in pale orange. Sites characterized for C3 vs. C4 grasses (Fox and Koch 2003) are shown by magenta circles; sites characterized for grassland indicator phytoliths (Strömberg 2005, 2011) are shown by blue and purple circles.

Although qualitatively the trend appears to point toward monodactyly being “optimal,” at least in grasslands, this pattern has yet to be quantitatively tested—and as we know from work showing that body size evolution was likely not directional in horses, apparent trends can be deceiving. Phylogenetic comparative methods offer a way to explicitly test the evolutionary mode of trends like these. Evolutionary model-fitting can compare the fit of models such as Brownian Motion (a random walk), an Ornstein–Uhlenbeck (OU) process (a model where the trait is being pulled with some strength toward an adaptive peak of some “optimal” value), or a multi-peak OU model, which allows for multiple optima that may correspond to another feature such as habitat (Hansen 1997; Butler and King 2004). In an in-progress study, we explicitly test digit reduction in this framework, investigating whether digit reduction is pulled to some adaptive optimum for all of equids (i.e., some degree of digit reduction is “optimal”) or whether that optimum varies based on habitat type (e.g., forest-dwelling species are pulled toward some moderate value of TRI whereas grassland dwellers are pulled toward monodactyly). Alternatively, if different habitats drive different rates of digit evolution but there is no trait optimum, a multi-rate model would be more appropriate (Collar et al. 2010). A study such as this would also be suited to a more broadly comparative context, evaluating whether the mode of digit reduction evolution is similar in other taxa (e.g., artiodactyls or litopterns).

The evolution of monodactyly in horses is remarkable and is unique among extant animals, but fortunately for scientists, the themes of digit reduction, habitat change, and body size change are repeated many times in the fossil record. Reviewing the available evidence makes it clear that we are unlikely to find a single evolutionary driver to be solely responsible for the evolution of monodactyly, because open habitat, changes in substrate, changes in foot posture, and changes in body size can all tie to one another and to broader ecological drivers such as changing climate. However, we argue that by combining finer-scale regional studies, quantitative biomechanical studies, and careful analysis of convergent clades, it will be possible to estimate the relative contributions of these evolutionary drivers. Even if digit reduction is ultimately not driven by the same factors in each clade (as may be the case with horses vs. litopterns), such a discovery would be a considerable leap forward in our understanding of how—and why—horses evolved a single toe.

From the symposium “Comparative Evolutionary Morphology and Biomechanics in the Era of Big Data” presented at the annual meeting of the Society for Integrative and Comparative Biology, January 3–7, 2019 at Tampa, Florida.

The authors would like to thank Zachary Morris and other members of the Pierce and Biewener laboratories for productive discussion; Talia Moore for help conceptualizing Fig. 6; Samantha Hopkins and Edward Davis, who first encouraged the equid evolutionary line of thinking and have provided ongoing thoughts; and Abigail Parker and Tristan Reinecke, whose work contributed to ideas mentioned in this review. Hayley O’Brien suggested looking into caviomorph rodents. Zhijie Jack Tseng and an anonymous reviewer provided helpful comments that improved the manuscript. Finally, the authors would like to thank Samantha Price and Martha Muñoz, who organized the symposium on Biomechanics in the Era of Big Data, to which this paper is a contribution.

This work was supported by the National Science Foundation [DGE-1144152 to B.K.M and DEB-1701656 to B.K.M and S.E.P].

Alexander RM.

Google Scholar

WorldCat

Anyonge W.

Google Scholar

WorldCat

Arnold SJ.

Google Scholar

WorldCat

Bai B, Wang Y-Q, Meng J.

Google Scholar

WorldCat

Barrey E, Desliens F, Poirel D, Biau S, Lemaire S, Rivero JLL, Langlois B.

Google Scholar

WorldCat

Biewener AA.

Google Scholar

WorldCat

Biewener AA.

Google Scholar

WorldCat

Biewener AA.

Google Scholar

WorldCat

Biewener AA.

Google Scholar

WorldCat

Biewener AA, Thomason J, Goodship A, Lanyon LE.

Google Scholar

WorldCat

Blob RW, Biewener AA.

Google Scholar

WorldCat

Browning RC, Modica JR, Kram R, Goswami A.

Google Scholar

WorldCat

Buckley M.

Google Scholar

WorldCat

Busbey AB.

Google Scholar

Google Preview

WorldCat

Butler MA, King AA.

Google Scholar

WorldCat

Camp CL, Smith N.

Google Scholar

WorldCat

Cantalapiedra JL, Prado JL, Fernández MH, Alberdi MT.

Google Scholar

WorldCat

Carrasco MA, Kraatz BP, Davis EB, Barnosky AD.

Google Scholar

Google Preview

WorldCat

Chen ST, Smith SY, Sheldon ND, Strömberg CA.

Google Scholar

WorldCat

Cifelli RL.

Google Scholar

WorldCat

Clifford AB.

Google Scholar

WorldCat

Codron D, Lee-Thorp JA, Sponheimer M, Codron J.

Google Scholar

WorldCat

Collar DC, Schulte IIJ, Losos JB.

Google Scholar

WorldCat

Collar DC, Schulte JA, O’Meara BC, Losos JB.

Google Scholar

WorldCat

Cooper KL, Sears KE, Uygur A, Maier J, Baczkowski K-S, Brosnahan M, Antczak D, Skidmore JA, Tabin CJ.

Google Scholar

WorldCat

de Bakker MAG, Fowler DA, den Oude K, Dondorp EM, Navas MCG, Horbanczuk JO, Sire J-Y, Szczerbińska D, Richardson MK.

Google Scholar

WorldCat

DeMiguel D, Azanza B, Morales J.

Google Scholar

WorldCat

Demment MW, Van Soest PJ.

Google Scholar

WorldCat

Doube M, Felder AA, Chua MY, Lodhia K, Kłosowski MM, Hutchinson JR, Shefelbine SJ.

Google Scholar

WorldCat

Endo H, Yoshida M, Nguyen TS, Akiba Y, Takeda M, Kudo K.

Google Scholar

WorldCat

Famoso NA, Davis EB.

Google Scholar

WorldCat

Fedak MA, Heglund NC, Taylor CR.

Google Scholar

WorldCat

Ferraz KMPMB, Bonach K, Verdade LM.

Google Scholar

WorldCat

Fox DL, Koch PL.

Google Scholar

WorldCat

Fraser D, Gorelick R, Rybczynski N.

Google Scholar

WorldCat

Froehlich DJ.

Google Scholar

WorldCat

Gnagey L, Clayton HM, Lanovaz JL.

Google Scholar

WorldCat

Gould F.

Google Scholar

WorldCat

Green DR, Olack G, Colman AS.

Google Scholar

WorldCat

Hansen TF.

Google Scholar

WorldCat

Hildebrand M.

Google Scholar

WorldCat

Hobbs SJ, Levine D, Richards J, Clayton H, Tate J, Walker R.

Google Scholar

WorldCat

Hutchinson JR, Garcia M.

Google Scholar

WorldCat

Illius AW, Gordon IJ.

Google Scholar

WorldCat

Jablonski D.

Google Scholar

WorldCat

Janis CM.

Google Scholar

WorldCat

Janis CM.

Google Scholar

Google Preview

WorldCat

Janis CM, Wilhelm PB.

Google Scholar

WorldCat

Jardine PE, Janis CM, Sahney S, Benton MJ.

Google Scholar

WorldCat

Jones K.

Google Scholar

WorldCat

Kilbourne BM, Andrada E, Fischer MS, Nyakatura JA.

Google Scholar

WorldCat

Kitts DB.

Google Scholar

WorldCat

Kristjansson T, Bjornsdottir S, Albertsdóttir E, Sigurdsson A, Pourcelot P, Crevier-Denoix N, Arnason T.

Google Scholar

WorldCat

Kubo T, Sakamoto M, Meade A, Venditti C.

Google Scholar

WorldCat

Lopez-Rios J, Duchesne A, Speziale D, Andrey G, Peterson KA, Germann P, Ünal E, Liu J, Floriot S, Barbey S, et al. .

Google Scholar

WorldCat

Losos JB.

Google Scholar

Google Preview

WorldCat

Lovegrove BG, Mowoe MO.

Google Scholar

WorldCat

MacFadden BJ.

Google Scholar

WorldCat

MacFadden BJ.

Google Scholar

Google Preview

WorldCat

Macfadden BJ, Cerling TE.

Google Scholar

WorldCat

MacFadden BJ, Hulbert RC.

Google Scholar

WorldCat

MacFadden BJ, Oviedo LH, Seymour GM, Ellis S.

Google Scholar

WorldCat

MacFadden BJ, Solounias N, Cerling TE.

Google Scholar

WorldCat

MacLaren JA, Nauwelaerts S.

Google Scholar

WorldCat

Maguire KC.

Google Scholar

WorldCat

Maguire KC, Stigall AL.

Google Scholar

WorldCat

Mainland IL.

Google Scholar

WorldCat

Marsh OC.

Google Scholar

WorldCat

Marshall CR, Finnegan S, Clites EC, Holroyd PA, Bonuso N, Cortez C, Davis E, Dietl GP, Druckenmiller PS, Eng RC, et al. .

Google Scholar

WorldCat

Martin PE.

Google Scholar

WorldCat

Martin RA.

Google Scholar

WorldCat

Matthew WD.

Google Scholar

WorldCat

McHorse BK, Biewener AA, Pierce SE.

Google Scholar

WorldCat

Mendoza M, Palmqvist P.

Google Scholar

WorldCat

Mihlbachler MC, Rivals F, Solounias N, Semprebon GM.

Google Scholar

WorldCat

Moore TY, Organ CL, Edwards SV, Biewener AA, Tabin CJ, Jenkins FAJr,, Cooper KL.

Google Scholar

WorldCat

Myers MJ, Steudel K.

Google Scholar

WorldCat

Nowak RM, Paradiso JL. 1999. Walker's Mammals of the World. 6th ed Baltimore: Johns Hopkins University Press.

Nyakatura JA, Melo K, Horvat T, Karakasiliotis K, Allen VR, Andikfar A, Andrada E, Arnold P, Lauströer J, Hutchinson JR, et al. .

Google Scholar

WorldCat

O’Meara BC, Beaulieu JM.

Google Scholar

Google Preview

WorldCat

Parker AK, McHorse BK, Pierce SE.

Google Scholar

WorldCat

Pierce SE, Clack JA, Hutchinson JR.

Google Scholar

WorldCat

Pierce SE, Hutchinson JR, Clack JA.

Price SA, Holzman R, Near TJ, Wainwright PC.

Google Scholar

WorldCat

Raia P, Carotenuto F, Passaro F, Fulgione D, Fortelius M.

Google Scholar

WorldCat

Retallack GJ.

Google Scholar

WorldCat

Rocha-Barbosa O, Loguercio MFC, Renous S, Gasc J-P.

Google Scholar

WorldCat

Schimel J.

Google Scholar

Google Preview

WorldCat

Schuchert C.

Google Scholar

Google Preview

WorldCat

Sears KE, Bormet AK, Rockwell A, Powers LE, Cooper LN, Wheeler MB.

Google Scholar

WorldCat

Secord R, Bloch JI, Chester SG, Boyer DM, Wood AR, Wing SL, Kraus MJ, McInerney FA, Krigbaum J.

Google Scholar

WorldCat

Shapiro MD, Shubin NH, Downs JP. 2007. Limb diversity and digit reduction in reptilian evolution. In: Hall BK, editor. Fins into limbs: evolution, development, and transformation. Chicago and London: University of Chicago Press. p. 225–245.

Shoemaker L, Clauset A, Hector A.

Google Scholar

WorldCat

Shotwell JA.

Google Scholar

WorldCat

Simpson GG. 1951. Horses: the story of the horse family in the modern world and through sixty million years of history. New York: Oxford University Press.

Solounias N, Danowitz M, Stachtiaris E, Khurana A, Araim M, Sayegh M, Natale J.

Google Scholar

WorldCat

Sondaar PY.

Google Scholar

WorldCat

Stigall AL.

Google Scholar

WorldCat

Stock C.

Google Scholar

WorldCat

Strömberg CA.

Google Scholar

WorldCat

Strömberg CA.

Google Scholar

WorldCat

Strömberg CA.

Google Scholar

WorldCat

Strömberg CAE.

Google Scholar

WorldCat

Strömberg CAE, Dunn RE, Madden RH, Kohn MJ, Carlini AA.

Google Scholar

WorldCat

Swartz SM, Bennett MB, Carrier DR.

Google Scholar

WorldCat

Therrien F.

Google Scholar

WorldCat

Thomason JJ.

Google Scholar

WorldCat

Thomason JJ.

Google Scholar

WorldCat

Van Valkenburgh B, Ruff CB.

Google Scholar

WorldCat

Welker F, Collins MJ, Thomas JA, Wadsley M, Brace S, Cappellini E, Turvey ST, Reguero M, Gelfo JN, Kramarz A, et al. .

Google Scholar

WorldCat

Weller R, Pfau T, Verheyen K, May SA, Wilson AM.

Google Scholar

WorldCat

Westbury M, Baleka S, Barlow A, Hartmann S, Paijmans JLA, Kramarz A, Forasiepi AM, Bond M, Gelfo JN, Reguero MA, et al. .

Google Scholar

WorldCat

Wickler SJ, Hoyt DF, Clayton HM, Mullineaux DR, Cogger EA, Sandoval E, McGuire R, Lopez C.

Google Scholar

WorldCat

Winkler DE, Schulz-Kornas E, Kaiser TM, Cuyper AD, Clauss M, Tütken T.

Google Scholar

WorldCat

Wood AR, Bebej RM, Manz CL, Begun DL, Gingerich PD.

Google Scholar

WorldCat

Zachos J, Pagani M, Sloan L, Thomas E, Billups K.

Google Scholar

WorldCat

Figure 5: Reprinted from Palaeogeography, Palaeoclimatology, Palaeoecology, Vol 511, Abigail K. Parker, Brianna K. McHorse, Stephanie E. Pierce, Niche modeling reveals lack of broad-scale habitat partitioning in extinct horses of North America, Pages 103–118, Copyright (2018), with permission from Elsevier.