Ideally, paleohabitats should be inferred based on fossil plants: but in many cases (and especially in older studies) accompanying paleobotanical information is not available for paleofaunal assemblages, and the types of animals present in the fauna have been used to make inferences about the faunal habitat. Quantitative studies have shown ungulates can certainly serve as reliable environmental proxies (e.g., Eronen et al., 2012). Savanna habitats have been inferred for faunas that, like modern savannas, contain a diversity of large ungulates comprising a mixture of browsers, grazers and mixed feeders, and also evidencing cursorial limb morphologies (see Webb, 1977). Morphological proxies are necessary to determine the probable diets and locomotor styles of extinct ungulates, especially if comparisons between fossil and recent ecosystems are to be made in a quantitative fashion. We discuss such proxies in more detail later when introducing the ones used in this study, but here we review a few of the issues.

The traditional indication of a grassland paleohabitat has been the relative height of the cheek teeth, or the “hypsodonty index,” of the ungulates in the fauna; animals with a high index have been assumed to be grazers, thus evidencing the presence of grasses. Grazing ungulates today (e.g., equids, many bovids) are all highly hypsodont: high-crowned teeth are more resistant to a lifetime of wear, and phytoliths in grasses have been presumed to be the agents of dental abrasion. However, more recent studies have shown that the actual food eaten by hypsodont ungulates may not be the major agent of dental wear, but rather the dust on the food and accidentally ingested soil (see Damuth and Janis, 2011, for review). Hypsodonty is indicative of an abrasive diet consumed in an open habitat, which is possibly but not necessarily a diet of grass. The average hypsodonty index of modern ungulate faunas has been shown to correlate with annual precipitation, which in turn correlates with habitat openness (Eronen et al., 2012).

A number of morphological features have been shown to functionally correlate with diet and locomotion in extant ungulates, and these may be used as proxies to determine the behavior of extinct ungulates; however, such proxies may not be entirely “taxon free” (as per Damuth, 1982), as is often assumed, and phylogenetic influences may have a profound effect (see Barr, 2018, and Kovarovic et al., 2018, for recent reviews of ecomorphological proxies, phylogenetic issues, and the use of such proxies in habitat reconstruction). Phylogenetic issues relating to this study are considered in more detail below.

Several craniodental proxies have been proposed for determining diets of extinct ungulates (see Fraser and Theodor, 2011). Hypsodonty index (which can only be estimated on unworn teeth) is a primary proxy; other dental proxies include various aspects of tooth wear, such as mesowear (i.e., lifetime wear as evidenced by abrasion and attrition; Fortelius and Solounias, 2000), and microwear (i.e., quantification of microscopic scratches and pits produced by the last few meals of the animal; Solounias and Semprebon, 2002). Because dental wear is a phenotypic phenomenon it is usually assumed to be free of phylogenetic influence (but see Fraser et al., 2018). Hypsodonty index is somewhat influenced by phylogeny (e.g., equids are more hypsodont than bovids of similar grazing diet; Janis, 1988), but categories of hypsodonty (e.g., brachydont, mesodont, hypsodont; Fortelius et al., 2002) correlate fairly well with diet and habitat across extant ungulates (Damuth and Janis, 2011; Eronen et al., 2012).

Cranial dietary proxies include the shape of the muzzle (e.g., narrower muzzles are indicative of more selective feeding; Janis and Ehrhardt, 1988; Dompierre and Churcher, 1996), and those describing the size of the jaw adductor muscles, which are in turn related to the masticatory forces required to orally process browse or grass (e.g., animals with a more fibrous diet have a larger masseter muscle, and hence a larger angle of the dentary: Solounias et al., 1995; Clauss et al., 2008). Likewise, several proxies have been proposed for determining the locomotor type of extinct ungulates. Ungulates in more open habitats have a more “cursorial” type of limb morphology, including relatively longer distal limb segments, and their limbs are more restricted in motion to the parasagittal plane. Specific postcranial proxies include ones related to the morphology of the femur (e.g., Kappelman, 1988; Kappelman et al., 1997; Janis et al., 2012), metapodials (e.g., Scott and Barr, 2014) and phalanges (e.g., Luoys et al., 2013).

Ungulate ecomorphological proxies have primarily been determined in African bovids: this assemblage offers a diversity of body sizes, diets and locomotor types, and these modern faunas have been of especial interest for inferring relatively recent (Plio-Pleistocene) African paleohabitats in connection with studies of human evolution (e.g., Kappelman et al., 1997; Plummer et al., 2015; Lehmann et al., 2016). But a problem in considering other ungulate families (e.g., equids) is that there is no comparable diversity of extant forms that can be used to establish ecomorphological proxies. In general, other ruminants can be compared with bovids (although bovids tend to be more hypsodont than cervids of similar diet; see Janis, 1988), but bovid-based proxies cannot be used to quantitatively determine the ecomorphology of suoid artiodactyls; suoids differ greatly from ruminants in their craniodental anatomy, and they have relatively shorter limbs. Such phylogenetic issues also arise when considering the other non-ruminant large herbivorous mammals of the North American Miocene savannas. Proboscideans and rhinos are very different to bovids in their craniodental and skeletal anatomy, as well as being much larger, and their ecomorphology obviously cannot be determined from bovid-based proxies. However, both camelids and equids are somewhat similar to bovids in overall size and general appearance, and we contrast these taxa below.

Equids were the most common and diverse ungulates in the North American savannas, comparable in their diversity of sizes and diets to African antelope today (see Webb, 1977). We attempted to include equids in this study but discovered that, while the hypsodont ones (subfamily Equinae) more or less clustered with the grazing bovids (as did the zebra, Equus burchelli, in the Serengeti fauna), the brachydont ones (subfamily Anchitheriinae, a little-appreciated yet important element of the North American savanna faunas until the latest Miocene) clustered in their own portion of the morphospace (Appendix 1), far removed from the browsing bovids that they resemble in hypsodonty and dental wear (see Semprebon et al., 2019). Although equids superficially resemble ruminants in their long faces and relatively elongated distal limb proportions, they differ in some critical ways. Equids, for example, being hindgut fermenters, have relatively larger masseter muscles than ruminants of similar diets, and so have a differently shaped mandible. This is not only apparent in extant animals, but the brachydont (presumed browsing) anchitheriine equids also have a relatively deep mandibular angle, very unlike browsing bovids. Regrettably, we found no way to adjust the parameters of our morphological proxies in such a way that the anchitheriine equids could be compared with bovids that they resembled ecologically and so we decided to limit our study to the artiodactyls.

The camelids proved to be more comparable to ruminants. Camelids have commonly been considered as the sister taxon to the ruminants based on craniodental characters (the grouping of the two forming the “Neoselenodonta” of Webb and Taylor, 1980), although many molecular studies show camelids to be less closely related to the ruminants than are suoids (e.g., Price et al., 2005; but see Zurano et al., 2019, which places camelids as the sister taxon to ruminants plus cetaceans). However, camelids resemble ruminants in having selenodont molars, relatively long limbs (albeit with a different, secondarily digitigrade, foot posture), and a generally similar cranial shape, although they are certainly not morphologically identical (e.g., camelids have a relatively shorter premolar row, and tend to have proportionally longer distal limb segments). The North American camelids mostly cluster in a different portion of the morphospace to the African ruminants, and such morphological differences might be considered to represent phylogeny rather than ecomorphology. However, extant camelids overlap with ruminants in the morphospace (see Figure 6B). We discuss extensively in the text why we consider that, although phylogeny may play a role, the position of camelids in our morphospace may indeed carry a genuine ecomorphological signal, indicative of habitat and ecosystem differences between the North American Neogene and present-day Africa.

Traditionally, the presence of paleosavannas in fossil ecosystems was determined using mammalian assemblages because little paleobotanical evidence was available. Webb (1977) suggested that the occurrence of large mammals with hypsodont cheek teeth and cursorial limb morphologies indicated the presence of savannas, as demonstrated especially in the middle Miocene of North America by the diversification of horses in the subfamily Equinae.

While macrofossils of plants are rare in paleofaunal assemblages, microfossils in the form of phytoliths may be plentiful; the research of Strömberg (2002, 2004, 2011) in the past couple of decades shows that phytoliths can provide direct evidence about the floral assemblage, and hence the habitat. For example, Strömberg et al. (2013) found evidence of hypsodonty predating the presence of open environments in South America, whereas hypsodonty postdated the origin of grasslands in North America (Strömberg, 2002). Strömberg et al. (2013) concluded that there is no simple correlation between hypsodonty and grassy and/or open habitats: rather, abrasive diets are the important factor, and such diets may be independent of open habitat environments. Therefore, one should be cautious when trying to reconstruct paleoenvironments using the hypsodonty of mammalian assemblages as the sole line of evidence. Paleobotanical evidence, like phytolith assemblages (e.g., Strömberg, 2002, 2004, 2011; Strömberg et al., 2013), and paleosols (e.g., Fox and Koch, 2004) should also be considered whenever possible.

Figure 1 highlights the different perspectives regarding the evolution of ecosystems in North America during the Cenozoic. The older hypothesis (Webb, 1977) proposed that changes in the morphology of ungulates throughout the Cenozoic are indicative of changing ecosystems. Based on the presence of relatively large vertebrates in the Rocky Mountain Region during the late Paleocene (e.g., sheep to pig-sized phenacodontid condylarths and tapir-sized ungulate-like mammals such as pantodonts), Webb (1977) postulated the existence of “protosavannas” – small patches of open country amid predominant sub-tropical forests covering most of the continent. Webb (1977) also suggested that more open habitats first appeared in North America as early as the middle Eocene, based on botanical and sedimentological evidence from the Rocky Mountain Region that points to an increase in seasonal aridity, as well as what were then the oldest known grasses. Further evidence stemming from mammalian assemblages includes the decline of arboreal species (especially primates) and the development of “special adaptations” in some taxa to open-habitat living (e.g., development of more lophodont dentitions and more elongated limbs). Additionally, he highlighted the immigration of relatively open-habitat lineages from the Old World into North America during the late Eocene, including camels, hornless ruminants, and rabbits.

Webb (1977) proposed that open woodlands were expanding by the Oligocene, as evidenced by the diversification of mammalian taxa, the presence of mesodont and hypsodont teeth in some species (e.g., the diminutive hypertragulid Hypisodus), and the existence of highly specialized fossorial insectivorans (Eulipotyphla) such as Proterix and Cryptoryctes. By the middle Miocene, the high diversity of ungulates in North America was indicative of a savanna ecosystem fully comparable to that of the present-day African savannas: the fauna included rhinos, proboscideans, and hypsodont ruminants and equids, and many Miocene ungulate taxa had open-habitat adaptations, both dental (i.e., hypsodonty) and locomotor (e.g., long distal limb segments). Nonetheless, a considerable diversity of brachydont, shorter-legged ungulates persisted well into the late Miocene, which Webb (1977) attributed to the presence of riparian forests interdigitating with the predominant woodland savanna. By the Pliocene, the replacement of woodland savanna by steppe in the Great Plains was evidenced by the near absence of browsing species, the decrease in diversity of hypsodont species, and the concomitant decrease in the overall diversity of ungulates (Webb, 1977).

A more recent hypothesis, proposed by Strömberg (2011), is based on paleobotanical evidence, including macrofossils and phytoliths, as well as paleosols (Figure 1) [see also earlier discussion of North American grassland evolution by Jacobs et al. (1999) and Retallack (1997)]. Macrofossil evidence and palynofloras suggest the presence of subtropical-tropical evergreen or semideciduous forests across North America for most of the Eocene, although the oldest grass macrofossils on this continent date back to the earliest Eocene (55 Ma) (Strömberg, 2011). Woodland savanna-like ecosystems developed in the middle Eocene and persisted into the early Miocene; however, these ecosystems were not dominated by grasses, but rather by small trees and shrubs (Strömberg, 2011). Macrofossil evidence suggests that it was not until the middle to late Miocene that grasses spread across North America (Strömberg, 2011). Phytolith assemblages tell a different story, with grasses being present in forests by the earliest Oligocene and woodland savannas dominated by pooid grasses by the late Oligocene-early Miocene (Strömberg, 2011), preceding the radiation of hypsodont horses by around five million years. Phytolith evidence indicates that open grasslands first appeared in North America by the latest Miocene (Strömberg, 2011). Paleosol data indicate the presence of open grassland to woodland environments between the late Eocene and the early Miocene, and the development of drier habitats by the early Miocene (Strömberg, 2011); however, paleosols indicative of more open (i.e., more arid) grasslands are not apparent in the North American Great Plains area until the latest Miocene (around 6 Ma) (Retallack, 1997). In addition, changes in Neogene soil oxygen isotopes reflect the shift in the hydrological cycle that was effected by the transition to grasslands from environments with abundant tree cover: the late Neogene expansion of grasslands had a feedback effect in increasing environmental patterns of seasonality and aridity, and grassland expansion in general has been a powerful driver of global environmental change (Mix et al., 2013; Chamberlain et al., 2014).

There has been much discussion over the past few decades about the point in time in which Neogene grasslands transitioned from C₃ photosynthesis to C₄ photosynthesis (generally held to have happened in the latest Miocene, around 6–8 Mya), and the role of declining atmospheric CO₂ in forcing this change (see Fox et al., 2018, for review). However, it is not clear how much this transition affected the grazing ungulates: certainly, as noted above, there is abundant evidence for C₃ grasslands, and grazers feeding on them, in the middle to early late Miocene, and a transition in mode of plant synthesis may not have had a significant effect on herbivore foraging.

Of special interest here are the woodland savannas and grass-dominated habitats present in North America during the Neogene (Miocene and Pliocene) and, in particular, the notion that the ungulate assemblages from the Miocene savannas of North America were fully analogous in their ecomorphology to those of the present-day African savannas as proposed by Webb (1977, 1983b). In terms of the taxonomic composition of the large herbivorous mammal fauna, modern African savannas are dominated by bovid ruminants; other artiodactyl taxa such as giraffes, suids (pigs) and hippopotamuses are also present, along with perissodactyls like equids (zebras) and rhinos, and proboscideans (elephants) (Shorrocks and Bates, 2015). The Neogene savannas of North America were also dominated by artiodactyls, and both perissodactyls (equids and rhinos) and proboscideans (gomphotheres and mastodons) were also present. However, the artiodactyl taxonomic composition differed from that seen in present-day Africa and included tayassuids (peccaries), camelids (and the extinct, possibly camel-related, oreodonts and protoceratids), and non-bovid ruminants such as dromomerycids, antilocaprids (pronghorns), and a diversity of small hornless taxa (Janis et al., 2004); see later detailed description of the taxa present. The extant diversity of North American cervids and bovids arose from two separate immigration events from Eurasia, during the Pliocene and Pleistocene (Webb, 1998a). Unlike the fauna of modern African savannas, which contains a single genus of equids (and rarely more than one sympatric species in any faunal assemblage), equids were prominent members of the Neogene North American ecosystems, with over 20 genera present in the Miocene and Pliocene, and up to eight sympatric species in paleoassemblages (MacFadden, 1998).

Despite the differences in taxonomic composition, many of the extinct ungulates from the North American savanna paleofaunas have been proposed as ecomorphological analogs of modern-day African ungulates: e.g., the “giraffe-like camel” Aepycamelus, the “gazelle-like camel” Stenomylus, and the “hippo-like” teleoceratinae rhinos (Webb, 1983b). While both extant and extinct faunas appear to contain a similar diversity of ungulate species and feeding categories, this notion of savanna faunal similarity has never been examined quantitatively. In this paper, we analyze the morphospace occupation of artiodactyls in the present-day Serengeti National Park and selected Neogene faunal assemblages from the North American Great Plains, using ecomorphological traits indicative of ungulate feeding and habitat preference. Additionally, we analyze changes in morphospace occupation through time in the selected North American artiodactyl faunas, in the context of changing prevailing environmental conditions. This study represents the first quantitative characterization of the Neogene savannas of North America in terms of the ecomorphological traits of ungulates, and investigates the similarities and differences in artiodactyl ecomorphological diversity between the savannas of present-day Africa and the Neogene of North America.

The correspondence analysis plots (Figure 4) show the morphospace positions on the first three principal axes of all taxa, including the extant ruminant artiodactyls from the present-day Serengeti and the fossil ruminant and camelid artiodactyls from the Neogene of North America. The resulting eigenvalues from the correspondence analysis are shown in Table 2 (for the row and column scores see Appendices 7 and 8). Euclidean similarity and distance indices between individual taxa can be found in the Supplementary Table S1, with values closer to zero indicating greater similarity/proximity in the morphospace and larger values indicating less similarity/proximity. Figure 4A shows the first two principal axes and Figure 4B shows axes 2 and 3; together these plots represent a three-dimensional morphospace seen from different vantage points. Axis one (Figure 4A) is dominated by body size and premolar row length and roughly arranges taxa in a continuum from large animals (e.g., giraffe, Giraffa camelopardalis, and common eland, Taurotragus oryx) with negative scores to small ones (e.g., duikers, genera Philantomba and Sylvicapra) with positive scores; additionally, taxa with shorter premolar rows have negative scores (e.g., camelids) and those with longer premolar rows have positive scores (e.g., most ruminants). Axis two (Figures 4A,B) is dominated by hypsodonty level and anterior face length: hypsodont mixed-feeders (e.g., impala, Aepyceros melampus; Grant’s gazelle, Nanger granti; Thomson’s gazelle, Eudorcas thomsonii) have negative scores (although the common eland has a slightly positive score), brachydont browsers have positive scores (e.g., giraffe and gerenuk, Litocranius walleri), and hypsodont grazers have scores close to zero (e.g., Bohor reedbuck, Redunca redunca; hartebeest, Alcelaphus buselaphus; African buffalo, Syncerus caffer). Axis three (Figure 4B) is dominated by the relative lengths of the temporalis moment arm and the metatarsals, and to a lesser extent by the relative lengths of the masseter moment arm and jaw corpus depth.

The extant Serengeti ruminants are well-discriminated in the three-dimensional morphospace on the basis of their dietary categories (see Appendix 9 for a full breakdown of where the taxa plot in the 3D morphospace). Dietary groups are statistically different from one another (Table 3). Browsers predominantly plot toward the front (i.e., negative scores on axis three) and top (i.e., positive scores on axis two) of the morphospace, with the exception of the blue duiker (Philantomba monticola), which has a weakly positive score on axis three, and Kirk’s dik-dik (Madoqua kirkii), which has a weakly negative score on axis two. Overall, browsers are spread out on the right-hand side of the morphospace (i.e., positive scores on axis one), with the exception of the largest browsers in the sample: the lesser and greater kudu (Tragelaphus imberbis and Tragelaphus strepsiceros, respectively), and the giraffe, which have weakly to moderately negative scores on axis one. Mixed feeders are very spread out on axis three but, overall, they tend to plot closer to the front of the morphospace (i.e., negative and weakly positive scores on axis three). They predominantly plot in the bottom (i.e., negative scores on axis two) right hand side (i.e., positive scores on axis one) area of the morphospace, with the exception of the common eland (weakly positive score on axis two and a strongly negative score on axis one) and the impala (weakly negative score on axis one). Grazers are fairly well-constrained in the three-dimensional morphospace. All of the grazing taxa plot in the back (i.e., positive scores on axis three) bottom (i.e., negative scores on axis two) area of the morphospace, and are evenly distributed between the left and right hand sides (i.e., moderately negative to moderately positive scores on axis one). In summary, the extant artiodactyl fauna from the Serengeti is well-discriminated in this analysis in terms of body size and dietary category, and therefore can provide a suitable basis for inferring the ecomorphological characteristics of the extinct artiodactyls from the Neogene of North America.

The phylogenetic flexible discriminant analysis (phylo FDA) results plot (Figure 5) also shows good separation between all three dietary groups among extant taxa, but shows particularly good separation between grazers and the other two dietary groups. Grazers tend to have more positive discriminant axis one (DA1) scores and more negative DA2 scores, browsers tend to have more negative DA1 and DA2 scores, and mixed feeders tend to have more negative DA1 scores and more positive DA2 scores. The posterior probability and discriminant axis values for all taxa are presented in Appendix 10.

The extinct camelids plot on the left hand side of the correspondence analysis plot (Figure 4, i.e., negative scores on axis one), in both the front and back of the morphospace (i.e., ranging from strong negative scores to strong positive scores on axis three). Overall, camelids tend to plot toward the middle of the morphospace (i.e., weakly negative to weakly positive scores on axis two), although Stenomylus has a strongly negative score on axis two. A number of camelids plot outside the area of morphospace occupied by any modern artiodactyl (e.g., Camelops sp., Stenomylus sp., Titanotylopus sp.), but some of them plot near the greater kudu (e.g., Procamelus grandis, Pleiolama magnifontis, and Oxydactylus sp.; Euclidean distances range from 0.330 to 0.363), the common eland (e.g., Homocamelus sp., Megatylopus sp.; Euclidean distances: 0.182 and 0.186, respectively), or the roan antelope (Hippotragus equinus) and waterbuck (Kobus ellipsiprymnus) (e.g., Paramiolabis sp., Homocamelus sp., Michenia sp.; Euclidean distances range from 0.239 to 0.352). The posterior probability obtained from the phylogenetic flexible discriminant analysis classifies most camelids as mixed feeders (Figure 5), with the exception of Aepycamelus, Oxydactylus and Procamelus (classified as browsers), as well as Camelops and Gigantocamelus (classified as grazers).

The sole representative of the family Protoceratidae, Lambdoceras sp., plots in the back (i.e., positive score on axis three), top (i.e., positive score on axis two), left hand side (i.e., negative score on axis one) octant of the morphospace, near extant browsers (Figure 4). Lambdoceras is classified as a browser in Figure 5.

For the most part, dromomerycids plot in the front (i.e., negative scores on axis three), top (i.e., positive scores on axis two), right hand side (i.e., positive scores on axis one) of the morphospace (Figure 4), with the exception of Yumaceras (a weakly positive score on axis three and a weakly negative score on axis one), and Pediomeryx and Sinclairomeryx (weakly positive scores on axis one and weakly to moderately positive scores on axis three). Overall, members of the Dromomerycidae plot close together in an area of the morphospace that somewhat resembles that occupied by medium-to-small to medium-sized browsers like the gerenuk, bushbuck (Tragelaphus scriptus), common duiker (Sylvicapra grimmia), and blue duiker. Most dromomerycids, except for Sinclairomeryx and Pediomeryx, are classified as browsers as shown in Figure 5.

The pronghorns (family Antilocapridae) predominantly plot in the back (i.e., positive scores on axis three), bottom (i.e., negative scores on axis two), right hand side (i.e., positive scores on axis one) octant of the morphospace (Figure 4), with the exception of Cosoryx furcatus and Merycodus (weakly positive scores on axis three), Submeryceros (a weakly positive score on axis two) and Texoceros (a weakly negative score on axis one). Merycodontine antilocaprids (an extinct subfamily of small forms) such as Cosoryx furcatus, Merycodus necatus and Paracosoryx sp., and the antilocaprine antilocaprids Capromeryx sp. and Texoceros sp., plot closely with mixed-feeding taxa such as the oribi (Ourebia oribi) and the steenbok (Raphicerus campestris), and with grazers such as the topi (Damaliscus lunatus) and the mountain reedbuck (Redunca fulvorufula) (Euclidean distances range from 0.190 to 0.338). The merycodontines Ramoceros sp. and Submeryceros sp. plot closer to the middle of axis two, somewhat near the Bohor reedbuck (Euclidean distances: 0.187 and 0.190, respectively). The phylogenetic flexible discriminant analysis classifies most pronghorns as mixed feeders, with the exception of Submeryceros (classified as a browser), and Merycodus and Ramoceros (classified as grazers) (Figure 5).

Among the hornless ruminants (all small taxa), Pseudoceras (family Gelocidae) plots in the front (negative score on axis three), bottom (negative score on axis two), right hand side (positive score on axis one) octant of the morphospace (Figure 4). The remainder of the hornless ruminants plot in the back (positive scores on axis three), right hand side (positive scores on axis one) area of the morphospace, with most taxa plotting in the top octant (positive scores on axis two), with the exception of Nanotragulus (family Hypertragulidae) and Longirostromeryx (family Moschidae, extinct subfamily Blastomerycinae), which have weakly negative scores on axis two. In general, the small hornless ruminants occupy an area of the morphospace unoccupied by any extant ruminants. The moschid Longirostromeryx wellsi plots somewhat near the small extant browsers, particularly the bushbuck and the common duiker (Euclidean distances: 0.209 and 0.326, respectively). Most hornless ruminants are classified as browsers in Figure 5, except for Nanotragulus, Blastomeryx gemmifer, Parablastomeryx, and Problastomeryx which are classified as mixed feeders.

Most extinct and extant artiodactyl families are statistically different from one another (Table 4), except for dromomerycids and moschids. Interestingly, extinct and extant ruminants are also statistically different from one another on the basis of their ecologically relevant morphological characters (Table 3), most likely because the extant ruminants are largely bovids.

Figure 6A shows the changes in morphospace occupation of artiodactyl communities in the Central Great Plains of North America from the early Miocene (Arikareaan 4) to the earliest Pleistocene (late Blancan) (axis one vs. axis two plots on the left, axis three vs. axis two plots on the right), and illustrates the prevailing conditions of temperature and precipitation during the transition from savanna woodlands (dominated by pooid grasses) into grass-dominated ecosystems. For the morphospace occupation plots of the individual faunas, showing how the individual taxa plot within the overall morphospace, see Appendix 11. The most taxonomically diverse faunas, which occupied the largest volumes of morphospace (Table 5 and Figure 7) across axes one, two and three – (i.e., 3-Thomson Quarry, 4-Echo Quarry, 5-Norden Bridge Quarry, 6-Burge Quarry) – were from late early Miocene to the middle Miocene localities. During this time [the Middle Miocene Climatic Optimum, which represents the peak of the Miocene warming period between 17 and 15 Ma; Zachos et al. (2001)], high temperatures and precipitation levels were evident. In contrast, the faunas spanning the late Miocene to the earliest Pleistocene (i.e., 7-Blue Jay Quarry, 8-Cambridge Local Fauna, 9-Edson Quarry, 10-Broadwater Local Fauna, and 11-Big Springs Local fauna) were comparatively less diverse, their morphospace occupancy was significantly smaller (with the exception of 8-Cambridge Local Fauna, whose morphospace occupation area was one of the largest in the study with respect to the first two axes, even though the fauna was not very diverse), and they tend to plot more negatively on axes one and three. The prevailing environmental conditions of the time were characterized by low mean annual temperatures and precipitation levels, which started with the Mid-Miocene Climate Transition (marked by the formation of the East Antarctic ice sheet at around 14 Ma and a concomitant decrease in temperatures) (Flowers and Kennett, 1993; Pierce et al., 2017).

Throughout the Neogene, mixed feeders were common in all artiodactyl faunas. During the mid-Miocene (specifically from the Hemingfordian to the Clarendonian), artiodactyl faunas were rich in browsing taxa (including many dromomerycids and hornless ruminants, and some camelids). Toward the late Miocene, the diversity of browsers started declining, until ultimately mixed-feeding taxa (and possibly some grazers) were all that remained.

The results of the phylogenetic flexible discriminant analysis indicate a change in diet among camelids throughout the Neogene: during the middle Miocene camelids were browsers and mixed feeders, toward the end of the Miocene they were exclusively mixed feeders, and by the Pliocene a couple of taxa possibly could have been grazers. We do not see a marked change in diet among individual ruminant groups through time, partly because they were neither as diverse nor as long lived as camelids. Some hornless ruminants (i.e., gelocids and leptomerycids) and most dromomerycids remained browsers (with the exception of Pediomeryx) throughout the Neogene, and pronghorns were mostly mixed feeders. All of these families decreased in diversity after the Mid-Miocene Climatic Optimum and went extinct by the end of the Miocene [in the Pliocene the role of browsing artiodactyls was taken by immigrant cervids Webb, 1998a)].

Figure 6A depicts the differential morphospace occupation of camelids and ruminants from localities in different NALMA subdivisions throughout the Neogene. In broad terms, the morphospace occupation of camelids and ruminants does not overlap along axis one; ruminants tend to occupy a larger morphospace area than camelids in all time periods except the late Blancan. However, while ruminants comprised several families and camelids only one, throughout the Neogene camelids were always more diverse than any individual ruminant family [except in the early Hemingfordian Flint Hill Local Fauna fauna No. 2; see Appendix 3)].

Figure 6B compares the morphospace occupation of the most diverse Neogene fauna in the study, Burge Quarry (fauna No. 6; bottom), with the extant Serengeti ruminants (top). The morphospace occupation of extant camelids (not present in the Serengeti) is also shown. In both extinct and extant faunas, camelids tend to plot toward the left-hand side of the plot (negative scores on axis one) and ruminants plot toward the right-hand side of the plot (positive scores on axis one). While the morphospace occupation of camelids and ruminants does not overlap in the extinct fauna, it does among the extant taxa: lamines plot near extant grazers and medium-sized mixed feeding bovids, and camelines plot near the common eland. Some modern ruminants also occupy areas of morphospace not occupied by their extinct counterparts (but remember that no Neogene North American ruminants were bovids), such as the blue duiker and large grazers like the African buffalo and the blue wildebeest. Likewise, extant camelids occupy areas of morphospace not occupied by Neogene North American camelids; for example, lamines have more positive scores on axis one and overlap with the ruminant morphospace. This indicates that, even though there is a phylogenetic component to the morphospace occupation of camelids and ruminants, overlap is not absent, suggesting that there is an ecomorphological signal in addition to a phylogenetic one.

Throughout the Neogene, most artiodactyl faunas were ecomorphologically similar to one another (Table 6), with a few exceptions. Flint Hill Local Fauna (fauna No. 2), dominated by ruminants and scarce in camelids, is statistically different to faunas 7-Blue Jay Quarry, 9-Edson Quarry, 10-Broadwater Local Fauna, and 11-Big Springs Local fauna (dominated by camelids and with few ruminants). Likewise, Big Springs Local Fauna (fauna No. 11- low diversity and disparity) is statistically different to faunas 3-Thomson Quarry and 5-Norden Bridge Quarry (very diverse and disparate faunas). When comparing the Neogene faunas with the extant Serengeti fauna, almost 50% of the former are statistically different from the modern African assemblage, namely faunas 2- Flint Hill Local Fauna, 7-Blue Jay Quarry, 9-Edson Quarry, 10-Broadwater Local Fauna, and 11-Big Springs Local fauna, which represent some of the least disparate and diverse faunas (with the exception of fauna 1-Morava Ranch Quarry).

Over the entire duration of the Neogene, it is clear that the faunas that have the highest ecologic and taxonomic diversity are those that lived during the Middle Miocene Climatic Optimum (MMCO) and prior to the formation of the East Antarctic ice sheet, which led to a concomitant drop in global temperature (Figure 6). In particular, the Early Clarendonian 6-Burge Quarry assemblage – the classical “Clarendonian Chronofauna” of Webb – constitutes the most diverse artiodactyl fauna included in the analysis (14 genera); interestingly, this assemblage dates to the start of the Neogene cooling and drying trend. Herbivorous artiodactyls displayed their highest diversity in North America during the MMCO (Janis et al., 2000, 2004). Additionally, the ecosystems of the first half of the Miocene produced abundant resources for browsers, which dominated artiodactyl communities prior to the middle Miocene (Janis et al., 2004). These highly rich, predominantly browsing (note that some browsing ungulates were excluded from this study, including artiodactyls such as tayassuids, and perissodactyls such as tapirs and anchitheriine horses) mammal communities from the late early to middle Miocene indicate a higher level of primary productivity than seen today, which might have been a consequence of elevated levels of atmospheric CO₂ (Janis et al., 2004; Kürschner et al., 2015). Phytolith evidence indicates that the prevailing ecosystems during this time were most likely savannah woodlands dominated by pooid grasses: plant macrofossils and palynofloras also suggest that grass-dominated habitats might have started to expand during this time (Strömberg, 2011). Faunas 3-Thomson Quarry, 5-Norden Bridge Quarry and 6-Burge Quarry, which occurred during or immediately after the MMCO, are the most diverse faunas of the North American Neogene (Figure 7) and are also the ones that most closely resemble the modern Serengeti artiodactyl fauna (Table 6).

Starting in the late Miocene, ungulate faunas became significantly less diverse, with both browsers and mixed feeders decreasing in numbers (Figure 6: only cameline camelids, some dromomerycids from the tribe Cranioceratini, some antilocaprine pronghorns, and the gelocid Pseudoceras remained; see Appendix 11); in fact, browsers disappeared altogether by the latest Miocene. These less diverse and less disparate faunas (faunas 7 through 11, with the exception of fauna 8-Cambridge Local Fauna) are all statistically different from the fauna of the modern-day Serengeti. Cambridge Local Fauna (fauna No. 8) is the only fauna of the late Miocene to still be very disparate (although not very diverse), due to the presence of the gelocid Pseudoceras. The decrease in diversity seen toward the end of the Neogene has generally been attributed to the increasingly drier and cooler conditions of the period, and the resulting change from woodland savannas, with abundant resources for browsers, to open grasslands, which offer more support to grazers (Janis et al., 2004). Phytoliths, plant macrofossils and palynofloras all indicate the spread of grass-dominated habitats in North America by the latest Miocene and Pliocene (Strömberg, 2011). Janis et al. (2004) highlight that toward the end of the Miocene, there was a decline in total species richness in the Great Plains: that is, a decrease in diversity occurred not only among browsers, but also among grazers and mixed feeders. The present analysis does not include any taxa previously considered as grazers, since this dietary category was largely occupied by equids in the North American Neogene and is thus beyond the scope of this study.

Throughout the Neogene of North America, the morphospace areas occupied by camelids and ruminants never overlapped (Figure 6). In fact, the only fossil taxa that occupied a similar morphospace area to camelids were the protoceratids, which are usually considered as camelid-related tylopods (Prothero, 1998). During the Miocene and Pliocene, camelids and ruminants belonged to equivalent dietary categories and coexisted within the same communities (Janis et al., 2004; Kita et al., 2014). Even so, they occupied divergent morphospaces on the basis of their phylogenetic and morphological differences, as previously mentioned. In the modern world, camelids rarely compete with ruminants. In South America, the dietary preference of camelids (i.e., opportunistic mixed feeding) only overlaps with that of domestic livestock (note that there are no native bovids in South America) and, given the ability of camelids to subsist on a lower food intake than ruminants (Franklin, 2011; Dittmann et al., 2014a, b), the two groups of animals can live alongside each other without competing for resources (Borgnia et al., 2008; Chammem et al., 2010). The fact that Neogene camelids and ruminants plot in different morphospace areas might indicate differential use of resources within the community, and might parallel the situation seen in modern-day faunas (note that no bovids or large mixed feeding ruminants, like the common eland, were present in the Neogene of North America); however, we cannot ignore the possible contribution of phylogeny to their separation in morphospace. The flexible discriminant analysis, which takes phylogenetic information into account, supports the notion that camelids and ruminants belonged to similar dietary categories.

Janis et al. (1994) note that, during the Miocene, the relative body sizes of equids and ruminants varied depending on their dietary preferences: browsing horses were larger (or smaller) than browsing ruminants, and mixed feeding and grazing artiodactyls were larger (camelids) or smaller (ruminants) than mixed feeding and grazing equids. These differences were attributed to their different digestive physiologies. Throughout the Neogene, the artiodactyls in our study were mostly browsers or mixed feeders, and camelids always displayed larger body sizes than ruminants, perhaps because of their different digestive physiologies and metabolic rates (camelids have lower metabolic rates than true ruminants, an attribute that makes them more efficient in arid conditions [Dittmann et al., 2014a]).

In a separate correspondence analysis (Appendix 13) the extant camelines and extinct camelids fall in comparable morphospace areas along all three axes, but the lamines do not plot in the same area of the morphospace. Some extinct camelids share similar scores on axis two (but not on axis one) with the extant lamines, including Stenomylus, Camelops and Hemiauchenia. The Bactrian camel (Camelus bactrianus) and the dromedary (Camelus dromedarius) plot near to the common eland, along with several extinct camelids. The guanaco (Lama guanicoe) and the llama (Lama glama) plot between the morphospaces of medium-sized browsing and grazing ungulates, and the vicuña (Vicugna vicugna) and the alpaca (Vicugna pacos) plot relatively close to the medium-sized grazing ungulates; some of the extinct antilocaprids also plot near to the lamines. It is notable that, even though camelids and ruminants can occupy similar areas of morphospace, this did not happen in the North American Neogene. Note, however, that there were no large ruminants in North America, but the eland of Africa (and to a lesser extent the African buffalo) does cluster with some extinct camelids. The discrepancies in morphospace occupancy between camelids and ruminants could be related to differences in morphology and physiology; camelids display a long-term pattern of adaptation to arid habitats with sparse food resources, in part due to their lower metabolic rates (Franklin, 2011; Dittmann et al., 2014a). Likewise, a comparative genomic analysis of extant camelids, including the Bactrian camel, the dromedary and the alpaca, revealed a number of physiological mechanisms related to the desert adaptations within the genus Camelus and those of their ancestors (Wu et al., 2014).

However, the lack of North American ruminants in the camelid area of the morphospace could also be a matter of incumbency. Schenk et al. (2013) suggest that the first colonizers of a particular region diversify faster than subsequent colonizers with similar niches. Camelids appeared in North America during the middle Eocene, approximately 45.9 million years ago (Honey et al., 1998), whereas the two earliest ruminant families, the small-sized Hypertragulidae and Leptomerycidae, are relatively younger and appear in North America approximately 41.3 million years ago (Webb, 1998b). This might have allowed camelids to diversify within particular niches that would have then become unavailable for later colonizers, such as the dromomerycids and antilocaprids, which did not appear in North America until the early Miocene (Janis and Manning, 1998a,b). The extraordinary diversification of equids from the Eocene onward, and their shift to predominantly grazing behavior toward the end of the Miocene (Wang et al., 1994; Semprebon et al., 2016), might have contributed to largely excluding camelids from the grazing niche. This stems from the fact that the digestive physiology of hindgut fermenters, such as equids, allows them to better tolerate lower quality herbage, which proved advantageous as forage quality decreased toward the end of the Miocene with increasing aridity and seasonality (Janis et al., 1994).

However, we consider that the occupancy of “camelid morphospace” may carry a real environmental signal. North America was evidently more arid than the Old World during the Neogene (Eronen et al., 2012): note that the common eland (the only bovid to cluster with the extinct camelids) is among the most arid-adapted of the African bovids, and is similar in body size and diet (mixed feeding) to these North American camelids. The morphophysiological adaptations of extant camelids must date back to at least the split between lamines and camelines, which would be around the early Miocene based on both anatomical (Honey et al., 1998) and molecular (Heintzman et al., 2015) data. While the Bactrian camel and the dromedary (tribe Camelini) have better physiological adaptations to harsh desert environments than alpacas (tribe Lamini), both tribes had an estimated increase in population size during transitions to colder and more arid conditions, such as the middle Pleistocene transition (Wu et al., 2014). Perhaps camelids were already adapted to arid conditions by the early Miocene, whereas the immigrant ruminants would not have been so fortunate.