1.1 Northeastern Mexico: San Josecito Cave and Cedral fossil sites

The fossil localities of San Josecito Cave (Nuevo León) and Cedral (San Luis Potosí) are two of the most studied late Pleistocene sites in northeastern Mexico. Stock [20, 21] considered all of the horse remains from San Josecito Cave to belong to a single species of Equus, which he thought was most similar to Equus conversidens Owen, 1869, but with sufficient morphological differences to be identified as a new subspecies: E. conversidens leoni. As pointed out by Dalquest [9] and Winans [4, 10], Stock [20, 21] did not select a type nor publish a formal description; thus, the name should be regarded as a nomen nudum. Moreover, Winans [4] proposed E. conversidens to be a nomen dubium, because she considered, in accordance with Hibbard [7], that the convergence of the cheek tooth rows toward the rostrum seen in the holotype (one of the main diagnostic characters for this species) was the result of a distorted restoration. Winans [10] assigned the specimens from San Josecito Cave to her species group E. alaskae (Hay), 1913. Contrary to Winans [4], Azzaroli [12] regarded E. conversidens as a valid species distinct from E. niobrarensis alaskae Hay, 1913 and identified the latter as a synonym of E. ferus Boddaert, 1785. He figured and described a partial skull from central Mexico, in which the two tooth rows converge toward the rostrum, suggesting that the holotype of E. conversidens was correctly mounted. In addition, Azzaroli [12] referred the fossil material from San Josecito Cave to E. conversidens, further stating that this species was closely related to South American horses, a relationship that has been suggested by other researchers (e.g., [8, 22]).

Three equid species have been recognized from the late Pleistocene deposits of Cedral, San Luis Potosí, Mexico, based on differences in size [23–27], metapodial proportions [24, 26], and features of the occlusal enamel pattern of the third and fourth upper premolars [25]. The large and medium-sized species have been tentatively identified as E. mexicanus (Hibbard), 1955 (originally described by Hibbard [7] as E. (Hesperohippus) mexicanus), and E. conversidens, respectively [23, 24, 26, 27]. The taxonomic identification of the smaller equid has been more problematic. Alberdi et al. [23] originally identified it as Equus sp. A, whereas Melgarejo-Damian and Montellano-Ballesteros [24] assigned it to Equus tau Owen, 1869. Recently, Alberdi et al. [27] have designated a new species, Equus cedralensis, for this material.

1.2 The American Southwest: New Mexico, western Texas and northern Chihuahua, Mexico

A number of important late Pleistocene fossil localities are known from New Mexico and western Texas, all of which have yielded large numbers of equid specimens, including Blackwater Draw Loc. 1, Dry Cave, Dark Canyon Cave, and U-Bar Cave in New Mexico as well as Scharbauer Ranch and Quitaque Creek in Texas. Blackwater Draw Loc. 1, New Mexico, is the type locality of the Clovis cultural complex and a large collection of bones as well as lithic artifacts and other cultural remains have been retrieved from this site [28]. The equid material from Blackwater Draw has been assigned to a variety of species. Stock and Bode [29] considered that only one species, E. excelsus Leidy, 1858, was represented in the material they studied. In contrast, Quinn [30] identified four taxa from this locality: Asinus conversidens, Equus caballus caballus Linnaeus, 1758, E. caballus laurentius Hay 1913, (originally described by Hay (1913) as E. laurentius), and E. midlandensis Quinn, 1957, a new species he named based on specimens from Scharbauer Ranch, Texas. Quinn [30] adhered to the proposal of dividing modern and fossil species of Equus into four genera, which consists of Equus for horses, Asinus for African asses and the domestic donkey, Onager for Asiatic asses, and Hippotigris for zebras. Lundelius [31], working with a larger sample from the Gray Sand unit of Blackwater Draw, agreed with Quinn [30] in identifying A. conversidens, but following a broader definition of the genus Equus he referred it to E. conversidens. In addition, Lundelius [31] reassigned the material identified by Quinn [30] as E. caballus laurentius to E. niobrarensis Hay 1913, whereas he reassigned the specimens identified as E. midlandensis and E. caballus caballus to E. scotti Gidley, 1900. A few years later, Harris and Porter [22] concluded that the specimens studied by Stock and Bode [29], Quinn [30], and Lundelus [31], with the exception of E. conversidens, appear to be assignable to E. niobrarensis. Recently, Harris [32] has revised his opinion and now considers E. niobrarensis a junior synonym of E. scotti.

In his study of fossil Equidae from Texas, Quinn [30] also examined, among other material, specimens from Scharbauer Ranch. Like Blackwater Draw Loc. 1, this locality has also yielded lithic artifacts and other cultural remains [33]. Quinn [30] identified some of the equid specimens he studied as A. conversidens and proposed a new species of large and stout-legged equid which he named E. midlandensis. This latter species is not considered to be valid by various authors. Lundelius [31] regarded E. midlandensis a synonym of E. scotti, Harris and Porter [22] proposed that it was synonymous with E. niobrarensis, whereas Winans [4] considered it a synonym of E. mexicanus, a species she thought was distinct from E. scotti. Winans [10] later proposed the name E. laurentius for the species group of E. mexicanus; however, it was recently shown that the holotype of E. laurentius belongs to a historic domestic horse and it is therefore a junior synonym of this species [34], a conclusion that had previously been expressed in the literature (e.g., [4, 6, 11, 12, 35]).

Dalquest [36] described an assemblage of fossils from a small tributary of Quitaque Creek, western Texas. Most of the equid remains collected were from a species of small horse, which Dalquest [36] identified as E. cf. conversidens based on similarities with specimens from the Valley of Mexico referred to E. conversidens by Hibbard [7]. There were also some remains of a larger horse slightly smaller than the average size of comparable elements identified as E. scotti from the Seymour formation of Knox County, Texas, which Dalquest [36] reported as Equus sp.

In one of the first studies that applied multivariate morphometrics to fossil equids, Harris and Porter [22] studied the equid remains from Dry Cave, southeastern New Mexico. They concluded that E. conversidens and E. niobrarensis were represented in the material they studied and also referred some specimens to E. occidentalis sensu Merriam, (1913), E. scotti, and a small zebrine species, which they called E. sp. A. Winans [10] studied specimens from Dry Cave and assigned them to the small, stout-legged species group of E. alaskae and the large, stout-legged species group of E. laurentius. Harris [32] has revised his interpretation of the equid remains from Dry Cave and currently recognizes E. conversidens, E. scotti (which he considers the senior synonym of E. niobrarensis), E. occidentalis (sensu Merriam, (1913); for the largest specimens in the fauna), E. sp. A (a small zebrine species), and a single partial upper tooth identified as E. francisci.

In addition to Dry Cave, there are several other cave sites from the American Southwest that have yielded equid remains. Two of these are U-Bar Cave, located in southwestern New Mexico, and Dark Canyon Cave, found south of Dry Cave, in southeastern New Mexico. Harris [37, 38] studied the fossil material collected from U-Bar Cave, separating it into mid-Wisconsin and late Wisconsin ages. He listed three equid species for the mid-Wisconsin of U-Bar Cave, namely E. conversidens, E. cf. niobrarensis, and E. cf. occidentalis, whereas for the late Wisconsin he considered that only E. conversidens and E. cf. niobrarensis were represented in the fauna [37]. Harris [32] maintains the same interpretation of the equid material from U-Bar Cave, with the exception that he considers E. niobrarensis to be a junior synonym of E. scotti. Regarding Dark Canyon Cave, Lundelius [22, 39] tentatively identified E. conversidens and E. scotii from this site, whereas Harris and Porter [22] referred to E. conversidens a small collection of equid remains from this locality housed at the University of Texas at El Paso. In his dissertation, Tebedge [40] described the fauna collected from excavations undertaken in the East Side Pocket of Dark Canyon Cave. He decided to identify the equid material as Equus sp. because of the confused nomenclature of Pleistocene equids [40].

The Vertebrate Paleobiology Collection of the University of Texas at El Paso houses specimens from different parts of Chihuahua, Mexico. Among these is a small collection of fossils from the ranch of Santa Barbara, located 9 km north of Villa Ahumada, northern Chihuahua [41]. In their report of this fossil locality, Comadurán et al. [41] identified the presence of Mammuthus sp. and Equus sp. Harris [32] examined the fossil material from this locality and identified the equid remains as Equus francisci.

1.3 Natural Trap Cave, Wyoming

The Natural Trap Cave fossil locality, Wyoming, has yielded thousands of vertebrate remains [42]. In a report of the excavations at Natural Trap Cave, Martin and Gilbert [43] mentioned the presence of three horse species for the equid material known at the time. They remarked that the most common species was a small, stilt-legged equid likely referrable to Hemionus [43], a group which has been treated as a genus or subgenus of Equus and which includes the extant Asiatic asses. Martin and Gilbert [43] indicated that the other two species were less abundant and that one of them is assignable to the subgenus Amerhippus. Winans [10] studied several specimens from Natural Trap Cave and assigned them to the species group of E. alaskae, which generally includes small, stout-legged horses. In a recent study using ancient mtDNA, Weinstock et al. [13] concluded that two clades were present at this locality, a caballine and a stilt-legged clade, each possibly representing a single species. The study by Weinstock et al. [13] further indicated that the stilt-legged clade is endemic to North America and that the presence of slender metapodials is, therefore, a convergent feature with extant Asiatic asses. In contrast, Eisenmann et al. [44] proposed that four equid species are represented in the material from Natural Trap Cave: a caballine, E. cf. conversidens, and a large and small Amerhippus, both with slender metapodials. According to Eisenmann et al. [44], the small Amerhippus is the most common species in the fauna.

1.4 Alberta, Canada: The Edmonton area gravel pits and Wally’s Beach site

The equid material from the Edmonton area gravel pits has not been described in detail. Burns and Young [45] listed two types of horses, which they referred to Equus cf. conversidens and E. cf. niobrarensis. Weinstock et al. [13] obtained ancient mtDNA from a large sample of specimens from the gravel pits around the Edmonton area. All of the specimens they studied were found to belong to the caballine clade, suggesting that only one species was represented in the sample they studied [13].

The archaeological-paleontological site of Wally’s Beach located in southern Alberta is remarkable in that it is the only known late Pleistocene horse and camel kill and butchering locality in North America [46–48]. Seven butchered horses were recovered associated with lithic artifacts [46]. McNeil [49] compared the equid material collected from Wally’s Beach to specimens from the Yukon Territory identified as E. lambei, as well as a skull from Papago Springs Cave, Arizona, identified by Skinner [50] as E. conversidens. McNeil [49] assigned the equid material from Wally’s Beach to E. conversidens and noted several differences between the Wally’s Beach sample and the sample of E. lambei, particularly in skull morphology and dentition.

1.5 The Yukon Territory: Bluefish Caves

The Bluefish Caves are located in northern Yukon Territory above the Arctic circle and have yielded, in addition to a large collection of vertebrate remains, some lithic artifacts and butchered bones, as well as other cultural evidence that extends from the late glacial to the LGM or possibly even earlier [51–53]. Burke and Cinq-Mars [54, 55] studied the horse remains from Bluefish Caves identified as E. lambei Hay, 1917. These authors documented the range of variation in cheek tooth morphology [54] and also constructed mortality profiles for each of the three caves [55]. Equus lambei has been identified as an onager [30], as a member of the genus Asinus [56, 57], and as a caballine equid [6, 58–61]. This species has also been considered a junior synonym of E. ferus caballus [4, 6] or E. asinus [4], it has also been assigned to the E. alaskae species group of Winans [10], and has been regarded as a possible subspecies of E. niobrarensis by Azzaroli [11, 12]. Burke and Cinq-Mars [54] concluded that E. lambei was a caballine horse, based on the morphology of the cheek tooth dentition. Weinstock et al. [13] successfully extracted, amplified, and sequenced, ancient mtDNA from one horse metatarsal from Bluefish Cave 3. Although the specimen was only identified as Equus sp., the sequence obtained by Weinstock et al. [13] placed it within the caballine group. Other late Pleistocene sites in Beringia have yielded fossil material of a horse with slender metapodials [62, 63], a feature that is present in extant hemione (Asiatic ass) species. However, molecular analysis of slender metapodials from the Yukon by Weinstock et al. [13] have placed this equid outside of the modern hemiones as a distinct species. These studies suggest that at least two species of Equus where present in Beringia during the late Pleistocene.

2.1 Linear morphometrics

2.1.1 Measurements and sample size.

The cheek tooth dentition of equids consists of three upper (P2, P3, and P4) and three lower (p2, p3, and p4) premolars, as well as three upper (M1, M2, and M3) and three lower (m1, m2, and m3) molars on each side of the dentition. We gathered linear measurements of the tooth crown dimensions of upper and lower cheek teeth using a Mitutoyo digital caliper with a measuring range of 0–150 mm, a resolution of 0.01 mm, and an accuracy of 0.003 mm. To account for measurement error, we took every measurement three times and used the mean of these measurements in all statistical analyses. All of the specimens were measured by the same researcher (CIB-O). The measurements collected are partially based on the methodology published by Eisenmann et al. [71]. For each cheek tooth studied, we measured the length and width at a crown height of 2 cm (Fig 3A and 3C). In the case of the lower teeth, these measurements were taken 2 cm above from the bifurcation of the protoconid and hypoconid columns measured on the buccal side of the tooth (Fig 3B), whereas in the upper teeth these measurements were taken 2 cm above from where the mesostyle ends on the buccal side of the tooth (Fig 3D). The occlusal dimensions of a tooth change as it wears down (e.g., [5, 72]) and taking measurements at a set tooth height compensates for this ontogenetic variation, especially in teeth with similar size and degree of hypsodonty. A potential drawback to this approach is that depending on the developmental stage of the tooth (and its state of preservation) sometimes a thin layer of cementum is present around the tooth crown where the measurements are taken. However, since this phenomenon was observed to occur in every sample studied, we do not expect it to be a source of systematic error.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3.

Lower (A and B) and upper (C and D) third premolars showing the measurements collected for lower and upper cheek teeth. Transverse width (Tr) and anteroposterior length (Ap) measurements are shown in figures A and C. These measurements were taken at a tooth crown height of 2 cm. In the lower teeth the measurements were taken 2 cm above from the bifurcation of the protoconid and hypoconid columns measured on the buccal side of the tooth (B). In the upper teeth the measurements were taken 2 cm above from where the mesostyle ends on the buccal side of the tooth (D). Figures A and B: left p3 (DhPg-8 3437.1). Figures C and D: right P3 (DP 3850).

https://doi.org/10.1371/journal.pone.0183045.g003

A total of 1,454 cheek teeth were measured (738 upper and 716 lower teeth; Tables A–H in S1 File). Most of the specimens measured were isolated teeth, although there were some tooth series and some complete or partial dentaries and maxillaries. We determined the side and tooth position for every individual tooth using the criteria presented by Bode [73] and Eisenmann et al. [71]. The upper and lower third and fourth premolars (P3/p3 and P4/p4) are sometimes difficult to distinguish, as is the case for upper and lower first and second molars (M1/m1 and M2/m2). As a result, Eisenmann et al. [71] suggest combining upper P3 and P4 as well as lower p3 and p4 into a single category, respectively. The same suggestion applies to the upper M1 and M2 as well as the lower m1 and m2 teeth [71]. Consequently, we arranged the data into eight tooth categories: Upper P2, P3/P4, M1/M2, and M3; and lower p2, p3/p4, m1/m2, and m3. For cases in which different tooth positions of the same tooth category were associated (i.e., they belong to the same individual), for instance a p3 and a p4 of the same tooth series, we selected one of the two specimens at random and excluded the other from the analysis. For situations in which the left and right sides of the same tooth position were associated, for example a right P2 and a left P2, we measured both specimens, calculated the average, and used the average measurements in the statistical analysis. The final data sets for each of the five geographic regions and each tooth category studied are shown in Tables A–H in S1 File.

2.2 Geometric morphometrics of the occlusal enamel pattern

The occlusal enamel pattern of equids is complex and has been used to varying degrees in the taxonomy of these ungulates. The occlusal surface and dimensions of hypsodont equid cheek teeth change with age as the teeth wear down (e.g., [5, 72, 77, 78]). This large ontogenetic variation has brought into question the utility of the cheek teeth in the determination of equid species. However, when comparing specimens at similar stages of wear, the enamel pattern can be taxonomically informative [25, 79]. In this study, we examined teeth with a tooth height representing 30–40% of the maximum crown height. This approximately corresponds to a crown height equivalent to the width of the tooth (or the length of the tooth for the lower teeth) measured at a tooth height of 2 cm as indicated above in the section on linear morphometrics.

2.2.1 Image acquisition.

We photographed specimens showing the selected stage of wear using a SONY Cyber-shot DSC-H9 digital camera. When taking the photograph, the occlusal surface of the tooth was oriented perpendicular to the camera lens. In addition, we placed a scale bar oriented parallel to the occlusal surface on the lingual side of the tooth for the upper teeth (Fig 4) and on the buccal side for the lower teeth (Fig 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Occlusal surface of a P3 (LACM 192/18109) showing the 24 landmarks used in the analysis.

Refer to the main text and Table C in S2 File for a description of the landmarks and details about how they were digitized.

https://doi.org/10.1371/journal.pone.0183045.g004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Digitized double knot (metaconid, linguaflexid, and metastylid) of a lower p4 (KU 50629) showing the 50 landmarks used in the analysis.

Refer to the main text for details about how the landmarks were digitized.

https://doi.org/10.1371/journal.pone.0183045.g005

Typically, fossil assemblages present teeth with varying degrees of wear; therefore, restricting the analysis to teeth with equivalent stages of wear reduces the effective sample size. In order to increase the sample size available for study we relied on X-ray Computed Tomography to digitally section specimens at the selected stage of wear. Due to limitations in CT-scanning time, we did not scan all of the tooth positions, but rather concentrated on the third and fourth premolars, as these tooth positions had proven to be taxonomically useful in a previous study of the occlusal enamel pattern [79]. As a result, our analysis focused solely on the upper P3/P4 and lower p3/p4 tooth categories.

In total, 139 specimens were CT-scanned, 64 upper P3/P4 and 75 lower p3/p4. All of the specimens were scanned using a SkyScan 1173 high-resolution micro-CT scanner at the Department of Comparative Biology and Experimental Medicine, University of Calgary. Three-dimensional (3-D) surface models of the teeth were created in AMIRA 5.3.3 and digitally sectioned at the selected tooth height. The specimens were not sectioned perpendicular to the long axis of the tooth, but rather the cutting plane was aligned with the occlusal surface of the tooth. Thus, the cutting plane was inclined lingually and mesially to varying degrees for the upper teeth and buccally and (generally) mesially on the lower teeth. The sectioned 3-D models were then oriented with the enamel pattern perpendicular to the screen and an image along with a scale bar was obtained.

In cases where there were images of associated specimens, for example left and right P3 of the same individual, we chose one of the two specimens at random for digitization. We also reflected all of the left teeth in the data set in order to have all of the specimens in the same orientation. We then renamed every image with a four digit identifier generated at random, with the objective of mixing the sample of images and removing the identity of each image to minimize any biases during digitization. The final data sets (including both photographed and CT-scanned specimens) consisted of 144 upper P3/P4 teeth (Table A in S2 File and data set in S3 File) and 128 lower p3/p4 teeth (Table B in S2 File and data set in S4 File).

2.3 Ancient mtDNA analysis

3.1 Linear morphometrics

In all analyses for both the upper and the lower cheek teeth, the first principal component (PC 1) accounted for over 87% of the variation in the data (Tables 1 and 2). The factor loadings indicate that this component reflects variation in size, with larger specimens showing more positive scores (Tables 1 and 2). Plotting the PC scores by geographic region for each tooth category reveals the presence of one to three size groups (Fig 6; Figs A–J in S2 File). These size differences do not correspond to sexual dimorphism as extant equid species and monospecific quarry samples of fossil species do not show sexual dimorphism in the cheek tooth dimensions investigated here [104].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Principal component plots and histograms of PC 1 scores resulting from PCA of the linear measurements of upper M1/M2 teeth.

The specimens come from five geographic regions of the Western Interior of North America: Mexico (C, Cedral; J, San Josecito Cave), the American Southwest (A, Algerita Blossom Cave; D, Dry Cave; F, Fresnal Canyon; K, Dark Canyon Cave; H, Nash Draw; I, Isleta Cave No. 2; L, Blackwater Draw; M, Big Manhole Cave; P, Imperial; R, Scharbauer Ranch; S, Salt Creek; U, U-Bar Cave); Wyoming (N, Natural Trap Cave); Alberta (E, Edmonton area gravel pits; W, Wally’s Beach), and the Yukon Territory (B, Bluefish Caves). A lower case “a” beside the specimen symbol indicates a tooth that yielded ancient mtDNA (EQ16 from Dry Cave). An asterisk (*) beside the specimen symbol denotes a tooth associated (i.e., it belongs to the same individual) with a specimen from which ancient mtDNA was obtained (teeth associated with EQ3 from Dry Cave, EQ9 from Natural Trap Cave, EQ29 and EQ43 from Wally’s Beach, EQ44 and EQ47 from Bluefish Caves). The dark line in the histograms corresponds to the kernel density estimation. Table C in S1 File lists all of the specimens included in this analysis.

https://doi.org/10.1371/journal.pone.0183045.g006

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Eigenvalues, percentage variance, and factor loadings for the principal components resulting from PCA of the linear measurements of the upper teeth (Ap = anteroposterior length; Tr = transverse width), taken at a crown height of 2 cm.

https://doi.org/10.1371/journal.pone.0183045.t001

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Eigenvalues, percentage variance, and factor loadings for the principal components resulting from PCA of the linear measurements of the lower teeth (Ap = anteroposterior length; Tr = transverse width), taken at a crown height of 2 cm.

https://doi.org/10.1371/journal.pone.0183045.t002

The upper and lower cheek teeth from northeastern Mexico tend to plot into three size groups: large, medium, and small (Fig 6; Figs A and B in S2 File). Moreover, the distribution of the specimens along PC 1 (Fig 6; Figs K and L in S2 File) statistically departs from normality in all tooth categories, except M3 and p2 (Table 3).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Results of Shapiro-Wilk test for normal distribution of principal component 1 scores for each tooth category and geographic region studied.

https://doi.org/10.1371/journal.pone.0183045.t003

The specimens from the American Southwest tend to plot into large and medium size clusters, except for the p3/p4 tooth category where there are three small-sized specimens that plot in the same region of the morphospace as the small-sized specimens from northeastern Mexico (Fig 6; Figs C and D in S2 File). Sample sizes for this geographic region are small in four of the eight tooth categories, namely P2, M3, p2, and m3 (Table 3). As a result, greater weight was given to the remaining tooth categories in the interpretation of the Shapiro-Wilk test of normality. The distribution of PC1 scores for P3/P4 and p3/p4 cheek teeth is statistically different from the expected normal distribution, whereas normality is not rejected for the M1/M2 and m1/m2 tooth categories; although, the p-value for the M1/M2 category is marginally greater than 0.05 (Table 3; Fig 6; Figs M and N in S2 File).

The vast majority of the specimens from Natural Trap Cave plot into one cluster that falls in the same region of the morphospace as the medium size cluster from northeastern Mexico and the American Southwest (Fig 6; Figs E and F in S2 File). There are, however, a few specimens that are of larger size, producing a right-skewed distribution of specimens along PC1 in all of the tooth categories except p2 and p3/p4 (Fig 6; Figs O and P in S2 File). Accordingly, the Shapiro-Wilk test is not significant for these two tooth categories (Table 3). The test is also not significant for the m3 tooth category. Significant departures from normality are detected for the remaining five tooth categories (Table 3).

The specimens from Alberta tend to plot on the right side of the graph in the same region of the morphospace as the large specimens from the American Southwest and northeastern Mexico (Fig 6; Figs G and H in S2 File). However, this does not apply to all of the tooth positions, and there are four specimens (one p2, two p3/p4, and one m1) that are smaller in size and that fall in the same region of the morphospace as the medium-sized specimens from Natural Trap Cave, the American Southwest, and northeastern Mexico. The sample size in four of the eight tooth categories (P2, M3, p2, and m3) is small and, therefore, greater weight was given to the interpretation of the Shapiro-Wilk test for the other tooth categories. Normality cannot be rejected for the distribution of specimens along PC1 for the P3/P4, M1/M2, and m1/m2 tooth categories, and the test is marginally not significant for the p3/p4 tooth category (Table 3; Fig 6; Figs Q and R in S2 File).

The specimens from Bluefish Caves form one cluster in all tooth categories (Fig 6; Figs I and J in S2 File). The specimens from this locality show a size range that is intermediate to that of the large-sized specimens from Alberta, the American Southwest, and northeastern Mexico and the medium-sized specimens from Alberta, Natural Trap Cave, the American Southwest, and northeastern Mexico. The Shapiro-Wilk test is not significant for any of the tooth categories, although it is only marginally not significant for the p2 and m3 tooth categories (Table 3; Fig 6; Figs S and T in S2 File).

3.2 Geometric morphometrics of the enamel pattern of upper premolars

There is a statistically significant relationship between shape (as defined by the Procrustes coordinates) and log centroid size (p-value < 0.0001). The regression on centroid size accounts for 6.045% of the total shape variation. Thus, it was necessary to standardize the data by computing the residuals from the regression to remove the shape variation due to allometry. The residuals were then used in further statistical analyses.

The first three Canonical Variates (CV 1 to CV 3) account for 81.84% of the relative between-group variation (Table 4). The different groups are arranged from small to large along CV 1 (Fig 7), largely reflecting the pattern seen in the PCA of the linear measurements. The transformation grids show that negative scores on CV 1 correspond to shallow parastyle-mesostyle and mesostyle-metastyle valleys, bucco-lingually expanded fossettes, relatively short protocones, and mesial displacement of landmark 11 (around the area where the pli caballin is located); the opposite is observed for positive CV 1 scores. The morphological characters associated with negative CV 1 scores are not typically found in extant caballine equids (e.g., [59]). As a result, we identify the groups that have negative CV 1 scores as possessing non-caballine tooth morphologies; these groups are: the small size group from Cedral, Mexico, as well as the intermediate size groups from northeastern Mexico (Cedral and San Josecito Cave), the American Southwest, and Natural Trap Cave. The remaing groups have caballine tooth morphologies (positive CV 1 scores) and these correspond to the large size groups from Cedral, Mexico, the American Southwest, Natural Trap Cave, and Alberta (Edmonton gravel pits and Wally’s Beach), as well as the specimens from Bluefish Caves, Yukon Territory (Fig 7). The second Canonical Axis (CV 2) clearly separates the small specimens from northeastern Mexico and the specimens from Bluefish Caves from the rest of the groups. Negative CV 2 scores reflect a mesial extension of the anterior margin of the protocone and a more prominent mesostyle (landmarks 3 and 4 are more separated from each other); the opposite is seen for positive CV 2 scores. The third Canonical Axis (CV 3) does not clearly separate any of the groups, but arranges the intermediate size groups from south to north: specimens from northeastern Mexico show negative scores, specimens from Wyoming have positive scores, and teeth from the American Southwest have an intermediate position (Fig 8). Examination of the transformation grid for the CV 3 axis, shows that negative scores correspond to a displacement away from the center of the tooth of the pli paraconule (landmark 17), pli postfossette (landmark 14), and landmark 11; the opposite is observed for positive CV 3 scores.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Plot of the first two Canonical Variates resulting from CVA of 24 landmark coordinates of the occlusal enamel pattern of the upper premolars (P3/P4).

Shown on the margins of the graph is the change in tooth shape along each corresponding axis. The groups included in the analysis are: 1) large specimens from Cedral, Mexico (Cl); 2) medium specimens from Cedral (Cm) as well as all teeth from San Josecito Cave (J), Mexico; 3) small specimens from Cedral, Mexico (Cs); 4) large specimens from different localities of the American Southwest (identified by a lower case “l” beside the specimen symbol; refer to Fig 1 for locality names); 5) medium specimens from different localities of the American Southwest (identified by a lower case “m” beside the specimen symbol; refer to Fig 1 for locality names); 6) medium specimens from Natural Trap Cave, Wyoming (N); 7) large specimens from Natural Trap Cave, Wyoming (Nl); 8) large specimens from the Edmonton area gravel pits (E) and Wally’s Beach (W), Alberta; and 9) all of the specimens digitized from Bluefish Caves, Yukon (B). A lower case “a” beside the specimen symbol indicates a tooth that yielded aDNA (these include EQ38 and EQ45 from Bluefish Caves). An asterisk (*) beside the specimen symbol denotes a tooth associated (i.e., it belongs to the same individual) with a specimen from which aDNA was obtained (including teeth associated with EQ3 from Dry Cave, New Mexico, EQ9 from Natural Trap Cave, EQ43 from Wally’s Beach, and EQ44 as well as EQ47 from Bluefish Caves). Table A in S2 File lists all of the specimens included in this analysis.

https://doi.org/10.1371/journal.pone.0183045.g007

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Plot of the first and third Canonical Variates resulting from CVA of 24 landmark coordinates of the occlusal enamel pattern of the upper premolars (P3/P4).

Shown on the margins of the graph is the change in tooth shape along each corresponding axis. Refer to caption of Fig 7 for details on the specimens included in this analysis.

https://doi.org/10.1371/journal.pone.0183045.g008

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Eigenvalues, percentage variance, and cumulative percentage variance of the first five Canonical Variates resulting from CVA of 24 landmark coordinates of the occlusal enamel pattern of the upper premolars (P3/P4).

https://doi.org/10.1371/journal.pone.0183045.t004

The pair-wise permutation tests identified significant differences in the Procrustes distance for all but eight comparisons (Tables 5 and 6). Two of these comparisons concern the large size group from northeastern Mexico, which is not significantly different from the large size groups of the American Southwest and Alberta. Likewise, these last two groups are not statistically different from each other. The medium size group from Natural Trap Cave is not significantly different from the medium size group of the American Southwest. The four remaining pair-wise permutation tests that are non-significant include the large size group from Natural Trap Cave, which has a sample size of only two specimens and, thus, the reliability of these results is questionable.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Procrustes distances among groups for the upper premolars (P3/P4).

https://doi.org/10.1371/journal.pone.0183045.t005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. P-values from permutation tests (10,000 permutation rounds) for Procrustes distances among groups of the upper premolars (P3/P4).

https://doi.org/10.1371/journal.pone.0183045.t006

3.3 Geometric morphometrics of the enamel pattern of lower premolars

The lower premolars show a marginally significant relationship between shape (as defined by the Procrustes coordinates) and log centroid size (p = 0.0437). The regression on centroid size accounts for 2.257% of the total shape variation. As for the case of the upper premolars, the residuals were calculated and used in further statistical analyses.

The first three Canonical Variates (CV 1 to CV 3) account for 71.93% of the relative between-group variation (Table 7). The groups generally plot along CV 1 from small to large (Fig 9), reflecting the same overall pattern observed in the CVA of the upper premolars and the PCA of the linear measurements. Examination of the transformation grids reveals that the CV 1 axis corresponds to a morphological gradient which goes from a caballine double knot, with a deep, U-shaped linguaflexid on the right side of the plot (i.e., CV 1 values greater than 0) to a hemione-like double knot with a shallow and more open, U-shaped linguaflexid on the left side of the plot (i.e., CV 1 values less than 0). Moreover, positive CV 1 scores also reflect a tooth morphology in which the metaconid is “constricted” (i.e., the bucco-distal margin of the metaconid is displaced towards the linguaflexid) and the metastylid is “open” (i.e., the bucco-mesial margin of the metastylid is displaced away from the linguaflexid); the opposite pattern is observed for specimens with negative CV 1 scores. As a result, we identify the groups that have positive CV 1 scores as possessing caballine tooth morphologies; these groups are: the large size groups from Cedral, Mexico, the American Southwest, Natural Trap Cave, and Alberta (Edmonton gravel pits and Wally’s Beach), as well as the specimens from Bluefish Caves, Yukon Territory. The remaing groups have non-caballine tooth morphologies (negative CV 1 scores) and these correspond to the small size groups from Cedral, Mexico, and the American Southwest (a small sample of teeth from northern Chihuahua, Mexico), as well as the intermediate size groups from northeastern Mexico (Cedral and San Josecito Cave), the American Southwest, Natural Trap Cave, and Alberta (a small sample of teeth from the Edmonton area gravel pits) (Fig 9). The small and large groups from northeastern Mexico, the small specimens from the American Southwest, and the specimens from Bluefish Caves all have positive CV 2 scores and are clearly separated from the rest of the groups along this axis (Fig 9). The transformation grids show that positive CV 2 scores correspond to a relatively rounded metastylid, whereas specimens with negative CV 2 scores reflect a triangular metastylid. The specimens from Bluefish Caves, the large and medium size specimens from Natural Trap Cave and the small groups from northeastern Mexico and the American Southwest all have negative CV 3 scores and plot separately from the remaining groups in the data set (Fig 10). Negative CV 3 scores correspond to a bucco-lingually compressed metastylid, whereas positive scores reflect a bucco-lingually expanded metastylid.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Plot of the first two Canonical Variates resulting from CVA of 50 landmark coordinates of the double knot (metaconid-linguaflexid-metastylid complex) of the lower premolars (p3/p4).

Shown on the margins of the graph is the change in shape along each corresponding axis. The groups included in the analysis are: 1) large specimens from Cedral, Mexico (Cl); 2) medium specimens from Cedral (Cm) as well as all teeth from San Josecito Cave (J), Mexico; 3) small specimens from Cedral, Mexico (Cs); 4) large specimens from different localities of the American Southwest (identified by a lower case “l” beside the specimen symbol; refer to Fig 1 for locality names); 5) medium specimens from different localities of the American Southwest (identified by a lower case “m” beside the specimen symbol; refer to Fig 1 for locality names); 6) small specimens from Villa Ahumada (Vs) and Highway 45 (Gs), Chihuahua, Mexico; 7) medium specimens from Natural Trap Cave, Wyoming (N); 8) large specimens from Natural Trap Cave, Wyoming (Nl); 9) large specimens from the Edmonton area gravel pits (E) and Wally’s Beach (W), Alberta; 10) medium specimens from the Edmonton area gravel pits (Em), Alberta; and 11) all of the specimens digitized from Bluefish Caves, Yukon (B). A lower case “a” beside the specimen symbol indicates a tooth that yielded aDNA (these include EQ1 from Dry Cave, New Mexico, EQ4 from the Edmonton area gravel pits, EQ13 as well as EQ22 from Natural Trap Cave, EQ43 from Wally’s Beach, and EQ39, EQ48, and EQ50 from Bluefish Caves). An asterisk (*) beside the specimen symbol denotes a tooth associated (i.e., it belongs to the same individual) with a specimen from which aDNA was obtained (including teeth associated with EQ42, EQ51, and EQ53 from Bluefish Caves). Table B in S2 File lists all of the specimens included in this analysis.

https://doi.org/10.1371/journal.pone.0183045.g009

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 10. Plot of the first and third Canonical Variates resulting from CVA of 50 landmark coordinates of the double knot (metaconid-linguaflexid-metastylid complex) of the lower premolars (p3/p4).

Shown on the margins of the graph is the change in shape along each corresponding axis. Refer to caption of Fig 9 for details on the specimens included in this analysis.

https://doi.org/10.1371/journal.pone.0183045.g010

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 7. Eigenvalues, percentage variance, and cumulative percentage variance of the first five Canonical Variates resulting from CVA of 50 landmark coordinates of the double knot (metaconid-linguaflexid-metastylid complex) of the lower premolars (p3/p4).

https://doi.org/10.1371/journal.pone.0183045.t007

In contrast to the upper premolars, there were fewer pair-wise comparisons in which the Procrustes distance between groups was statistically different (Tables 8 and 9). This is partially due to the inclusion of groups with small sample sizes, namely the small groups from northeastern Mexico and the American Southwest, the large size group from Natural Trap Cave, and the medium size group from Alberta. Of the remaining groups in the data set, the most relevant differences are: 1) the Bluefish Caves group is significantly different from all other groups; 2) the medium size group from northeastern Mexico differs from the medium size group of Natural Trap Cave as well as the large size groups from northeastern Mexico, the American Southwest, and Alberta; 3) the medium size group from the American Southwest is statistically different from the large size groups of Alberta and the American Southwest; and 4) the medium size group from Natural Trap Cave differs from the large size group of Alberta.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 8. Procrustes distances among groups for the lower premolars (p3/p4).

https://doi.org/10.1371/journal.pone.0183045.t008

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 9. P-values from permutation tests (10,000 permutation rounds) for Procrustes distances among groups of the lower premolars (p3/p4).

https://doi.org/10.1371/journal.pone.0183045.t009

3.4 Ancient mtDNA

We were able to extract and amplify ancient mtDNA from 22 of 50 late Pleistocene specimens we sampled (Table D in S2 File), including specimens of the different morphological groups identified in the morphometric analyses, except for the small non-caballine equids from northeastern Mexico and the American Southwest. The DNA sequence data we generated was submitted to GenBank (accession numbers KX137124 –KX137148).

The Bayesian phylogenetic analysis using a 588 bp fragment of the HVR I yields the two lineages of late Pleistocene North American equids that have been identified in previous molecular studies [13, 97]: caballine and New World stilt-legged lineages (identified as clades A and B, respectively in Fig 11). The phylogenetic analysis also recovers 16 of the 18 extant horse haplogroups identified by Achilli et al. [96]. The two horse haplogroups that are not recovered in the analysis are haplogroups O and F.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 11. Consensus tree of Bayesian (Markov chain Monte Carlo) phylogenetic analysis displaying relationships between mitochondrial control region (HVR 1) haplotypes of extinct and extant equids, rooted with domestic donkey (Equus africanus asinus (L., 1758)) and Somali Wild Ass (Equus africanus somaliensis (Noack, 1884)) as the outgroup.

The tree was constructed using 588 bp fragments of the HVR I. Posterior probabilities of the major nodes are listed for each of the branches. The groups discussed in the text are indicated by the letters. Tables D and F in S2 File list all of the specimens included in this analysis.

https://doi.org/10.1371/journal.pone.0183045.g011

Interestingly, the medium-sized specimens with a non-caballine tooth morphology from the Edmonton area, Alberta; Natural Trap Cave, Wyoming; Dry Cave, New Mexico; and San Josecito Cave, Mexico fall in the stilt-legged clade. The large-sized specimens with a caballine tooth morphology from Dry Cave, New Mexico and Wally’s Beach site, Alberta; as well as the specimens from Bluefish Caves, Yukon Territory, which also have a caballine tooth morphology cluster within the caballine clade (Fig 11). The caballine specimens from Wally’s Beach site and Dry Cave fall outside of the caballine crown group (clade C in Fig 11). The specimens from Bluefish Caves fall within the caballine crown group and may comprise an extinct horse haplogroup along with specimens from Siberia, Alaska, and other sites in the Yukon Territory (clade D in Fig 11).

4.1 Taxonomic nomenclature and geographic distribution of late Pleistocene equids from the Western Interior of North America

The taxonomy of North American Equus is highly confused and its resolution is beyond the scope of this study, which would require careful re-evaluation of every single holotype. Previous researchers (e.g., [4, 10]) have lamented that several holotypes consist of isolated teeth or partial tooth rows and have questioned the diagnostic value of these elements, regarding the names based on them as nomina dubia. The methodology presented here can be applied to evaluate many of these holotypes. This new research direction can potentially help to clarify the nomenclature of North American Pleistocene equids. Until such a study is completed, the name we use (i.e., Equus cedralensis) for the putative non-caballine species (morphological group 3 above) is considered provisional.

The caballine equid species appears to be conspecific with E. ferus Boddaert, 1785, and this is the name we propose should be assigned to this material. We regard the morphological differences between the enamel pattern of the caballine specimens from Bluefish Caves (morphological group 1) and the caballine specimens from the other geographic regions (morphological group 2) as the product of geographic variation. Equus lambei Hay, 1917, is the name that has been widely applied in the literature for the caballine equid material from Bluefish Caves (e.g., [54, 55]). Winans [10] suggested that E. lambei might be a junior synonym of E. alaskae (Hay), 1913, along with the specimens from San Josecito Cave referred as E. conversidens leoni by Stock [20, 21]. The synonymy with E. alaskae may or may not be correct, but the material from San Josecito Cave is clearly distinct based on the morphological and molecular analyses reported here. Azzaroli [11, 12] considered E. lambei as a valid species, but thought it was probably a subspecies of E. niobrarensis Hay, 1913. The caballine equid remains we studied from Alberta, Natural Trap Cave, the American Southwest, and northeastern Mexico have been identified under several names including Equus caballus caballus Linnaeus, 1758, E. caballus laurentius Hay 1913, E. excelsus Leidy, 1858, E. laurentius Hay, 1913, E. mexicanus (Hibbard), 1955, E. midlandensis Quinn, 1957, E. niobrarensis, and E. scotti Gidley, 1900 [4, 10, 22–24, 26, 27, 29–32]. Extant caballine equids have historically been assigned to two species, E. caballus Linnaeus, 1758, and E. przewalskii Poliakov, 1881, but several studies point to the inclusion of E. przewaslkii in E. caballus (e.g., [106–109]). The name E. ferus was proposed by Gentry et al. [110] to differentiate wild caballines from domestic forms (i.e., E. caballus). The International Commission on Zoological Nomenclature has approved this proposal [111, 112], and “implementation of the ruling means that names based on wild populations will continue to be used for wild species and will include those for domestic forms if these are considered conspecific” (p. 649 in [112]). We follow this proposal in the present study; however, we point out that there is still some disagreement about the status of E. ferus as a wild rather than a feral horse [113].

The non-caballine equid species (morphological group 4) whose ancient mtDNA corresponds to the NWSL clade of Weinstock et al. [13] is referred to Equus conversidens Owen, 1869. This name has been widely used in the literature of North American late Pleistocene equids, although not without some confusion (see Scott [114], for different morphological concepts of this species). The morphological and molecular data sets we analyzed for this species included several specimens studied by previous authors and which were consistently identified as E. conversidens, including material from San Josecito Cave (e.g., [11, 12, 20, 21, 114]), specimens from Dry Cave [22, 32], U-bar Cave [37], Scharbauer Ranch [30], and Blackwater Draw [31]. Alberdi et al. [27] also report the presence of E. conversidens from Cedral, Mexico; however, several of the specimens that Alberdi et al. [27] identify as E. cedralensis we identify here as E. conversidens. The specimens from Wally’s Beach, Alberta, were identified as E. conversidens by McNeil [49]; however, this assignment is not supported by the morphological and molecular analyses of the specimens from this site included in our study. The results show that the specimens from Wally’s Beach are caballine equids and are, therefore, identified as E. ferus. The NWSL equid from Natural Trap Cave, Wyoming, has been identified under different names, including Amerhippus sp., E. alaskae, and Hemionus sp. [10, 43, 44], but our results suggest that it is not distinct morphologically nor genetically from the NWSL equid of the American Southwest and Mexico; thus, it is here re-identified as E. conversidens. The same is suggested by the phylogenetic analysis for the NWSL equid from Beringia. The proposal put forward by different authors (e.g., [8, 12, 22]) regarding the close phylogenetic affinity of E. conversidens to South American equids of the subgenus Amerhippus (sometimes regarded as a distinct genus (e.g., [44])), based primarily on specimens from San Josecito Cave, is not supported by the molecular analysis. The specimens of Equus (Amerhippus) neogeus cluster well within the caballine clade (Fig 11) as it was originally reported by Orlando et al. [98]. Nevertheless, this should be tested further with additional mitochondrial and nuclear DNA data.

Fossil material referred here to Equus conversidens was recognized in four of the five geographic regions studied. It is well represented in northeastern Mexico (including Cedral and San Josecito Cave), the American Southwest (i.e., Algerita Blossom Cave, Blackwater Draw, Dark Canyon Cave, Dry Cave, Lubbock Lake, Quitaque Creek, Salt Creek, Scharbauer Ranch, and U-Bar Cave), and Natural Trap Cave, Wyoming. This species is much less common in Alberta, where it was identified based on at least four specimens from the Edmonton area gravel pits, but not from Wally’s Beach, and it was not found in the material examined from Bluefish Caves, Yukon. The presence of this species in the Edmonton area gravel pits is further supported by the association of some of the specimens studied (right and left p2 as well as left p3) as part of a partial dentary (RAM P98.5.480) in which all of the incisors lack an infundibulum (a funnel-like cup of enamel filled with cementum). Our examination of partial mandibles and mandibular symphyses with lower incisors from San Josecito Cave (e.g., LACM 18404, 18383, 18802, 120758, 18644) and those identified as E. conversidens from Dry Cave (UTEP 22–955, 26–1064) by Harris and Porter [22] and Harris [32] revealed that all of the incisors lack an infundibulum. In contrast, the mandibles and partial mandibles we observed with associated lower incisors that we assign to E. ferus from Bluefish Caves (e.g., CMH MgVo-2 B3-3-23, MgVo-2 C3(E)-3-19, MgVo-2 H6-3-7, MgVo-3 85–95, MgVo-3 85–76, MgVo-3 85–64, MgVo-3 M-9-83), Wally’s Beach (RAM DhPg-8 876.1, DhPg-8 863, DhPg-8 3437.2), the Edmonton area gravel pits (RAM P97.11.2A), Dry Cave (UTEP 22–1657), Salt Creek (UTEP 34–5), and Scharbauer Ranch (TMM 998–1) have an infundibulum on the first and second lower incisors and this feature is more variable on the third lower incisors. This pattern is certainly consistent with the results obtained for the molecular and morphological analyses of the cheek teeth; nevertheless, the sample size represented by these specimens is not adequate to fully document the frequency of this morphological trait in each species and further study is required. Eisenmann [115] has noted that the frequency of infundibula in the lower incisors of modern equid species can show important intraspecific variation. Moreover, as with other morphological characters of the enamel pattern of equid teeth, the morphology of the infundibulum changes as the tooth wears down until it completely disappears; therefore, the assessment of this character has to take into consideration the stage of tooth wear.

The taxonomic assignment of the small non-caballine equid (morphological group 3) from Cedral, northeastern Mexico, and northern Chihuahua, Mexico, here grouped with the American Southwest samples, is not completely clear. Morphologically, it appears to represent a separate species, but this needs to be validated with the sequencing and analysis of ancient DNA. Alberdi et al. [27] considered that the small equid from Cedral represents a new species, which they named Equus cedralensis, but the enamel pattern of the premolars as well as the tooth dimensions are comparable to those of E. tau Owen, 1869. The maxillary figured and described by Owen [116] (designated the lectotype of E. tau by Mooser and Dalquest [117]) has the third premolar damaged, but the fourth premolar shows many of the traits found in the group 3 teeth from Cedral identified in the geometric morphometric analysis: mesostyle and parastyle not prominent, shallow parastyle-mesostyle valley (the mesostyle-metastyle valley is not preserved in Owen’s [116] specimen), and the region of the occlusal enamel corresponding to landmark 11 displaced mesially (CV1 transformation grid in Fig 7). Other morphological traits commonly present in the group 3 teeth from Cedral and shared with the cheek teeth figured by Owen [116] are a straight (flat) lingual border of the protocone and the absence of a pli caballin. All of the features mentioned above are also present in the holotype of E. francisci figured by Lundelius and Stevens [105] and Eisenmann et al. [44]. The holotype of E. tau has been lost [117]; nevertheless, different researchers have assigned material from a variety of localities in Mexico and the United States to this species (e.g., [9, 24, 117]), some of which may or may not correspond to the species described by Owen [116]. Recently, Eisenmann et al. [44] suggested a partial skull from the Cedazo fossil assemblage, central Mexico, as the neotype of E. tau. Our examination of this specimen confirmed that the teeth are within the size range of the group 3 specimens and those figured by Owen [116], but they are in an advanced stage of wear and the premolars are too worn for reliable comparisons of their enamel pattern with the teeth in our analyses. The most common morphological concept of E. tau in the literature is that of a small-sized equid with slender metapodials (e.g., [9, 44]) and, as a result, some researchers have synonymised E. francisci with E. tau (e.g., [9]). It is not clear, however, whether E. tau possessed slender metapodials as the material described by Owen [116] was not associated with metapodials or any other postcranial elements. The same is true for the neotype proposed by Eisenmann et al. [44]. Although a slender metatarsal of the size range expected for the neotype skull of E. tau has been recovered from Cedazo [44, 117], its association with the skull cannot be firmly established as neither possess precise stratigraphic information. Stratigraphic studies at several of the localities that have produced the Cedazo fossil assemblage identified different fossiliferous strata, which range biochronologically from the latest Blancan to the Rancholabrean NALMAs (early to late Pleistocene) [118, 119]. This work suggests that what was originally described as the Cedazo local fauna by Mooser and Dalquest [117] actually includes specimens of different ages. It is, therefore, possible that the proposed neotype skull of E. tau and the small slender metatarsal are from different strata. According to Alberdi et al. [27], Equus cedralensis does not possess slender metapodials. Until the exact taxonomic status of E. tau and other small North American equids (e.g., E. littoralis Hay, 1913; E. achates Hay and Cooke, 1930) is clarified we provisionally refer to the morphological group 3 equid as E. cedralensis.

Extant equid species are mostly allopatric, except for small areas where two species are known to coexist (e.g., [120–122]). In contrast to the modern distribution of equids, the fossil record shows that throughout much of the history of the group in North America the coexistence of two or more species was the norm rather than the exception [104]. This pattern may have extended into the late Pleistocene. The coexistence of the species identified here as E. ferus and E. conversidens in some areas of the Western Interior is suggested by stratigraphic and biochronologic information, with associated radiocarbon dates, from localities such as Dry Cave [38], U-bar Cave [38], and Blackwater Draw [28, 31]. At Cedral, Mexico, these two equids co-occur with E. cedralensis [27] and recent paleodietary reconstructions suggest that resource partitioning may have facilitated their coexistence [123]. The restriction of E. cedralensis to lower latitudes of the Western Interior may seem surprising compared to the distribution of E. ferus and E. conversidens; however, our study evaluated specimens from a paleontologically short time interval (primarily mid- to late-Wisconsin glacial stage, ca. 50,000 to 10,000 radiocarbon years BP) and it is possible that this species had a wider distribution at other times. Alternatively, E. cedralensis may have been a resident of lower latitudes throughout most of its evolutionary history. Some North American Pleistocene mammals appear to be restricted to southern localities including pronghorn (Capromeryx spp. and Stockoceros conklingi), gomphotheres (Cuvieronious spp.), capybaras (Hydrochoerus spp. and Neochoerus spp.), glyptodonts (Glyptotherium spp.), camelids (Palaeolama mirifica), and pampatheres (Pampatherium mexicanum and Holmesina spp.) [1–3].