Ancient feeding ecology inferred from stable isotopic evidence from fossil horses in South America over the past 3 Ma

Background Stable isotope ratios (13C/12C and 18O/16O) in fossil teeth and bone provide key archives for understanding the ecology of extinct horses during the Plio-Pleistocene in South America; however, what happened in areas of sympatry between Equus (Amerhippus) and Hippidion is less understood. Results Here, we use stable carbon and oxygen isotopes preserved in 67 fossil tooth and bone samples for seven species of horses from 25 different localities to document the magnitude of the dietary shifts of horses and ancient floral change during the Plio-Pleistocene. Dietary reconstructions inferred from stable isotopes of both genera of horses present in South America document dietary separation and environmental changes in ancient ecosystems, including C3/C4 transitions. Stable isotope data demonstrate changes in C4 grass consumption, inter-species dietary partitioning and variation in isotopic niche breadth of mixed feeders with latitudinal gradient. Conclusions The data for Hippidion indicate a preference varying from C3 plants to mixed C3-C4 plants in their diet. Equus (Amerhippus) shows three different patterns of dietary partitioning Equus (A.) neogeus from the province of Buenos Aires indicate a preference for C3 plants in the diet. Equus (A.) andium from Ecuador and Equus (A.) insulatus from Bolivia show a preference for to a diet of mixed C3-C4 plants, while Equus (A.) santaeelenae from La Carolina (sea level of Ecuador) and Brazil are mostly C4 feeders. These results confirm that ancient feeding ecology cannot always be inferred from dental morphology. While the carbon isotope composition of horses skeletal material decreased as latitude increased, we found evidence of boundary between a mixed C3/C4 diet signal and a pure C4 signal around 32° S and a change from a mixed diet signal to an exclusively C3 signal around 35°S. We found that the horses living at high altitudes and at low to middle latitude still have a C4 component in their diet, except the specimens from 4000 m, which have a pure C3 diet. The change in altitudinal vegetation gradients during the Pleistocene is one of several possibilities to explain the C4 dietary component in horses living at high altitudes. Other alternative explanations imply that the horses fed partially at lower altitudes.


Background
In South America, horses are represented by two groups: equidiforms and hippidiforms. Hippidion are characterized by a retracted nasal notch which has been interpreted as an adaptation to the presence of a proboscis and limbs with robust metapodials. The upper teeth present an elongate-oval protocone with simple enamel plication and lower teeth have a deep ectoflexid, penetrating the isthmus (Figure 1). On the other hand, in Equus (Amerhippus) a retracted nasal notch is not present and the metapodials are slender. The upper teeth present a triangular protocone and multiple internal posfossette plications. There are features that are common to both of these groups, such as differentiation into horses of both small and large body size, which are possibly a consequence of convergence due to adaptation to similar environments. The three species of hippidiforms are included within the genus Hippidion Owen, 1869 * Correspondence: malberdi@mncn.csic.es 2 Museo Nacional de Ciencias Naturales, CSIC. José Gutiérrez Abascal, 2. 28006-Madrid, Spain Full list of author information is available at the end of the article [1,2] whereas the five species of equidiforms are included in the subgenus Equus (Amerhippus) [3]. Each of these groups has different ecological adaptations that are evident in the cranial morphology, robustness of the limbs and body size [4,5].
A technique that has proven useful for investigating the ecology of fossil horses is through examination of stable isotope values found in teeth and bone in combination with dental wear [6][7][8][9][10][11][12][13][14][15][16]. Isotopic analyses can reveal information about resource use and resource partitioning among species and is also able to determine diet and habitat use [15,17]. Here, we will address: (1) whether stable isotope values permit identification of resource use and partitioning among horse species and (2) if resource use and partitioning are determined, do the results support the ecology predicted by morphology or body size? Finally, we compare carbon isotope values and evaluate the hypothesis that dietary niches, inferred from the mean and variation of carbon isotope values, did not change throughout time in the same latitude.
Stable carbon and oxygen isotopes are incorporated into the tooth and bone apatite of horses and are representative, respectively, of the food and water consumed while alive. The carbon isotope ratio is influenced by the type of plant material ingested, which is in turn influenced by the photosynthetic pathway utilized by the plants. During photosynthesis, C 3 plants in terrestrial ecosystems (trees, bushes, shrubs, forbs, and high altitude and high latitude grasses) discriminate more markedly against the heavy 13 C isotope during fixation of CO 2 than C 4 plants (tropical grasses and sedges). Thus C 3 and C 4 plants have distinct δ 13 C values. C 3 plants usually have δ 13 C values of -30 per mil (‰) to -22‰, with an average of approximately -26‰, whereas C 4 plants have δ 13 C values of -14 to -10‰, with an average of about -12‰ [18][19][20][21][22]. Animals incorporate carbon isotopes from food into their teeth and bone with an additional fractionation of~12 to 14‰ [23,24]. Mammals feeding on C 3 plants characteristically have δ 13 C values between -14 and -8‰, while animals that eat C 4 tropical grasses have δ 13 C values between +2 and -2‰. A mixed-feeder would fall somewhere in between these two extremes [25,26]. Hence, the relative proportions of C 3 and C 4 vegetation in the diet of an animal can be determined by analyzing the δ 13 C value of its teeth and bones.
A number of previous studies have used the carbon and oxygen isotopic abundance of fossils and paleosols from South America to reconstruct the diets of extinct herbivores and the paleoenvironmental parameters of ancient terrestrial communities and ecosystems [27][28][29][30][31][32]. Carbon isotope data for horses from South America have been presented in several papers [8,[33][34][35]. In 1999, MacFadden et al. [36] presented the ancient distributions and latitudinal gradients of C 3 and C 4 grasses based on isotopic data from New World Pleistocene horses. In addition, some papers have investigated the application of geochemical techniques in conjunction with morphological and dental wear data to reconstruct the feeding ecology and niche characterization of individual species [34,36,37].
All equid taxa from South America were sampled for teeth (n:29) and bone (n:38) stable carbon and oxygen isotopes (table 1 and 2). Additional data of thirty samples were taken from MacFadden et al. [36]. Together the data represent five species within the subgenus Equus  Figure 2). H. saldiasi is restricted to a particular habitat in southern Patagonia [38,39]; while H. principale and H. devillei come from different localities in South America, such as Tarija in Bolivia and the Pampa region in Argentina, that cover a broad range of altitudes from 10 to 4000 m. One restriction to our study is the chronological control of the sample. Most of the samples were collected from old museum collections in Ecuador, Bolivia and Argentina. These old collections were recovered without sufficient stratigraphic control. Anyway we considered this limitation may condition the interpretations about altitudinal and latitudinal gradients for fossil samples, but do not invalidate the suggested patterns. In order to demonstrate how dietary resources were partitioned we divided the samples into 10 different groups taking into account the genus, as well as the age and the altitude of the corresponding deposit. Data for these groups and descriptive statistics are listed in

Materials
Fossil samples were collected from specimens stored at the following institutions in Argentina: Museo de La Plata, Museo Argentino de Ciencias Naturales "Bernardino Rivadavia" in Buenos Aires and Museo de Ciencias Naturales y Antropológicas "Juan Cornelio Moyano"in Mendoza. Museo de la Escuela Politécnica Nacional of Quito, in Ecuador. The museum specimen number, locality, country, age, skeletal tissue (enamel, bone and dentine) and the altitudinal and latitudinal distribution of each sample are given in Tables 1 and 2. We analyzed 26 samples of Hippidion and 41 carbon and oxygen Samples with the same "specimen number" correspond to the same individuals. The bones correspond to the mandibular or maxillary remains that contained the teeth that were analyzed. Plio = Pliocene; Pl = Pleistocene; e = enamel; d = dentine; b = bone; t = tooth (enamel + dentine). Ar = Argentina; Bo = Bolivia; Br = Brazil; Ch = Chile. A = Andean; PL = Plain landscape. 1 Taken from MacFadden et al. [22], 2 taken from MacFadden and Shockey [27] and 3 taken from MacFadden et al. [20]. isotope composition samples of Equus (Amerhippus). Additional 32 samples values (Table 1 and 2) were taken from MacFadden et al. [31,33,36] and MacFadden and Shockey [34].

Pre-treatment of the samples
The samples were finely ground in an agate mortar. The chemical pre-treatment of the samples was carried out as described in Koch et al. [58] in order to eliminate secondary carbonate. About 40-50 mg of powdered enamel and bone samples were soaked in 2% NaOCl for three days at room temperature to oxidize organic matter. Residues were rinsed and centrifuged five times with de-ionized water, and then treated with buffered 1M acetic acid for one day to remove diagenetic carbonates. Pre-treatment of the enamel was slightly different because samples were soaked in 2% NaOCl for only one day.  [59] to calculate the magnitude of the oxygen isotopic fractionation between apatite CO 2 and H 3 PO 4 at 50°C. The analytical variation for repeated analyses was 0,1‰ for δ 13 C and 0,2‰ for δ 18 O. For the analysis of phosphate we followed the chemical treatment procedure described by Tudge [60], which resulted in the precipitation of the phosphate ions in the form of BiPO 4 . CO 2 was obtained by reacting BiPO 4 with BrF 5 as described by Longinelli [61]. All the samples were run in duplicate and the reported results are the mean of at least two consistent results. The analytical precision for repeated analyses was 0,2‰. We performed both parametric (t-test) and nonparametric (Wilcoxon Signed-Rank) statistical tests to evaluate δ 13 C and δ 18 O differences in Middle and Late Pleistocene populations. SPSS 15.0 software was used for the statistical analysis. Samples with the same "specimen number" correspond to the same individuals. The bones correspond to the mandibular or maxillary remains that contained the teeth that were analyzed. Abbreviations as in Table 1. * taken from MacFadden et al. [25].

Results
Preservation state of the enamel, dentine and bone in the specimens analyzed We checked the diagenetic alteration between the different skeletal tissues under the assumption that primary values were similar between different skeletal tissues from one individual. The phosphate oxygen isotope composition is usually considered to be more robust against diagenetic alteration than the carbonate oxygen at least if no bacteria are involved during the alteration processes. We measured both δ 18 O CO3 and δ 18 O PO4 values on the same specimens and obtained a regression between both, and compared those with known isotope equilibrium-relationships for biogenic apatite of modern bones and teeth [40,41]. If samples fall in the range of modern biogenic apatite this would argue for a preservation of primary δ 18 O CO3 and δ 18 O PO4 values, and thus likely also δ 13 C values as they are less easily altered than δ 18 O CO3 values. This was taken as an indicator that even the bone and dentine samples may be reasonably well-preserved and can be interpreted to infer feeding ecology and habitat use of these ancient horses. Several authors [42][43][44] suggest a high correlation between these two phases in some European and North American equids. In South America, the equation obtained for Equus(Amerhippus) from Argentina [45] was: δ 18 O CO3 = 16.74 δ 18 O PO4 + 0.64; R 2 = 0.91. More recently, Sánchez et al. [8,32] also found that a In the first case it appears that the sample might have been modified by diagenesis because the carbonate results among the three skeletal phases (enamel, dentine, and bone) are different. In fact, the range of variation is 2.6‰ in the δ 18 O PO4 and 5.6‰ in the δ 18 O CO3 for the samples from the same specimen. Results of δ 13 C obtained from the bone are similar to those obtained from the dentine and are, in general, equal or more negative than results obtained from the enamel. We observed the same pattern in the δ 18 O results as we did in the δ 13 C results. The range of variation between the three skeletal phases (enamel, dentine, and bone) was small and we suppose that they have not been significantly altered by diagenesis. For these reasons, we consider these δ 13 C values as representative of each group of horses.

Dietary Partitioning
The carbon isotopic ratio of Equus (Amerhippus) and Hippidion remains indicate significant ecological differences ( Figure 4). The Hippidion samples analyzed here are more homogeneous than the Equus (Amerhippus) samples, with a δ 13 C range between -12,9 and -8,0 ‰ (Table 1 and 2).  All the species of Hippidion were almost exclusively C 3 feeders but some individuals from Bolivia and Argentina fall at the lower end of the mixed C 3 /C 4 range. For instance, H. principale from the Eastern corridor (at sea level) and H. devillei from the Andes corridor yield similar δ 13 C values suggesting that they ate mainly C 3 plants. The same pattern of dietary partitioning was obtained when comparisons were made between the same taxa at different latitudes (between 22°S and 52°S). As mentioned before, a few outliers (e.g. δ 13 C values of 9,2; 6,1 and 5,4‰ from La Carolina) cannot be easily explained. These extremely high δ 13 C values (above 3‰) cannot be explained by consumption of C 4 vegetation, which should impart an upper limit of about 3‰. These outliers could be the result from one of several possibilities, such as individuals living in costal peninsula areas of Ecuador during the time in which C 4 grasses were abundant and may have produced δ 13 C values not observed in the modern ecosystem, or the sample presents taphonomic alteration. Specimen number V.3037 has a high variation between dentine, enamel, and bone (9,2; 1,5 and 6,1‰ respectively). MacFadden et al [36] obtained a δ 13 C value of 1,8‰ for the enamel of the same specimen but obtained 5,4‰ for the enamel of

Areas of sympatry
As might be expected from ecological theory, finer-scale preliminary results from this study suggest feeding and niche differentiation within coexisting horses. Hippidion and Equus (Amerhippus) are sympatric in the same stratigraphic level in two localities: Tarija (22°S) in Bolivia and the Pampas in Argentina (38°S). Data from both localities suggest some difference in dietary partitioning. In a previous study, MacFadden and Shockey [34] presented carbon isotopic results from Tarija herbivores that, based on dental morphology, span the spectrum from presumed browsers (e.g., tapirs) to presumed grazers (e.g., horses) coexisting in the same localities. They suggested than the horses clearly occupied different dietary niches and can be separated using the carbon isotopes and hypsodonty index. Hippidion is the least hypsodont taxon and has the most negative mean δ 13 C value, whereas the relatively hypsodont Equus has the most positive mean δ 13 C value. As has been shown in another article [36], Equus is usually among the most grazing adapted mammalian herbivore in Pleistocene terrestrial ecosystems. This position in the herbivore community is similar at Tarija, whereas Hippidion are more adapted to the browsing end of the spectrum and Equus are primarily C 4 grazers based on their carbon isotopic signature.
Our data suggests that the range of δ 13 C values for H. devillei and H. principale from Tarija falls in the low end of the mixed C 3 /C 4 range (-10,8 to -6,6‰), whereas E. (A.) insulatus falls in the higher end (-8,3 to -2,3‰). There is some overlap with at least the two values of -8,3 and -6,6‰ indicating some differences in isotopic niche, with H. devillei consuming less grass than E. (A.) insulatus at this place and time [36].
In the data from Arroyo Tapalque (38°S), one of the localities in the Pampas, the range of δ 13 C values for H. principale (-11,6 to -9,3‰) and E. (A.) neogeus (-8,5 to -7,9‰) do not overlap, but both are close that they suggest the same pattern of dietary partitioning as in Tarija.
If we look at the morphology, these taxa are very different. Several previous papers based on cranial and limb morphology associate these differences with browsing and grazing diet preference [e.g. [46]]. In general, the teeth of Equus (Amerhippus) are more hypsodont than Hippidion, and the enamel patterns are more complicated in Equus (Amerhippus).
Another difference between sympatric species is body sizes. For instance, H. principale and E. (A.) neogeus are sympatric in the Pampas in Argentina. Both have a large body size, adapted to open habitat, but they differ in body mass (460 kg and 378 kg, respectively) [47]. The skulls of E. (A.) neogeus are big and show an enlarged preorbital and nasal region. The limb bones are large and robust, but more slender than in the other South American Equus species [3,4]. On the other hand, H. principale has a retracted nasal notch might signify some sort of proboscis. The skeleton is large and bulky, and the extremities are robust, mainly the metapodials and phalanges. It is the largest and strongest of the South American hippidiforms. These characteristics are classically associated with dietary and habitat preferences. The morphology indicated that H. principale may have been a browser but was able to live in open grasslands and that E. (A.) neogeus is the most hypsodont horse, and has the relatively straight muzzle characteristic of grazing horses [46].

The effect of latitude
Some of the most fundamental patterns in global biogeography are those that are structured by the latitudinal gradients that extend from pole to equator ( Figure 6). The proportion of C 3 to C 4 grasses in most modern ecosystems rises with increasing latitude. C 3 grasses are predominant (>90%) in high-latitude steppes and prairies, whereas C 4 grasses are predominant (>70-90%) in most low-elevation, and equatorial grasslands. The transition between C 3 and C 4 dominance in grasslands occurs at about 40-45°latitude in the Northern Hemisphere [48][49][50]. Exceptions to this general rule include grasses at high-elevations or in climates with cool-growing seasons (e.g. the Mediterranean basin), where the grasses are predominantly C 3 regardless of latitude [50][51][52], and the occasional C 4 grass species that are found in Arctic regions [53].
MacFadden et al. [36] used the Pleistocene distribution of Equus (Amerhippus) in the Americas to present a general δ 13 C gradient that seems to be symmetrical on either side of the Equator. They found that the isotopic transition between a full C 4 signal and full C 3 signal is observed at about 45°N in the northern hemisphere. Although there are considerably fewer data points in the southern hemisphere they [36] have found a signal of exclusively C 3 feeders at around 35°S (province of Buenos Aires). A similar pattern in the Southern Hemisphere was found by Sánchez et al. [32] in the distribution of gomphotheres in South America. Samples of gomphotheres from the province of Buenos Aires, at around 39°S latitude show a mean δ 13 C value of -10,8‰, while samples from Chile (from several localities around 35°to 41°S) show a mean δ 13 C value of -12,3‰. This fact would confirm the existence of a latitudinal gradient for the Southern Hemisphere, and places the transition between a full C 3 signal and mixed C 3 /C 4 signal around 35°to 41°S.
The new data for the genus Equus (Amerhippus) analyzed in this paper are in agreement with the δ 13 C latitudinal pattern postulated by MacFadden et al. [36], while data for Hippidion also seem to follow a clear pattern. Samples of Hippidion show a boundary between an exclusively C 3 to a mixed C 3 /C 4 diet at Tarija (22°S), but in this case there is an effect of altitude combined with latitude. At the highest latitude specimens from Patagonia (52°S), the mean value is -12.2‰. The middle latitude samples, from the province of Buenos Aires (34°S), show δ 13 C values that range between -12,9 and -8,1‰, while the lowest latitude specimens, from Tarija (22°S), show δ 13 C values that range between -6,6 to -0,8‰.

Altitudinal gradient
There are also significant differences between the δ 18 O CO3 values for the two genera, even when the range of δ 18 O values for high altitude and equatorial samples are taken into account ( An important point concerns altitudinal gradients. It seems that the horses living at high altitudes, 1866 up to 2780 m above sea level, and at low to middle latitude   [55]. The fossil pollen records show that such oscillations in patterns of plant distribution were repeated many times during the Pleistocene Ice Ages [56]. During the Last Glacial Maximum, when atmospheric pCO 2 was reduced by some 50%, C 4 plants dominated the Paramo vegetation, while only the highest mountain tops were covered by C 3 grasses because of the low temperatures. Such small patches of C 4 -rich vegetation are probably relicts from the last ice age during which paramo vegetation was mainly composed of small tussocks and tufts of C 4 grasses [57]. The change in altitudinal vegetation gradients during the Pleistocene is one of several possibilities to explain the C 4 dietary component in horses living at high altitudes. Another alternative explanation might be that the horses partially fed at lower altitudes.

Conclusions
Based on modern analogues, Pleistocene horses are inferred to be grazers but none of the grazing horses were interpreted as consumers of only C 4 grasses. Our data shows that Equus feeders. These results confirm that ancient feeding ecology cannot always be inferred from dental morphology.
The record from South America suggests that Hippidion is in general a higher latitude taxon than Equus (Amerhippus). The highest latitude occurrence of Hippidion appears to be in southern Bolivia, in contrary to Equus (Amerhippus) where the occurrences are further north, and alone goes a long way in explaining the isotopic differences.
The data for Hippidion indicates a preference for C 3 plants and mixed C 3 -C 4 plants, but most of this data came from high altitude or latitude specimens. One possible, but unconfirmed explanation is that Hippidion were living in a "C 4 World" and were browsers as indicated by their morphology, but more southern individuals were living at latitudes high enough to support C 3 and C 4 grasses.
The current study demonstrates the utility of using wide-ranging fossil mammals to explore latitudinal gradients and patterns of C 3 /C 4 grass distribution and continental palaeotemperature during the Pleistocene. The carbon isotope composition of horses decreased as latitude increased. In Equus (Amerhippus) we found a change in signal between a mixed C 3 /C 4 diet and a pure C 4 diet around 32°S and a boundary between mixed diet and exclusively C 3 signals at 35°S.
We also found that the horses living at high altitudes and at low to middle latitudes still have a C 4 component in their diet, except for those specimens living at 4000 m, which have a pure C 3 diet. The change in altitudinal vegetation gradients during the Pleistocene is one of several possibilities to explain the C 4 dietary component in horses that lived at high altitudes. Another alternative explanation implies that the horses fed partially at lower altitudes.