Scholarly article on topic 'Cats in the forest: predicting habitat adaptations from humerus morphometry in extant and fossil Felidae (Carnivora)'

Cats in the forest: predicting habitat adaptations from humerus morphometry in extant and fossil Felidae (Carnivora) Academic research paper on "Biological sciences"

0
0
Share paper
Academic journal
Paleobiology
OECD Field of science
Keywords
{""}

Academic research paper on topic "Cats in the forest: predicting habitat adaptations from humerus morphometry in extant and fossil Felidae (Carnivora)"

li BioOne"

RESEARCH EVOLVED

Cats in the forest: predicting habitat adaptations from humerus morphometry in extant and fossil Felidae (Carnivora)

Author(s): Carlo Meloro, Sarah Elton, Julien Louys, Laura C. Bishop, and Peter Ditchfield

Source: Paleobiology, 39(3):323-344. 2013. Published By: The Paleontological Society DOI: http://dx.doi.org/10.1666/12001 URL: http://www.bioone.org/doi/full/10.1666/12001

BioOne (www.bioone.org) is a nonprofit, online aggregation of core research in the biological, ecological, and environmental sciences. BioOne provides a sustainable online platform for over 170 journals and books published by nonprofit societies, associations, museums, institutions, and presses.

Your use of this PDF, the BioOne Web site, and all posted and associated content indicates your acceptance of BioOne's Terms of Use, available at www.bioone.org/page/ terms_of_use.

Usage of BioOne content is strictly limited to personal, educational, and non-commercial use. Commercial inquiries or rights and permissions requests should be directed to the individual publisher as copyright holder.

BioOne sees sustainable scholarly publishing as an inherently collaborative enterprise connecting authors, nonprofit publishers, academic institutions, research libraries, and research funders in the common goal of maximizing access to critical research.

Paleobiology, 39(3), 2013, pp. 323-344 DOI: 10.1666/12001

Cats in the forest: predicting habitat adaptations from humerus morphometry in extant and fossil Felidae (Carnivora)

Carlo Meloro, Sarah Elton, Julien Louys, Laura C. Bishop, and Peter Ditchfield

Abstract.—Mammalian carnivores are rarely incorporated in paleoenvironmental reconstructions, largely because of their rarity within the fossil record. However, multivariate statistical modeling can be successfully used to quantify specific anatomical features as environmental predictors. Here we explore morphological variability of the humerus in a closely related group of predators (Felidae) to investigate the relationship between morphometric descriptors and habitat categories. We analyze linear measurements of the humerus in three different morphometric combinations (log-transformed, size-free, and ratio), and explore four distinct ways of categorizing habitat adaptations. Open, Mixed, and Closed categories are defined according to criteria based on traditional descriptions of species, distributions, and biome occupancy Extensive exploratory work is presented using linear discriminant analyses and several fossils are included to provide paleoecological reconstructions.

We found no significant differences in the predictive power of distinct morphometric descriptors or habitat criteria, although sample splitting into small and large cat guilds greatly improves the stability of the models. Significant insights emerge for three long-canine cats: Smilodon populator, Paramachairodus orientalis, and Dinofelis sp. from Olduvai Gorge (East Africa). S. populator and P. orientalis are both predicted to have been closed-habitat adapted taxa. The false "sabertooth" Dinofelis sp. from Olduvai Gorge is predicted to be adapted to mixed habitat. The application of felid humerus ecomorphology to the carnivoran record of Olduvai Gorge shows that the older stratigraphic levels (Bed 1,1.99-1.79 Ma) included a broader range of environments than Beds II or V, where there is an abundance of cats adapted to open environments.

Carlo Meloro* and Sarah Elton.** Hull York Medical School, University of Hull, Loxley Building, Cottingham Road Hull HU6 7RX, UK. ""Corresponding author. Present address: Dipartimento di Scienze della Terra, Universita degli Studi di Napoli, Federico II, Naples, Italy. E-mail: carlo.meloro@hyms.ac.uk. ""Present address: Department of Anthropology, Durham University, Durham, U.K. Julien Louyst and Laura C. Bishop. Research Centre in Evolutionary Anthropology and Palaeoecology, School of Natural Sciences and Psychology, Liverpool John Moores University, Byrom Street, Liverpool L3 3AF, U.K. tPresent address: School of Earth Sciences, University of Queensland, Brisbane, Australia Peter Ditchfield. Research Laboratory for Archaeology and the History of Art, School of Archaeology, University of Oxford, Dyson Perrins Building, South Parks Road, Oxford OX1 3QY, U.K.

Accepted: 8 January 2013 Published online: 18 March 2013

Supplementary materials deposited at Dryad: 10.5061/dryad.s587t

Introduction

Adaptations in the functional morphology of the postcranial skeleton can be powerful indicators of locomotion and habitat exploitation. For fossil species, whose behavior cannot be observed directly, identifying such adaptations and linking them to habitat are important aspects of paleobiological reconstruction. This approach also informs paleoecological and paleoenvironmental reconstruction, with these ''ecomorphological'' methods shedding light not only on the animals themselves but also on the environments they inhabited. Numerous ecomorphic studies, focused mostly on bovids from Plio-Pleistocene African paleontological sites (Kappelman 1988; Plummer and Bishop 1994; Kappelman et al. 1997;

© 2013 The Paleontological Society. All rights reserved.

DeGusta and Vrba 2003, 2005a,b; Kovarovic and Andrews 2007; Plummer et al. 2008), have informed paleohabitat reconstruction. Although other taxa, such as primates (Elton 2001, 2002, 2006), marsupials (Bassarova et al. 2009), and suids (Bishop 1999), have been subject to similar analyses, terrestrial carnivo-rans (fissiped Carnivora) are generally under-represented in ecomorphic studies (Gonyea 1976; Lewis 1997). Conventionally, it is assumed that morphological diversity in the carnivorans reflects adaptations to specific functions (e.g., foraging and feeding, posture) more than the environment they occupy (Ewer 1973; Van Valkenburgh 1985,1987,1988,1989, 1999, 2007; Biknevicius and Van Valkenburgh 1996; Anyonge 1996; Janis and Wilhem 1993;

0094-8373/13/3903-0001/$1.00

Garland and Janis 1993; Carrano 1999; Farlow and Pianka 2002; Wroe et al. 2005; Meloro 2011a,b; Samuels et al. 2012; Walmsley et al. 2012). Because they occupy large geographic ranges and high trophic levels, carnivorans tend to be more eurybiomic—able to exploit numerous habitats and biomes—than other mammalian clades (Hernandez Fernandez and Vrba 2006). This reinforces the largely unexplored notion that carnivorans are "gen-eralists'' in their skeletal adaptations to habitat and hence have limited value when included in studies that aim to reconstruct paleohabitat.

Neglecting carnivoran fauna when undertaking ecomorphic-based paleoenvironmental reconstructions may exclude important information about local and regional habitats and how these are exploited by different members of mammalian communities. It is thus important to address whether the carnivoran skeleton can yield sufficiently detailed information about functional morphological adaptations related to particular habitats. Variations in habitat preferences within the felids suggest the utility of this group. Within the genus Panthera, there are obvious differences in habitat exploitation between the extant lion (Panthera leo), which tends to hunt in an open, savanna environment, and the tiger (Panthera tigris), which is more restricted to tropical and temperate forested areas. This indicates that even though carnivorans are eurytopic and eurybiomic, niche differentiation does occur. It can further be inferred that if this differentiation has reasonably deep evolutionary roots, there may be morphological adaptations, even if subtle, to these different habitats. Indeed, distinct skeletal metrics that correlate with habitat exploitation in large carnivorous predators have already been identified and used to explore adaptation to habitat in both extant and extinct taxa (Lewis 1997; Meloro 2011b). These have rarely been followed up with morphometric surveys that specifically examine the correlations between skeletal morphology and habitat adaptations in terrestrial carnivorans (but see Polly 2010), but several studies have indicated the strong relationship between appendicular skeleton morphometry and locomotion or behavior (Anyonge 1996; Andersson and

Werdelin 2003; Andersson 2004; Schutz and Guralnick 2007; Polly 2008; Polly and MacLeod 2008; Meachen-Samuels and Van Val-kenburgh 2009; Lewis and Lague 2010; Walmsley et al. 2012).

Here, we explore the relationships between the functional morphology of the carnivoran postcranial skeleton and habitat preferences, focusing on a single family of fissiped carniv-orans, the Felidae. We develop models based on modern species and apply these to fossil felids. The felids, or cats, are a speciose and widespread family of "hypercarnivores" (Ewer 1973; Gittleman 1985; Martin 1989; Kitchener 1991; Turner and Anton 1997; Sunquist and Sunquist 2002). Although this hyper-carnivory has resulted in relative dental homogeneity within the family (Holliday and Steppan 2004; Meloro and Raia 2010), other features do differ according to the prey on which they specialize (Christiansen 2008; Slater and Van Valkenburgh 2008; Meachen-Samuels and Van Valkenburgh 2009; Meloro 2011a). The most exceptional skull and post-cranial morphologies, seen in the extinct saber- and dirk-tooth cats (distinguished by extremely long canine teeth), are possibly a result of extreme adaptations to a specialized hunting technique (Christiansen 2008; Slater and Van Valkenburgh 2008; Anton et al. 2004, 2005; Palmqvist et al. 2007; McHenry et al. 2007; Meachen-Samuels and Van Valkenburgh 2010; Meachen-Samuels 2012). Felids vary greatly in size, with the smallest members (such as the black-footed cat, Felis nigripes) having body masses under 2 kg and the largest extant forms (such as the tiger, Panthera tigris) weighing as much as 300 kg (Kitchener et al. 2010). Some extinct taxa, such as the dirk-tooth Smilodon populator, were likely to have been even larger than this, exceeding 400 kg (Christiansen and Harris 2005). This diversity in body mass is reflected in locomotion, with smaller taxa generally being much more arboreal than bigger forms (Gittleman 1985; Kitchener 1991; Turner and Anton 1997; Sunquist and Sunquist 2002; Kitchner et al. 2010; Samuels et al. 2012). Felids also exploit an array of habitats, commensurate with their near-cosmopolitan distribution. Many species are adapted to a broad range of habitats (e.g.,

the leopard, Panthera pardus, is a typical habitat generalist that can be found in woodlands as well as deserts), whereas others occur only in association with specific environmental conditions (e.g., the Andean cat, Leopardus jacobita, occurs only in association with rocky outcrops in the arid zones of the high Andes, typically above 4200 m) (Macdonald et al. 2010). This diversity of habitats seems to be reflected in a variety of adaptations in the appendicular skeleton, suggesting that the latter evolved in response to the former (Kitchener et al. 2010).

We restrict our analyses to a single bone, the humerus. The humerus has been shown in primates to be highly informative about locomotor adaptations and habitat preferences (Elton 2001, 2002, 2006), and forelimb bone proportions (radius/humerus length) have been used in previous studies of felids and carnivorans to distinguish adaptation to different habitats (Gonyea 1978; Lewis 1997; Meachen-Samuels and Van Valkenburgh 2009; Meloro 2011b; Samuels et al. 2012). The humerus is one of the three long bones of the forelimb, and articulates proximally with the scapula, providing information about shoulder function including rotation, extension and flexion, and distally with the radius and ulna, reflecting elbow flexion and extension. The shape of the distal portion of the humerus discriminates cursorial from non-cursorial Carnivora, and Andersson and Werdelin (2003) argued that this feature might be indicative of adaptations to open/closed habitats. Cursorial carnivores are generally expected to be better adapted to open environments (Janis and Wilhem 1993; Samuels et al. 2012) and this generalization applies also to Felidae (Walmsley et al. 2012). Thus, considering the ecological diversity of the Felidae (Macdonald et al. 2010) we expect the mor-phometry of the humerus to correlate with habitat adaptations.

Our focus on a single bone does not imply that only this bone may be informative, but rather aims to identify its potential for paleobiological and paleoenvironmental reconstruction. Associated skeletons and skeletal regions are rare in the fossil record, so any method that aims to reconstruct the paleobi-

ology of fossil specimens must take this into account. Most ecomorphic studies focus on single bones (see Polly 2010), and some on epiphyses only, given that these are the long bone segments most likely to be preserved (e.g., Elton 2001, 2002, 2006). We therefore present a broad range of statistical analyses designed to improve the resolution of existing methods.

We also aim to identify an objective habitat classification for use in ecomorphic analyses. It is usual for three or four habitat categories to be defined a priori (Kappelman et al. 1997; Bishop 1999; Elton 2001, 2002; DeGusta and Vrba 2003, 2005a,b; Plummer et al. 2008; Louys et al. 2012), although some studies have used as many as seven (Kovarovic and Andrews 2007). Notwithstanding the relatively large number of studies correlating habitat adaptations and long bone morphology, there is no consensus about how to categorize large mammals in discrete and distinct habitats objectively. Most studies rely on reviews of biology and ethology to categorize the most common environments exploited by different species. These are often defined as Open (for example, grassland), Mixed (mixture of grassland and tree cover), and Closed (forest). In her study of the Plio-Pleistocene East African carnivore guild, Lewis (1997) assigned carniv-oran species to these three habitat types, defining Mixed as having around 20% canopy cover, Open as having less than this, and Closed as having more than 20%. An alternative advocated by some authors (e.g., Hernandez Fernandez 2001; Hernandez Fernandez and Pelaez-Campomanes 2003) is to categorize environmental preferences by number of biomes occurring in a species' geographical range. However, few published studies have addressed the issue of habitat categorization in detail. Here, we explore how to define habitat categories, quantifying presence/absence in particular environments and using number of biomes overlapped by the geographical range of a taxon to examine a species' environmental preferences.

Early ecomorphic studies (e.g., Kappelman 1988) used ratios as a means of size-correcting morphometric data. Later studies questioned this approach (DeGusta and Vrba 2005a,b;

Kovaric and Andrews 2007; Louys et al. 2012), arguing that simple linear measurements were equally informative as ratios in the discriminant analyses that form the basis of ecomor-phic studies. Given that previous studies of carnivorans similar to ours used only ratios (Van Valkenburgh 1987; Lewis 1997; Samuels et al. 2012), here we examine the utility of simple raw measurements versus ratios and residuals (another common way of generating ''size free'' data) as predictors of habitat preference.

In short, we seek to assess whether the humeral morphometry of one family of carnivorans, the felids, allows the recovery of useful information about habitat exploitation. We assess methods surrounding ecomorphic reconstruction, in order to examine particular types of scaling methods and modern comparative data sets and to determine whether the way in which habitat is categorized for taxa in a modern comparative sample influences analytic results. In addition, we investigate whether it is possible to recover accurate information from highly fragmentary material, using data from the epiphyses of modern specimens as a proxy for the data collected from incomplete fossils. Finally, we use the methods we develop to reconstruct the habitat preferences of three extinct felids, Paramachair-odus orientalis, Smilodon populator, and Dinofelis sp., as well as those of taxa represented by fragmentary fossil material from different stratigraphic intervals of the hominin East African fossil site Olduvai Gorge, Tanzania. This approach provides an example of how felid humerus ecomorphology can be used to inform paleoenvironmental reconstruction.

Materials and Methods

Sample Size

Complete and incomplete humeri belonging to both extant and extinct members of the Felidae, housed in the Natural History Museum London (BMNH) London; Royal Museum for Central Africa (RMCA), Tervuren; National Museum of Scotland (NMS) Edinburgh; and Kenya National Museum (KNM), Nairobi, were included in the ecomorphological analyses (Supplementary Table 1). For each mod-

ern specimen, taxonomy was reassessed following species accounts in the IUCN Red List (IUCN 2009). When accurate geographic information was available, modern specimens belonging to species with large geographic ranges—the wild cat (Felis silvestris), lion (Panthera leo), leopard (Panthera pardus), and tiger (Panthera tigris)—were assigned to subspecies (Table 1). A total of 111 extant specimens across 11 genera were included in the analyses (Supplementary Table 1). Sample size was not equally distributed across taxa. Inevitably, most of the extant sample was biased in favor of trophy-hunted species (e.g., lions and leopards). To get maximum taxo-nomic and hence environmental coverage, non-pathological captive specimens were included (13% of the sample). Several of these specimens derive from captive breeding centers where general conditions for the animal approximate their natural environment (A. Kitchener personal communication 2009). For this study, captivity was determined to be a negligible source of morphometric variation (Supplementary Table 2, but see O'Regan and Kitchener 2005 for caveats). Thirty-one percent of the specimens either had no locality recorded or were only located to a continent; the rest of the sample being wild-caught with good locality data. Approximately half the extant sample could not be assigned to sex; within the rest of the sample, males and females were equally distributed. Sexual dimorphism is generally high in felids, but because it is uncorrelated with habitat adaptation at interspecific level (Gittleman and Van Valkenburgh 1997), we assume it to be a negligible source of morphometric variation. Both sexes were pooled in the analyses.

Two complete specimens (one fossil, one cast) of cats of the subfamily Machairodonti-nae plus five humeri from Olduvai Gorge were analyzed. The sabertooth Paramachairo-dus orientalis (BMNH M8960) is represented by a complete but slightly deformed humerus from Pikermi, Greece (a late Miocene fossil site); the specimen of Smilodon populator, the biggest Pleistocene dirk-tooth cat from South America, was a cast from a complete skeleton housed in the Natural History Museum, London (BMNH). The material from Olduvai

Table 1. Habitat categories and basic ecological data for felid species analyzed (body weight averaged for males plus females, from IUCN 2009; locomotion as in Meachen-Samuels and Van Valkenburgh 2009). Habitat C is summarized as percentage of specimens recorded in Forest or Grassland.

Body weight Habitat C % specimens

Species (kg) Locomotion Habitat A Habitat B Forest Grass Habitat D

Acinonyx jubatus 40.917 Terrestrial Open Open 50 50 Open

Caracal aurata 6.200 Terrestrial Closed Closed 100 0 Closed

Caracal caracal 11.500 Scansorial Open Open 50 50 Mixed

Felis chaus 5.150 Terrestrial Mixed Mixed 100 0 Closed

Felis margarita 2.500 Terrestrial Open Open 0 100 Open

Felis nigripes 1.525 Terrestrial Open Open 0 100 Mixed

Felis silvestris grampia 4.167 Scansorial Closed Mixed 100 0 Closed

Felis silvestris lybica 4.833 Scansorial Open Open 33 67 Mixed

Leopardus geoffroyi 4.350 Terrestrial Mixed Mixed 100 0 Open

Leopardus guigna 2.200 Unknown Closed Closed 100 0 Closed

Leopardus pardalis 10.131 Scansorial Closed Closed 0 100 Closed

Leopardus wiedii 3.200 Arboreal Closed Closed 100 0 Closed

Leptailurus serval 12.250 Terrestrial Open Open 0 100 Mixed

Lynx canadensis 10.025 Terrestrial Mixed Closed 100 0 Mixed

Lynx lynx 20.100 Scansorial Mixed Mixed 67 33 Closed

Lynx pardinus 11.050 Terrestrial Mixed Mixed 100 0 Closed

Lynx rufus 9.300 Scansorial Open Mixed 0 100 Open

Neofelis nebulosa 15.500 Arboreal Closed Closed 100 0 Closed

Panthera leo 150.529 Terrestrial Open Open 20 80 Open

Panthera leo persica 147.500 Terrestrial Open Open 100 0 Open

Panthera onca 79.167 Scansorial Closed Closed 100 0 Closed

Panthera pardus 35.042 Scansorial Mixed Mixed 50 50 Mixed

Panthera pardus fusca 49.667 Scansorial Mixed Mixed 100 0 Mixed

Panthera tigris 169.375 Terrestrial Closed Closed 100 0 Closed

Panthera tigris altaica 243.000 Terrestrial Closed Closed 100 0 Closed

Panthera uncia 42.188 Scansorial Closed Open 0 100 Open

Pardofelis badia 1.950 Unknown Closed Closed 100 0 Closed

Pardofelis marmorata 4.000 Arboreal Closed Closed 100 0 Closed

Pardofelis temminckii 11.750 Scansorial Mixed Mixed 100 0 Closed

Prionailurus bengalensis 5.050 Scansorial Closed Closed 100 0 Closed

Prionailurus planiceps 2.000 Terrestrial Closed Closed 100 0 Closed

Prionailurus rubiginosus 1.350 Unknown Mixed Mixed 100 0 Closed

Prionailurus viverrinus 9.625 Terrestrial Closed Closed 100 0 Closed

Puma concolor 57.125 Scansorial Mixed Mixed 100 0 Closed

Puma yagouaroundi 5.150 Scansorial Closed Mixed 100 0 Closed

Gorge includes two complete humeri belonging to Dinofelis sp. indet D (OLD 74/54, OLD 74/348 [Werdelin and Lewis 2001]) from Bed I measured in the Kenya National Museum (Nairobi), and three distal fragments housed at the Natural History Museum, London: M20240, recorded from DKI 25 IV 35 and tentatively assigned to Panthera sp. from Bed I; M14676, belonging to Panthera leo from Bed II (cf. Leakey 1965); and M14677, classified as Panthera leo from Bed V (Upper Pleistocene). Only one humerus from Bed I (OLD 5067 FLK NII 4, Panthera pardus) was too incomplete to be included here.

Linear Measurements and Error Estimation

Forty linear measurements of the humerus (Table 2) were taken to 0.5 mm by a single

observer (C.M.) using an osteometric board (for greatest bone length), spreading calipers (for physiological length), or Sylvac digital calipers interfaced to a laptop computer. Most measurements were taken on the left humerus. If that was not available, we substituted the right, assuming that because asymmetry was fluctuating and not directional, no systematic bias would be introduced.

Measurement error was calculated by measuring the same specimen of serval (Leptailu-rus serval BMNH 1981.988) three times on separate occasions (cf. DeGusta and Vrba 2003, 2005a,b). Overall, the mean error was less than 5%, consistent with that seen in other studies. The mean error estimate for each measurement is given in Table 2. Given that

Table 2. Measurement error and description for 40 humerus linear measurements. Only absolute values of differences in multiple measurement comparison are reported. Abbreviations: L, length; ML, mediolateral; AP, anterior-posterior.

ID Description Error

L Length 0.12%

PhL Physiological L—from central tip of epiphyses 0.62%

DtL Deltopectoral crest max L 1.32%

DtPhL Deltopectoral crest physiological L 2.57%

Mds ML Midshaft ML 0.10%

Mds AP Midshaft AP 1.77%

APmH Head max AP 0.14%

APartH Head articular surface AP 5.07%

APsH Head shaft AP 1.05%

ML H Head max ML 0.66%

ML artH Head articular surface ML 1.69%

H H Head surface height 0.42%

BcG W Bicipital groove width 4.48%

BcG D Bicipital groove depth 1.45%

GT AP Greater tubercle max AP 1.47%

GT ML Greater tubercle max ML 0.00%

Sb ML Subspinosus scar ML 2.89%

Sb AP Subspinosus scar AP 3.32%

LT AP Lesser tubercle max AP 0.16%

LT ML Lesser tubercle max ML 0.00%

Dst ML Distal epiphysis maximum ML 0.14%

Dst AP1 Distal epiphysis medial articular surface AP 5.15%

Dst AP2 Distal epiphysis lateral articular surface AP 1.27%

TrL Trochlea max L 1.52%

CpL Capitulum max L 2.90%

TrAP Trochlea AP at the midpoint 0.11%

Cd L1 Trochlea superior-inferior maximum L 1.15%

Cd L2 Trochlea superior-inferior medium L 1.85%

Cd L3 Trochlea superior-inferior minimum L 5.24%

Pj_Tr Projection of trochlea in vertical plane 2.13%

Dst art ML Distal articular surface ML 2.38%

Of ML Olecranon fossa ML 0.72%

Of H Olecranon fossa height 8.22%

Of Pr Olecranon fossa projection 0.64%

PrTb L Pronator tubercle L 3.16%

UmF L Ulnar medial fossa L 1.00%

UlF ML Ulnar lateral fossa ML 5.78%

UlF AP Ulnar lateral fossa AP 12.23%

UlF pj Ulnar lateral fossa depth 5.20%

ExC L Extensor carpii scar L 1.14%

the sample represents cats having a wide range of body mass (1-200 kg), the measurements were log-transformed for statistical analyses. This enables an assumption of normality to be met and also scaled the data (cf. Kovarovic and Andrews 2007). Initial examinations showed that when data are log transformed for discriminant analyses, the percentages of correctly classified cases were always higher than when raw data were used.

Habitat Categorization

Four different ways of determining habitat categories were examined and compared (Table 1): (A) presence or absence in particular biomes (based on raw data from Ortolani and

Caro 1996) to assign species to one of three categories (Open, Mixed, Closed); (B) descriptions from the IUCN Cat Specialist Group (IUCN 2009) that were then used to assign species to the above categories; (C) a GIS-based approach to assign each specimen to grassland or forest biome; (D) a similar GIS-based method assigning specimens to open or closed biome.

Presence or Absence in Particular Biomes (A).—We used data from Ortolani and Caro (1996), who recorded the presence or absence of each carnivoran species in a series of broad biomes, to assign each felid species to one of three categories (Open, Mixed, or Closed). The biomes used by Ortolani and Caro (1996) were

temperate forest, tropical forest, grassland, arctic, riparian, and desert. Species were first given a habitat score, by recording presence in tropical or temperate forest as +1 and presence in grassland, arctic, or desert as —1; riparian was not considered because it is generally associated with semiaquatic species like otters and no felids occurred exclusively in this category. Absence was recorded as 0. These scores were summed, with positive values used to indicate preference for Closed environments (presence in forests), zero values preference to Mixed environments (balanced between forest and open environments), and negative scores preference for Open environments. For example, the lion is recorded in grassland (—1) and desert (—1), which sum to —2 and is thus categorized as Open. The European lynx (Lynx lynx) occurs in temperate forest (+1) and grassland (—1), summing to zero, and thus is assigned to Mixed. For several taxa we did not possess the Ortolani and Caro (1996) categorizations, and consequently we based their categorization on the IUCN species description (see B).

Descriptions from the IUCN Specialist Group

(B).—We used descriptions of habitat preferences from the IUCN Cat Specialist Group (IUCN 2009) to help assign each as Open, Mixed, or Closed. For instance, the lion, described as preferring ''open woodland-thick bush, scrub, grass complexes'' was scored as Open. The European lynx, described as preferring ''forested areas'' as well as being present in ''more open, thinly wooded, thick scrub woodland and barren, rocky areas above the treeline, alpine tundra,'' is scored as Mixed. In many cases, categorization using this method correlates with results of method (A) but some differences exist; e.g., the Scottish wild cat (Felis silvestris grampia) is considered as Mixed (coniferous forest + Mediterranean shrubland) using this method, rather than Open, as was the case using method (A).

GIS-based Approach: Grassland versus Forest

(C).—The third method used a specimen-specific rather than a broad species-based approach. Each individual skeleton with accurate locality data (longitude and latitude) was plotted using DIVA-GIS software, version

5.2.0.2. We used open-source shape files showing species-specific range maps for the Carnivora, taken from Greneyer et al. (2006), to check for potential outliers. We also used these maps to identify species centroid coordinates (where the geographic range was continuous and not fragmented), assumed to be a representative locality for captive specimens or for those with no recorded locality. By overlaying the WWF world ecoregion polygon (Olson et al. 2001), which describes 14 ecoregions (plus two categories: rock and ice, and lake), on the range maps, we were able to extract an ecoregion (biome) for each specimen locality. The ecoregions extracted for the sampled localities were ascribed to either ''forest'' (e.g., tropical and subtropical moist, dry broadleaf forest, temperate coniferous forest) or "grassland/shrubland" (e.g., montane grassland and shrubland, tropical and subtropical grassland, savanna and shrub-lands, deserts and xeric shrublands). In this way each individual extant specimen could be confidently ascribed to ''forest'' or ''grass-land/shrubland.''

GIS-based Approach: Open versus Closed (D).—The fourth method builds on the principles of method (C), but instead of simply extracting a single habitat category for each specimen, we used the Xtools of ESRI software ArcGIS 3.2 (ESRI, Redlands, Calif.) to assess the interaction between a species' geographic range and ecoregion polygons to quantify the relative proportion of each biome occupied by an individual taxon. In order to assign species to either Open, Mixed, or Closed, we classified the 14 biomes as either ''forest'' or ''grassland/ shrubland'' as in method C. For each species, the relative percentages of occurrence in forest or grassland biomes were summarized in a ratio that is equal to % occurrence grassland divided by % occurrence forest. Species with a ratio between 0 and 0.9 were classified as Closed, between 0.9 to 1.1 Mixed, and above 1.1 Open. Seventy-two percent of the geographic range of the lion was found in grassland/shrubland biomes and 25% in forest biomes (3% is in the biome ''lake'' or ''rocks & ice,'' which are not considered here), giving a ratio of 2.88. Consequently, all skeletal specimens of lion are classified as

"open." The Eurasian lynx (Lynx lynx), in contrast, has a ratio of 0.63 (38% in grassland and 61% in forest biomes) and is considered Closed.

Scaling Measurements

Three separate data sets, derived from different measurement scaling/size correction methods, were analyzed to examine their relative efficacy. Data set 1 contained log-transformed linear measurements. Data set 2 was obtained by applying univariate regression models of log maximum length versus all the other logged variables, then using the unstandardized residuals as "size-free" variables. Only one of these residuals, log Physiological Length (n = 112, rs = 0.230, p = 0.012) was correlated with log Maximum Length. Data set 3 used log-transformed linear measurements (Table 2) combined in 27 functional ratios (after Bishop 1999; Elton 2001, 2002). Six ratios exhibited no correlation with log Maximum Length but the majority were negatively influenced by humerus length (see Supplementary Table 3). All the ratios were retained for statistical analyses.

We used the software PASW Statistic 18 to perform linear discriminant analysis (LDA) with a forward stepwise procedure (with an F entry probability of p = 0.05). This option allows for the selection of the most appropriate morphometric variables for discriminating preassigned categories (Hair et al. 1998). Although the more relaxed criterion of F = 0.15 used by several previous studies (Kappel-man et al. 1997; Bishop 1999; Elton 2001, 2002; Plummer et al. 2008) generally leads to the inclusion of more variables in the models, the more stringent criterion used here should select a smaller number of the most powerful discriminatory variables, which could be an advantage when applying the methods to fragmentary fossils. To assign fossil specimens, which lack an a priori habitat classification to a category, we used the modern specimens of known habitat as a "training set."

Twelve separate LDAs were therefore undertaken, combining each scaled data set and each habitat categorization method. For all models, we interpreted the validity of LDA on the basis of the Wilks' lambda statistics and

the percentage of correctly classified specimens after cross-validation. In addition, we ran LDA models on a subset of variables that could simulate bias introduced by analyses of fossils. In these cases, we selected a subsample of variables from the proximal or distal epiphyses of specific bones and re-ran the discriminant analyses on that subset (cf. Elton 2001, 2002).

Nested Ecological Analyses

Meloro (2011a) recently demonstrated that LDA model performance can be improved if data sets are split according to ecologically meaningful groups. Because the purpose of our study is to predict habitat adaptation of large fossil species, we performed a nested ecological analysis splitting our felid sample at a body mass threshold of 7 kg, based on the findings of studies by Van Valkenburgh (1985, 1988,1989) and Meloro and O'Higgins (2011). The majority of small cats (<7 kg) tend to hunt prey smaller than themselves and are capable of an arboreal life style (Meachen-Samuels and Van Valkenburgh 2009: Table 1). Consequently, the morphology of their humerus mor-phometry is expected to correlate to specific habitat adaptations differently than in larger species. Lewis and Lague (2010) have also demonstrated that long bone allometry of felids (including the humerus) is better described by a second polynomial regression, suggesting that allometric differences occur between small and larger taxa.

Because the number of ''small'' cat (<7 kg) specimens is relatively low (n = 29), the large number of variables relative to the sample size precluded any meaningful LDAs. However, the data set of large taxa (>7 kg, n = 82 specimens) was reanalyzed separately to test whether LDA on these taxa improved when compared to the overall data set. Because the big cat LDA models performed so much better on data set 1 only (log-transformed data set), we restricted our analyses to this data set.

Sensitivity Analyses

In order to validate the efficacy of our LDA models in making predictions irrespective of unequal taxonomic sample size (Kovarovic et al. 2011), we performed two kinds of sensitivity

analyses. First, we repeated the most accurate LDA after removing all the specimens belonging to a particularly abundant taxon from the original sample. We repeated the LDA by excluding first Felis silvestris grampia (n = 9, the most abundant small felid), then Panthera pardus (n = 12, representative medium-size felid of Mixed habitat), and finally Panthera leo (n = 17, the most abundant large felid).

A second sensitivity analysis was conducted to test for the effect of sample size (number of specimens) or body mass (in grams, log transformed) on percentage of correctly classified specimens for the 32 extant species sampled. On the basis of the results from all the LDA models, nonparametric Spearman correlation was applied to identify positive or negative significant correlations.

Results

Whole Humerus

For the whole humerus data set, all 12 data combinations yielded statistically significant (p < 0.0001) Wilks' lambda values in the linear discriminant analysis (Table 3).

The stepwise procedure reduced the number of variables selected in the models. For models using habitat categorization methods A, B, or D (using Open/Mixed/Closed categories), six to nine variables were selected. For models using method C (which uses only two habitat categories—forest and grassland) the number of variables decreased to four, three, or two depending on the measurement-scaling data set used. Certain variables were selected in most of the LDA models regardless of habitat categorization or measurement-scaling data set (Fig. 1, section ''logged data''). Mediolateral head articular surface, bicipital groove depth, head surface height, and mediolateral subspinosus scar were commonly selected for the proximal end, whereas distal epiphysis maximum mediolateral, olecranon fossa projection, trochlea superior-inferior medial border, and extensor carpi scar length were commonly selected measurements from the distal portion (Fig. 1). These variables also tended to be selected in the models of the ''size-free'' data set (see Fig. 1, section ''size-free''). Data from the proximal humerus were selected more frequently than

those from the distal humerus (Fig 1). In the analyses using ratios, anteroposterior maximum head diameter/mediolateral head articular surface width, the bicipital groove ratio, the olecranon fossa ratio, and three of the trochlea ratios are all frequently selected. (Fig. 2). Both proximal and distal data are equally selected in ratio-based models (Fig. 2).

Table 4 shows percentages of correctly classified specimens for each of the 12 models using ''leave-one-out'' classification in the LDA. A combination of logged linear measurements with habitat categorization method A gives the most consistent and accurate classifications (see also Fig. 3). Residuals and ratios are less consistent and effective overall. Analyses using habitat categorization methods C and D are less accurate than those using the other two methods.

Proximal and Distal Epiphyses

The proximal and distal humerus models were derived from logged linear measurements and ratios; residuals were not used because corrections were based on whole humerus length, which would not be available for fragmented fossil specimens. Sixteen linear discriminant models (using each of the four habitat categorization methods, and two data sets divided into proximal and distal elements) were therefore derived. All but one of the discriminant functions were significant (p < 0.001). The exception was the model using habitat categorization method B (IUCN categorization) and ratios for the distal humerus, in which the second extracted function was nonsignificant (p = 0.32). Again, the stepwise procedure considerably reduced the number of variables selected (Figs. 1, 2). For logged linear measurements, six to nine variables were selected for proximal models (with greater tubercle mediolateral length selected most often) and three to six for distal (with distal mediolateral width and trochlea superior-inferior medial border being frequently selected) (Fig. 1); for ratios the number of variables selected ranged from one—antero-posterior maximum head diameter/mediolateral head articular surface width—and six variables for proximal models, and up to five for distal models (Fig. 2).

Table 3. Wilks' lambda for 12 discriminant analyses based on overall measurements. For all values, p < 0.0001.

Habitat A Habitat B Habitat C Habitat D

Logged (1) 0.241 0.299 0.794 0.271

Size-free (2) 0.297 0.349 0.859 0.385

Ratios (3) 0.368 0.557 0.86 0.415

The percentages of correctly classified specimens differed markedly between the models (Table 5). For the proximal humerus, the model with the best classification rate used habitat categorization method A (Ortolani and Caro habitat) and data set 1 (logged linear measurements). For the distal humerus, habitat categorization methods C (GIS sample based) and D (geographic range based) along with data set 3 (ratios) gave the best results (Table 5). In general, the proximal humerus models gave better classification results than those for the distal humerus.

Big Cats

Discriminant functions were significant at p < 0.05, except for the distal humerus model using logged linear measurements and habitat

categorization method C. The stepwise procedure reduced the number of variables to as few as one (for proximal humerus only, logged greater tubercle mediolateral width for habitat categorization method D) to as many as ten (logged humerus measurements on the whole humerus for habitat categorization method D). Proximal humerus measurements were selected more frequently than distal ones (Fig. 4), with bicipital groove descriptors being one of the most informative in the whole humerus and proximal humerus analyses (see Fig. 4). Mediolateral width at the distal end is the most frequently selected variable in the distal models. The highest percentage of correctly classified specimens is in models for the whole humerus using logged linear measurements, with habitat categorization methods A and D

Logged data MethA MethB MethC MethD Size free MethA MethB MethC MethD Epiphyses logged data MethA MethB MethC MethD

Length

Deltopectoral crest max L Deltopectoral crest physiological L Midshaft ML Midshaft AP Head max AP Head max ML Head articular surface ML Head articular surface AP Head surface height Bicipital groove width Bicipital groove depth Greater tubercle max ML Subspinosus scar ML Subspinosus scar AP Lesser tubercle max ML

Distal epiphysis max ML Distal epiphysis lateral articular surface AP Trochlea AP at the midpoint Capitulum max L Trochlea superior-inferior rain L Distal articular surface ML Ulnar lateral fossa ML Olecranon fossa projection Ulnar lateral fossa depth Extensor carpii scar L

Figure 1. Variable selection by each categorization scheme (=Meth) is indicated in gray. Results from different data sets are reported, including the whole bone LDA models (=logged data), the ''size-free'' LDA model after regressing out bone length, and the epiphyses logged data (proximal and distal separated by line). Abbreviations: Meth A, presence or absence in particular biomes; Meth B, descriptions from IUCN specialist group; Meth C, GIS-based approach, grassland versus forest; Meth D, GIS-based approach, open versus closed; L, length; ML, mediolateral; AP, anterior-posterior.

Ratios whole bone MethA MethB MethC MethD Ratios Epiphyses MethA MethB MethC MethD

Deltopectoral crest max L / Deltopectoral crest physiological L Midshaft ML/AP Head max AP / Head maxML Head max AP / ML Head articular surface Head max AP / Head surface height Head articular surface AP / Head surface height Bicipital Groove width / Groove depth Greater tubercle max AP / Greater tubercle max ML Subspinosus scar ML / Subspinosus scar AP Lesser tubercle max AP / Lesser Tubercle max ML

Distal epiphysis max ML / Capitulum L Trochlea max L / Trochlea AP at the midpoint Trochlea max L / Capitulum L Trochlea max L / Trochlea superior-inferior max L Trochlea superior-inferior max L / Trochlea AP at the midpoint Trochlea superior-inferior max L / Trochlea superior-inferior min L Olecranon fossa height / Olecranon fossa projection Ulnar lateral fossa ML / Ulnar lateral fossa AP Pronator tubercle L / Ulnar medial fossa L Pronator tubercle L / Extensor carpii scar L

FIGURE 2. Variable selection by each categorization scheme (= Meth) is indicated in gray. Results from the ratio data sets (whole bone or epiphyses only) are shown. Abbreviations as in Figure 1.

giving the best results (Table 6) and method C markedly worse. Accuracy declines with the use of proximal and distal elements on their own, with distal models being the least effective classifiers (Table 6).

Summary of Models

When the percentage of correctly classified cases is compared across different data sets without taking different habitat methodologies into account there are no differences in predictive power for the categories Open (Kruskal-Wallis *2(9) = 14.820, p = 0.096) and Mixed (K-W v2(6) = 16.075, p = 0.065). The predictive power for the Closed category changes according to the data set used (K-W

V2(9) = 22.393, p = 0.008), with models based on data set 1 (logged linear measurements) having the highest percentage of correctly classified cases (Fig. 5A). To compare the predictive accuracy of distinct habitat categorizations, we pooled percentage of correctly classified cases from different data sets. There are no statistically significant differences between reclassification rates for the different habitat categorization methods (all K-W p > 0.1, Fig. 5B). In general, the models using habitat criterion D obtain the highest correct prediction rate for the category Open but not for Closed (Fig. 5B). When only big cats are considered, the category Closed is predicted at approximately 90% accuracy when habitat

TABLE 4. Percentage of correctly classified cases after leave-one-out procedure, for 12 discriminant models based on the overall data set.

Habitat A Habitat B Habitat C Habitat D

Logged (1)

Open 74.4 71.4 70.6 69.7

Mixed 75.9 69.4 69.0

Closed 78.6 71.9 61.9 75.0

Size-free (2)

Open 82.1 69.1 64.7 60.6

Mixed 62.1 58.3 62.1

Closed 66.7 75.0 40.5 70.8

Ratios (3)

Open 71.8 71.4 67.7 69.7

Mixed 65.5 44.4 75.9

Closed 64.3 59.4 64.3 60.4

ÎS 0.0-

OOpen Mixed • Closed

X Ungrouped Cases # Group Centroid

• t • ••••

о • •

• о* °

—i— 0.0

Function 1

—I— 2.5

Figure 3. Plot of the discriminant functions extracted for Habitat A (Ortolani and Caro 1996) when using logged linear measurements (number of variables selected = eight).

criterion A is used; this result is never achieved in any other LDA models. Consequently, the Ortolani and Caro (1996) criterion is probably the best for fitting ecomorphology data of big cats (Fig. 5B).

We used UPGMA cluster analysis to summarize differences and similarities in LDA model performance when different data sets (Fig. 5C) and methods (Fig. 5D) were used.

Figure 6 illustrates the major differences between the proximal and distal epiphyses of

three taxa close to the centroids for the different habitat categories (Open, Mixed, Closed). Open-adapted specimens exhibit on average a larger subspinosus scar, a higher head, a larger distal epiphysis and a higher trochlea relative to the Closed-adapted taxa.

Sensitivity Analyses

The LDA models excluding specific taxa were applied to discriminate habitat categorization method A (Ortolani and Caro) for data set 1 (logged data). Excluding Felis silvestris grampia, the LDA yields two significant functions (at p < 0.0001) associated with ten measurements. The percentage of correctly classified cases is highest for Closed (81.8%), followed by Open (79.5%) and then Mixed (69.0%). The exclusion of the leopard also shows little effect on the LDA model (significant after a selection of eight variables) with percentage of correctly classified cases improving for Closed (92.7%) and Mixed (70.6%) but not for Open (66.7%). On the other hand, the exclusion of all lion specimens rendered the LDA model nonsignificant (significant only after adding at least seven lion specimens).

Species show distinct percentages of correctly classified cases depending on the analyses (Table 7). A positive correlation is recorded between number of specimens and log body weight (rspearman = 0.547, p < 0.0001). Number of specimens also correlates positively with percentage of correct cases when habitat A

TABLE 5. Percentage of correctly classified cases after leave-one-out procedure, for 16 discriminant models using proximal or distal humerus region.

Habitat A Habitat B Habitat C Habitat D

Proximal log (1)

Open 66.7 64.3 73.8 81.8

Mixed 65.5 63.9 44.8

Closed 69 65.6 64.7 68.8

Proximal ratio (3)

Open 66.7 64.3 52.4 75.8

Mixed 72.4 47.2 72.4

Closed 57.1 59.4 66.2 43.8

Distal raw (1)

Open 66.7 71.4 57.1 66.7

Mixed 41.4 44.4 62.1

Closed 67.4 30.3 69.6 57.1

Distal ratio (3)

Open 69.2 78.6 76.2 81.8

Mixed 58.6 19.4 34.5

Closed 53.5 39.4 58 49

Logged data MethB MethC MethD

Epiphyses logged only MethA MethB MethC MethD

Length

Deltopectoral crest max L Midshaft ML Head max AP Head articular surface Head shaft AP Head max ML Head articular surface ML Head surface height Bicipital groove width Bicipital groove depth Greater tubercle max ML Subspinosus scar ML Distal epiphysis maximum ML Distal epiphysis lateral articular surface AP Trochlea max L Trochlea AP at the midpoint Trochlea superior-inferior max L Trochlea superior-inferior min L Distal articular surface ML Olecranon fossa ML Olecranon fossa projection Pronator tubercle L Ulnar lateral fossa ML Ulnar lateral fossa depth

Figure 4. Variable selection by each categorization scheme (= Meth) is indicated in gray. Results from the whole log data set and epiphyses log data set of a subsample of large cats are reported. Abbreviations as in Figure 1.

(Ortolani and Caro 1996) is applied to the logged data set (rspearman = 0.362, p < 0.004) and to the ''size-free'' data set (rspearman = 0.522, p < 0.002). For the latter data set there is also a positive correlation between log body weight and percentage of correct cases (rspearman = 0.357, p < 0.044). No other significant correlations emerged, suggesting that sample size and

body mass had no influence on the other LDA models.

Application to Fossil Specimens

Habitat prediction of fossil specimens varies according to the method and data set used (Fig. 7). The sabertooth/ dirk-tooth cats Para-machairodus orientalis and Smilodon populator

TABLE 6. Percentage of correctly classified cases after leave-one-out procedure, for 11 discriminant models based on subsample of large felids.

Habitat A Habitat B Habitat C Habitat D

Log (1)

Open 81.2 85.7 67.6 85.7

Mixed 91.7 90.5 97.5

Closed 95.5 72.7 61.4 80.8

Proximal log (1)

Open 75 71.4 73.5 78.6

Mixed 79.2 61.9 41.7

Closed 63.6 72.7 68.2 19.2

Distal log (1)

Open 75 78.6 n.s 74.3

Mixed 58.3 70.8 57.1

Closed 54.5 57.7 n.s 54.5

Figure 5. A, B, Box and whisker plots of percentage of correctly classified cases summarized in 40 LDA models. Central bar indicates the mean value, top and bottom of box indicate the 25% and 75% quartiles, whiskers indicate maximum and minimum values. C, D, UPGMA tree based on absolute differences in percentage of correctly classified cases. Cophenetic correlation is 0.8549 for data sets (C) and 0.9626 for methods (D). Abbreviations: DistalBig, data set of distal logged data for only big cats; DistalRatio, data set of distal ratios; DistalRaw, data set of distal logged data; ProximalBig, data set of proximal logged data for only big cats; ProxRatio, data set of proximal ratios; ProxRaw, data set of proximal logged data; Ratio, data set of ratios; Raw, data set of logged data; RawBig, data set of logged data for big cats only; Residual, data set ''size-free.'' Habitat definitions as in Figure 1.

are generally predicted to be adapted to Closed habitat. Dinofelis sp., on the other hand, exhibits a broader range of adaptations: depending on the analysis used, it is classified,

by similar proportions, as Open, Mixed, or Closed.

Because all fossil specimens are large felids, the models based on the big cats data set 1

Figure 6. Schematic representation of proximal and distal humerus epiphyses for three taxa that closely resemble in the discriminant function scores the centroid of each habitat categorization. Abbreviations: Sb_ML, subspino-sus scar mediolateral; H_H, head surface height; Dst_ML, distal epiphysis maximum ML; Cd_L3, trochlea superior-inferior minimum length.

(logged linear measurements) using habitat method A are, as expected, the most accurate for the complete specimens. With this classification scheme, both Paramachairodus orientalis and Smilodon populator are classified as Closed, and Dinofelis sp. (OLD 74/01) is classified as Mixed. For the analyses of proximal humerus we chose data set 1/ method A for big cats, which yielded the best rate of cross-validation accuracy (average 72.6 %) when compared with the other methods. Again, P. orientalis and S. populator are classified as Closed, and both Dinofelis sp. from Olduvai Bed I as Mixed. Data set 1/ method B (IUCN classification scheme) had the best rate of classification for the distal humerus of big cats (average 69.03%) and it validates adaptation to Closed habitat in P. orientalis and S. populator; both Dinofelis specimens from Bed I are classified as Open. The distal fragment Panthera sp. (M 20240) from Bed I is predicted as Closed whereas the two distal fragments of lions from Bed II and Bed V are predicted as Open. The fossil humeri of cats from Olduvai Gorge show the breadth of habitats present at Bed I relative to Bed II and Bed V (Upper Pleistocene).

Discussion

Our results clearly indicate that accurate information about habitat exploitation can be recovered from the felid humerus, notwith-

standing the cosmopolitan and eurybiomic nature of the family (Kitchener 1991; Turner and Anton 1997; Sunquist and Sunquist 2002; Kitchener et al. 2010). Single bones, even if fragmentary, can be ecologically informative. Previous work (Gonyea 1976; Anyonge 1996; Lewis 1997; Meloro 2011b) has demonstrated that comparative long bone indices, such as intermembral index, can be used to reconstruct habitat preference and locomotor strategy in a broad range of large carnivores. However, the probability of fossilization for mammalian carnivores is generally low (Dam-uth 1982) and even lower for particularly large felids (Gittleman and Harvey 1982; Gittleman 1985; Turner and Anton 1997), so few relatively complete and associated skeletons are recovered. Developing accurate models based on single bones and bone elements is thus important. The resubstitution rates for the felid humerus in this study are similar to those observed in discriminant analyses from other studies of large mammals, including bovids, suids, and primates (Kappelman et al. 1997; Bishop 1999; Elton 2001, 2002; DeGusta and Vrba 2003, 2005a,b; Kovarovic and Andrews 2007; Plummer et al. 2008). This indicates that carnivorans, important components of past and present biotas, can be as paleoecologically informative as their prey (Hernandez Fernandez 2001; Hernandez Fernandez and Pelae-Campomanes 2003; Hernandez Fernandez et al. 2006; Hernandez Fernandez and Vrba 2006).

One promising avenue of research seeks to combine ecomorphology-based reconstructions of past habitats for different mammalian groups likely to be sympatric and contemporaneous to construct a more holistic picture of the environmental context of ecological communities. This approach was recently used by Polly (2010), who examined calcaneum eco-morphology in different North American carnivoran communities and found strong correlations between community "ecometrics" and environmental variables. Taking this further, using multiple carnivorans and prey species from the same locality may provide a wealth of information about biome and paleoenvironments that cannot be recovered

Table 7. Percentage of correctly classified cases (numbers are in decimals, 1.00 = 100%) for each species based on a selection of LDA analyses performed on the whole humerus measurement data set. Meth = habitat categorization. All of the 100% correct cases are in bold.

Species Logged data Size-free data Ratios

Meth A Meth B Meth C Meth D Meth A Meth B Meth C Meth D Meth A Meth B Meth C Meth D

Acinonyx jubatus 1.00 1.00 0.67 1.00 1.00 1.00 0.67 1.00 1.00 1.00 0.33 0.83

Caracal aurata 0.50 0.50 0.50 1.00 0.00 0.50 1.00 1.00 0.00 0.00 0.50 0.00

Caracal caracal 1.00 1.00 1.00 1.00 1.00 1.00 0.00 0.50 0.50 1.00 0.50 1.00

Felis chaus 0.00 0.00 0.00 0.50 0.00 0.00 0.00 0.00 0.50 0.50 0.50 0.00

Felis margarita 0.00 0.00 0.50 0.50 1.00 1.00 0.50 1.00 0.00 0.00 0.50 0.00

Felis marmorata 1.00 0.00 1.00 1.00 0.00 0.00 1.00 1.00 1.00 1.00 1.00 0.00

Felis nigripes 0.50 0.50 0.00 0.00 1.00 1.00 1.00 0.00 0.00 0.50 0.00 0.00

Felis silvestris grampia 1.00 1.00 1.00 0.89 0.78 0.67 0.78 0.56 1.00 0.67 1.00 0.89

Felis silvestris lybica 0.00 0.33 0.33 0.33 1.00 1.00 0.67 0.00 0.67 0.33 0.33 0.67

Leopardus geoffroyi 0.50 1.00 0.00 0.00 0.50 0.00 0.50 0.50 0.50 0.50 0.50 0.00

Leopardus guigna 0.00 1.00 1.00 1.00 0.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Leopardus pardalis 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00

Leopardus wiedii 0.00 0.00 1.00 1.00 0.00 0.00 1.00 0.00 0.00 0.00 0.00 0.00

Leptailurus serval 0.83 0.50 0.50 1.00 0.67 0.83 0.83 0.83 0.50 0.33 0.67 0.83

Lynx canadensis 1.00 1.00 1.00 1.00 1.00 0.75 0.25 1.00 1.00 0.75 1.00 0.75

Lynx lynx 1.00 0.67 0.67 0.67 1.00 1.00 0.67 0.33 1.00 1.00 0.33 0.00

Lynx pardinus 1.00 0.50 1.00 1.00 1.00 0.50 0.50 1.00 1.00 0.00 1.00 1.00

Lynx rufus 0.00 1.00 0.00 0.00 0.00 1.00 0.00 1.00 1.00 1.00 1.00 0.00

Neofelis nebulosa 1.00 0.67 0.00 0.33 0.67 0.67 0.33 0.67 0.67 0.33 0.67 0.67

Panthera leo 0.88 0.94 0.47 0.88 0.76 0.53 0.18 0.53 0.88 0.82 0.65 0.94

Panthera onca 1.00 0.67 0.67 1.00 1.00 1.00 1.00 1.00 0.00 0.67 1.00 0.33

Panthera pardus 0.92 0.67 0.67 0.58 0.50 0.75 0.75 0.67 0.50 0.17 0.42 0.83

Panthera tigris 0.00 0.25 1.00 0.25 0.75 0.50 1.00 0.75 0.00 0.25 0.75 0.25

Panthera uncia 0.75 0.25 1.00 0.25 0.50 0.00 1.00 0.25 0.75 1.00 1.00 0.50

Pardofelis badia 1.00 1.00 1.00 1.00 0.00 1.00 1.00 0.00 0.00 0.00 1.00 1.00

Pardofelis temminckii 0.00 0.00 0.00 1.00 1.00 0.00 0.00 1.00 1.00 0.00 0.00 1.00

Prionailurus bengalensis 1.00 0.67 1.00 0.33 1.00 0.67 0.67 0.33 1.00 0.67 0.67 1.00

Prionailurus planiceps 1.00 1.00 1.00 1.00 0.00 1.00 0.00 1.00 1.00 1.00 1.00 1.00

Prionailurus rubiginosus 0.00 1.00 0.00 0.00 0.00 0.00 0.00 0.00 0.00 1.00 1.00 1.00

Prionailurus viverrinus 1.00 1.00 1.00 1.00 1.00 1.00 1.00 1.00 0.75 0.75 0.75 0.75

Puma concolor 0.50 0.00 1.00 0.50 0.50 0.50 1.00 1.00 0.50 0.50 0.50 0.00

Puma yagouaroundi 0.00 1.00 1.00 0.00 0.00 0.00 1.00 1.00 0.00 0.00 1.00 0.00

by focusing on specimens from a single species, genus, or even family.

The analyses we present here highlight several methodological issues. Attempts to correct the data using residuals or ratios to take size into account did not increase the accuracy of the statistical models, and indeed in many cases yielded resubstitution rates that were lower than the logged linear measurements. These results therefore lend support to the use of minimally manipulated data in ecomorphic analysis of large mammals (sensu DeGusta and Vrba 2003). Investigations on the teeth of much smaller mammals—voles— suggested similar conclusions, indicating that residual or ratio-based scaling of morphomet-ric data is not always justified, must be validated through experimentation, and must be appropriate to the question being ad-

dressed (Navarro et al. 2004). Using logged linear measurements retains a significant size signal. For ecomorphic reconstruction using felids, it is likely that size, known to be hugely

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0%

□ Open □ Mixed ■ Closed

P. orientalis S. fatalis \Panthera sp. Dinofelissp.

Figure 7. Percentage of predicted cases for fossil specimens. Stratigraphy of specimens from Olduvai is also indicated.

influential both biologically and ecologically in carnivorans (Gittleman 1985; Carbone et al. 1996, 2007) as well as in mammals generally (Damuth and McFadden 1990), is an important explanatory and discriminatory variable when considering habitat adaptation. This has also been noted for primates (Elton 2001,2002) and bovids (Plummer et al. 2008; Louys et al. 2012).

One challenge when retaining a size signal by using logged linear measurements is accounting for allometric effects, especially if scaling factors differ between taxonomic groups or adaptive grades. In the discriminant analyses, resubstitution rates improved on the whole when the modern sample was divided according to size, with 7 kg the threshold between "small" and "big" cats. UPGMA cluster analysis suggests that data set 1 (log linear measurement) for big cats clearly outperforms the others and is placed outside all other clusters (Fig. 5C). This clearly points to scaling differences within the felids, as noted in other studies (Bertram and Biewener 1990; Christiansen 1999; Lewis and Lague 2010). There is no hard-and-fast rule about where the threshold should be drawn, however, with one study on interspecific scaling of the carnivoran postcranium pegging the size threshold at 100 kg (Bertram and Biewener 1990) and another at 50 kg (Christiansen 1999). Another potential threshold is at 25 kg, based on metabolism and hunting behavior, because carnivorans bigger than this tend to kill prey larger than themselves (Carbone et al. 1999, 2007). The choice of threshold should be appropriate to the question being addressed. In our case, 7 kg is the most meaningful threshold, as shown in previous research on locomotor behavior (Van Valken-burgh 1985, 1987). Specifically, arboreality is primarily dictated by body mass (Van Val-kenburgh 1987); in Felidae, most taxa with mean body masses <7 kg tend to use arboreal substrates frequently, whereas bigger forms such as caracal, serval, lynxes, and most of the pantherine cats tend to be more terrestrial, albeit with many taxa being scansorial as well arboreal (Meachen-Samuels and Van Valken-burgh 2009). This does not preclude the exploitation of closed habitats in taxa >7 kg,

as seen in Table 1, but it does highlight a grade shift in the felids. Separation of these two grades results in more accurate discrimination for modern specimens, so a similar separation method (using estimates of mass based on the size of the bone) should be used for fossil felid specimens, as supported by the more consistent allocation of fossils to habitat types in our study when using the big cat modern training set rather than the full modern sample.

There were no statistically significant differences in the discriminatory power of GIS-based habitat categories (methods C and D) versus more traditional ways of assigning taxa to groups (methods A and B) based on species' biology (see Fig. 5B). However, in the discriminant analyses, some habitat categorization methods worked better than others, and this was also evident in the UPGMA cluster analysis (Fig. 5D). For the whole humerus and proximal humerus, method A gave the most accurate and consistent resubstitution results when the whole modern training set was used. When we used the big cats training set, methods A and D yielded the best results. The GIS-based methods (C and D) developed here, which use specimen-specific information based on geographic location, may have a significant drawback in that their accuracy relies on the sample's having a representative geographical distribution. This is often problematic with museum collections, as specimens tend to be drawn from a relatively small number of localities, often chosen for ease of access (Cardini et al. 2007). Smaller or forest specimens may thus be more poorly represented than the larger, charismatic open landscape animals, and that certainly seems to be the case for our sample. The overall habitat classification may thus be skewed. Cluster analysis demonstrates that methods C and D (both GIS-based) stand apart from the others, in having a lower percentage correctly classified. On the other hand, methods A and B perform in a similar way to each other (Fig. 5D).

On balance, therefore, method A, based on data from Ortolani and Caro (1996), who recorded the presence or absence of carnivo-ran species, rather than individual specimens in broad biomes, seems to provide the best

habitat categorization for extant species of big cats.

Unsurprisingly, the jackknifed classification rates were higher for models using the whole humerus than for those employing either the proximal or distal epiphyses separately. The proximal humerus was the most functionally informative in our felid sample, demonstrated both by the dominance of proximal humerus measurements in the stepwise LDA and by the better resubstitution results from the analyses using only the proximal region compared with distal. The proximal humerus is also highly informative in cercopithecid primates, although the distal humerus is also a good discriminator (Elton 2001). In felids, variables such as greater tubercle mediolateral width, bicipital groove depth and width, head articular surface height and width of subspinosus tuberosity (Fig. 6) are constantly selected in most models, and generally have a low degree of measurement error. Similar measurements, albeit contained in ratios, were selected in primate models (Elton 2001), suggesting that they may be functionally informative across a number of mammalian groups. The greater tubercle is the attachment site for the supra and subspinosus muscle complex, part of the rotator cuff of the shoulder. The bicipital groove (or sulcus intertubercularis) is the passage for the tendon of the biceps muscle (Reighard and Jennings 1901; Barone 1980). These muscles facilitate movement between the scapula and the humerus in the vertical and lateral planes. They are implicated in both prey capture and climbing, activities that can be linked to habitat adaptations in the Felidae (e.g., arboreal taxa occur more in forested biomes [Ewer 1973; Nowell and Jackson 1996; Kitchener et al. 2010]).

In felids, models based on the distal humerus were less consistent and accurate, and cluster analysis indicated a clear partition between proximal (better performance) and distal humerus fractions (Fig. 5C). Relatively few measurements were selected in the step-wise procedure for the whole humerus. Measurements that most frequently emerged were maximum mediolateral width of the distal humerus and trochlea superior-inferior medial length (Fig. 6), as well as the size of the

extensor carpii tuberosity to a lesser extent. The error is relatively low for these measurements. Both distal mediolateral width and trochlea superior-inferior length provide information about elbow functional morphology. The elbow joint has been identified as functionally informative in other studies of Carnivora as a whole (Andersson and Werde-lin 2003; Andersson 2004) because of its role in supination as well as its high correlation with body mass. The extensor carpii muscle influences movements of the forepaw that are needed in prey grappling and climbing (Barone 1980).

One major purpose of ecomorphic discriminant analysis is to reconstruct the habitat preferences of extinct taxa. The results from our study give useful insights into the paleoecology of the saber/dirk toothed cats Smilodon populator and Paramachairodus orientalis and the enigmatic false sabertooth Dinofelis sp. The most robust combination of logged linear measurements, habitat categorization method A and the big cat data set, supported strongly the assignment of both saber cats to the Closed category while predicting Dinofelis sp. as "mixed." This is in line with previous research that suggested that the large, specialist stalker S. populator needed to exploit environments with extensive forest cover (Gonyea 1976; Kurten and Werdelin 1990; Christiansen and Harris 2005). Paleoeco-logical data on P. orientalis are generally scanty, although its European counterpart (Pristinosmilus ogygia) has been suggested to be a better climber than the leopard (Salesa et al. 2005, 2006, 2009). Polly and MacLeod (2008) also predicted semidigitigrade locomotor behavior for this big cat, indicating climbing ability and possibly suggesting an adaptation to mixed-closed environments. Lewis (1997) and Werdelin and Lewis (2001) reported adaptations in Dinofelis sp. to tree climbing for carcass transport and possibly to "mixed/closed" habitat exploitation. Our analyses suggest that this taxon was probably more eurybiomic and capable of adapting to a range of environments including open grassland. These findings are supported by the allometric investigation of Lewis and Lague (2010), who recognized that the Dinofelis

humerus had similarities with medium-sized Mixed cats such as the leopard and the puma. Predictions for the fragmentary material of Panthera sp. of Olduvai Gorge Bed I are more enigmatic and possibly suggest that the specimen does not belong to P. leo but to a large cat with adaptations to Closed habitats. The fossil lions included in our analyses are consistently assigned to the Open category, the same category as the modern lion. This suggests that the habitat preferences of African Pleistocene lions were broadly similar to those of extant forms.

Overall, these results strongly encourage the inclusion of felids in paleoenvironmental reconstructions. It clearly emerges that a broad range of habitats existed at Olduvai Bed I, with possibly more forested conditions than in later intervals of Bed II and Bed V (Upper Pleistocene). This conclusion is supported by other studies (Fernandez-Jalvo et al. 1998, Plummer and Bishop 1994; Plummer et al. 2009), and if the ecomorphologies of a broader range of mammalian species were examined more robust results might emerge. Louys et al. (2011) recently demonstrated that broad ecological categories within mammalian communities correlate with percentage of vegetation cover in extant tropical ecosystems; thus future "taxon-free" studies have the potential to predict multidimensional environmental variables.

Inevitably, a small number of inconsistencies emerged in the multiple models used to assign the fossil specimens to habitat category, with some resubstitutions into the Mixed category for both sabertooth cats and lions. This may reflect measurement error or statistical inaccuracy. Alternatively, it may hint at adaptations in extinct animals to environments that are unknown in the modern world. Fossils may inhabit functional categories of their own (Albrecht 1992), which may lead to unexpected results in discriminant analysis. Similarly, habitats in the past may not resemble those seen today. This has been suggested by faunal evidence for Olduvai Bed I, which had a more species-rich woodland than that seen today. It is therefore possible that significant differences exist between the function and structure of past and present ecosystems (Fernandez-Jalvo et al. 1998).

Conclusion

The results presented in this paper show that the use of carnivorans has great promise in aiding paleoecological reconstructions. For felids, the best linear discriminant function model, which predicted habitat categories at a very high rate of accuracy (over 90% for Mixed and Closed categories in jack-knifed classifications), used logged linear measurements in a training set comprising big (>7 kg) species. This suggests that the use of logged linear measurements could be preferable to residual or ratio-based scaled data, although this should be verified on a study-by-study basis and with reference to the question being addressed. It also indicates that a narrower, grade-based modern comparative sample might be more appropriate than one comprising a larger number of species. The choice of habitat categorization method is less straightforward, but a broad, biome-based classification using robust and comparable data from the literature may be the preferable option. The proximal humerus appears to be functionally informative, in line with earlier ecomorphic research (Elton 2001). The high reclassification results for the proximal humerus on its own indicates that even isolated, fragmentary bones can yield useful information about habitat preferences in fossils. The fossil sabertooths Smilodon populator and Para-machairodus orientalis show adaptations to Closed environments whereas Dinofelis sp. from Olduvai Bed I clearly belongs to the Mixed category. This interpretation, together with results emerging from other fragmentary fossil cats, validates previous paleoenviron-mental reconstructions of Olduvai Gorge, with older stratigraphic intervals (Bed I) being characterized by higher abundance of forest-adapted taxa.

Acknowledgments

We are grateful to museum curators and staff of The Natural History Museum of London, National Museum of Scotland, Royal Museum for Central Africa (Tervuren, Belgium) and National Museums of Kenya (Nairobi) for providing access to museum specimens. In particular, we would like to

thank P. Jenkins, L. Tomsett, R. Portela-Miguez, A. Salvador, D. Hills, J. J. Hooker, P. Brewer, and A. Currant (The Natural History Museum, London); A. Kitchener and J. Herman (National Museum of Scotland, Edinburgh); E. Gilissen and W. Wendelen (Royal Museum for Central Africa, Tervuren); E. Mbua, M. Mungu, F. Nderitu and O. Mwebi (Kenya National Museum, Nairobi). A visit to the Royal Museum of Central Africa was supported by the Synthesys grant ''Ecomor-phology of extant African carnivores'' (BE-TAF 4901) to C. Meloro. We are grateful to the Governments of Kenya and Tanzania for kindly providing permission to study Olduvai and Koobi Fora fossil material.

C. P. Egeland and M. Domínguez-Rodrigo kindly provided assistance in verifying strati-graphic interval for Olduvai Gorge fossil material. The associate editor and several anonymous reviewers greatly improved the quality of our manuscript. Our research was generously supported by the Leverhulme Trust project ''Taxon-Free Palaeontological Methods for Reconstructing Environmental Change'' (F/00 754/C).

Literature Cited

Albrecht, G. H. 1992. Assessing the affinities of fossils using canonical variates and generalized distances. Journal of Human Evolution 7:49-69. Andersson, K. 2004. Elbow-joint morphology as a guide to forearm function and foraging behaviour in mammalian carnivores. Zoological Journal of the Linnean Society 142:91-104. Andersson, K., and L. Werdelin. 2003. The evolution of cursorial carnivores in the Tertiary: implications of elbow-joint morphology. Proceeding of the Royal Society of London B 270:S163-S165. Anton, M., M. J. Salesa, J. F. Pastor, I. M. Sanchez, S. Fraile, and J. Morales. 2004. Implications of the mastoid anatomy of larger extant felids for the evolution and predatory behaviour of sabertoothed cats (Mammalia, Carnivora, Felidae). Zoological Journal of the Linnean Society 140:207-221. Antoín, M., A. Galobart, and A. Turner. 2005. Co-existence of scimitar-toothed cats, lions and hominins in the European Pleistocene: implications of the post-cranial anatomy of Homo-therium latidens (Owen) for comparative paleoecology. Quaternary Science Review 24:1287-1301. Anyonge, W. 1996. Locomotor behaviour in Plio-Pleistocene saber-tooth cats: a biomechanical analysis. Journal of Zoology 238:395-413.

Barone, R. 1980. Anatomia Comparata dei Mammiferi Domestici,

Vol. 1. Osteologia. Edagricole, Bologna. Bassarova, M., C. M. Janis, and M. Archer. 2009. The calcaneum— on the heels of marsupial locomotion. Journal of Mammalian Evolution 16:1-23. Bertram, J. E. A., and A. A. Biewener. 1990. Differential scaling of the long bones in the terrestrial Carnivora and other mammals. Journal of Morphology 204:157-169.

Biknevicius, A. R., and B. Van Valkenburgh. 1996. Design for killing: craniodental adaptations of predators. Pp. 393-428 in Gittleman 1989.

Bishop, L. C. 1999. Suid paleoecology and habitat preference at African Pliocene and Pleistocene hominid localities. Pp. 216-225 in T. G. Bromage and F. Schrenk, eds. African biogeography, climate change and human evolution. Oxford University Press, Oxford.

Carbone, C., G. M. Mace, S. C. Roberts, and D. W. Macdonald. 1999. Energetic constraints on the diet of terrestrial carnivores. Nature 402:286-288.

Carbone, C., A. Teacher, and J. M. Rowcliffe. 2007. The costs of carnivory. PLoS Biology 5(2):e22.

Carrano, M. T. 1999. What, if anything, is a cursor? Categories vs. continua for determining locomotor habit in mammals and dinosaurs. Journal of Zoology 247:29-42.

Cardini, A., A.-U. Jansson, and S. Elton. 2007. Ecomorphology of vervet monkeys: a geometric morphometric approach to the study of clinal variation. Journal of Biogeography 34:1663-1678.

Christiansen, P. 1999. Scaling of the limb long bones to bodymass in terrestrial mammals. Journal of Morphology 239:167-190.

--. 2008. Evolution of skull and mandible shape in cats. PLoS

ONE 3(7):e2807.

Christiansen, P., and M. Harris. 2005. Body size of Smilodon (Mammalia: Felidae). Journal of Morphology 266:369-384.

Damuth, J. 1982. Analysis of the preservation of community structure in assemblages of fossil mammals. Paleobiology 8:434-446.

Damuth, J., and B. J. MacFadden. 1990. Body size in mammalian paleobiology. Cambridge University Press, Cambridge.

DeGusta, D., and E. S. Vrba. 2003. A method for inferring paleohabitats from the functional morphology of bovid astragali. Journal of Archaeological Science 30:1009-1022.

--. 2005a. Methods for inferring paleohabitats from discrete

traits of the bovid postcranial skeleton. Journal of Archaeological Science 32:1115-1123.

--. 2005b. Methods for inferring paleohabitats from the

functional morphology of bovid phalanges. Journal of Archaeological Science 32:1099-1113.

Elton, S. 2001. Locomotor and habitat classification of cercopith-ecoid postcranial material from Sterkfontein Member4, Bolt's Farm and Swartkrans Members 1 and 2, South Africa. Palae-ontologia Africana 37:115-126.

--. 2002. A reappraisal of the locomotion and habitat

preference of Theropithecus oswaldi. Folia Primatologica 73:252280.

--. 2006. 40 years on and still going strong: the use of the

hominin-cercopithecid comparison in human evolution. Journal of the Royal Anthropological Institute 12:19-38.

Ewer, R. F. 1973. The carnivores. Cornell University Press, Ithaca, N.Y.

Farlow, J. O., and E. R. Pianka. 2002. Body size overlap, habitat partitioning and living space requirements of terrestrial vertebrate predators: implications for the paleoecology of large theropod dinosaurs. Historical Biology 16:21-40.

Fernandez-Jalvo, Y., C. Denys, P. Andrews, T. Williams, Y. Dauphin, and L. Humphrey. 1998. Taphonomy and palae-oecology of Olduvai Bed-I (Pleistocene, Tanzania). Journal of Human Evolution 34:137-172.

Garland, T., Jr., and C. M. Janis. 1993. Does metatarsal/femur ratio predict the maximal running speed in cursorial mammals? Journal of Zoology 229:133-151.

Gittleman, J. L. 1985. Carnivore body size: ecological and taxonomic correlates. Oecologia 67:540-554.

--, ed. 1989. Carnivore behavior, ecology, and evolution, Vol.

1. Cornell University Press, Ithaca, N.Y.

Gittleman, J. L., and P. H. Harvey. 1982. Carnivore home-range size, metabolic needs and ecology. Behavioral Ecology and Sociobiology 10:52-63.

Gittleman, J. L., and B. Van Valkenburgh. 1997. Sexual size dimorphism in the canines and skulls of carnivores: effects of size, phylogeny, and behavioural ecology. Journal of Zoology 242:97-117.

Gonyea, W. J. 1976. Behavioral implications of saber-toothed felid morphology. Paleobiology 2:332-342.

Hair, J. F., R. E. Anderson, R. L. Tatham, and W. C. Black. 1998. Multivariate data analysis, 5th ed. Prentice Hall, Upper Saddle River, N.J.

Hernandez Fernandez, M. 2001. Bioclimatic discriminant capacity of terrestrial mammal faunas. Global Ecology and Biogeography 10:189-204.

Hernandez Fernandez, M., and P. Pelaez-Campomanes. 2003. The bioclimatic model: a method of palaeoclimatic qualitative inference based on mammal associations. Global Ecology and Biogeography 12:507-517.

Hernandez Fernandez, M., and E. Vrba. 2006. Plio-Pleistocene climate change in the Turkana Basin (East Africa): evidence from large mammal faunas. Journal of Human Evolution 50:595-626.

Hernandez Fernandez, M., M. Alberdi, B. Azanza, P. Montoya, J. Morales, M. Nieto, and P. Pelaez-Campomanes. 2006. Identification problems of arid environments in the Neogene-Quaternary mammal record of Spain. Journal of Arid Environments 66:585-608.

Holliday, J. A., and S. J. Steppan. 2004. Evolution of hyper-carnivory: the effect of specialization on morphological and taxonomic diversity. Paleobiology 30:108-128.

IUCN 2009. The IUCN red list of threatened species, Version 2009.1.

Janis, C. M., and P. B. Wilhem. 1993. Were there mammalian pursuit predators in the Tertiary? Dances with wolf avatars. Journal of Mammalian Evolution 1:103-125.

Kappelman, J. 1988. Morphology and locomotor adaptations of the bovid femur in relation to habitat. Journal Morphology 198:119-130.

Kappelman, J., T. Plummer, L. Bishop, A. Duncan, and S. Appleton. 1997. Bovids as indicators of Plio-Pleistocene paleoenvironments in East Africa. Journal Human Evolution 32:229-256.

Kitchener, A. C. 1991. The natural history of the wild cats. Comstock Associates, Ithaca, N.Y.

Kitchener, A. C., B. Van Valkenburgh, and N. Yamaguchi. 2010. Felid form and function. Pp. 83-106 in D. W. Macdonald and A. J. Loveridge, eds. Biology and conservation of wild felids. Oxford University Press, Oxford.

Kovarovic, K. M., and P. Andrews. 2007. Bovid postcranial ecomorphological survey of the Laetoli paleoenvironment. Journal of Human Evolution 52:663-680.

Kovarovic, K., Aiello, L. C., Cardini, A., and C. A. Lockwood. 2011. Discriminant function analyses in archaeology: are classification rates too good to be true? Journal of Archaeological Science 38:3006-3018.

Kurten, B., and L. Werdelin. 1990. Relationships between North and South American Smilodon. Journal of Vertebrate Paleontology 10:158-169.

Lewis, M. E. 1997. Carnivoran paleoguilds of Africa: implications for hominid food procurement strategies. Journal of Human Evolution 32:257-288.

Lewis, M. E., and M. R. Lague. 2010. Interpreting sabretooth cat (Carnivora; Felidae; Machairodontinae) postcranial morphology in light of scaling patterns. Pp. 411-465 in A. Goswami and A. R. Friscia, eds. Carnivoran evolution: new views on phylogeny, form and function. Cambridge University Press, Cambridge.

Louys, J., Meloro, C., Elton, S., Ditchfield, P., and L. C. Bishop. 2011. Mammal community structure correlates with arboreal heterogeneity in faunally and geographically diverse habitats:

implications for community convergence. Global Ecology and Biogeography 20:717-729.

Louys, J., Montanari, S., Plummer, T., Hertel, F., and L. C. Bishop, 2012. Evolutionary divergence and convergence in shape and size within African antelope proximal phalanges. Journal of Mammalian Evolution. doi: 10.1007/s10914-012-9211-4.

Macdonald, D. W., A. J. Loveridge, and K. Nowell. 2010. Dramatis personae: an introduction to the wild felids. Pp. 3-58 in D. W. Macdonald and A. J. Loveridge, eds. Biology and conservation of wild felids. Oxford University Press, Oxford.

Martin, L. D. 1989. Fossil history of terrestrial Carnivora. Pp. 536568 in Gittleman 1989.

McHenry, C. R., S. Wroe, P. D. Clausen, K. Moreno, and E. Cunningham. 2007. Supermodeled sabercat, predatory behavior in Smilodon fatalis revealed by high-resolution 3D computer simulation. Proceedings of the National Academy of Sciences USA 104:16010-16015.

Meachen-Samuels, J. 2012. Morphological convergence of the prey-killing arsenal of sabertooth predators. Paleobiology 38:114.

Meachen-Samuels, J., and B. Van Valkenburgh. 2009. Forelimb indicators of prey-size preference in the Felidae. Journal of Morphology 270:729-744.

-. 2010. Radiographs reveal exceptional forelimb strength in

the sabertooth cat, Smilodon fatalis. PLoS ONE 5:e11412.

Meloro, C. 2011a. Feeding habits of Plio-Pleistocene large carnivores as revealed by their mandibular geometry. Journal of Vertebrate Paleontology 31:428-446.

-. 2011b. Locomotor adaptations in Plio-Pleistocene large

carnivores from the Italian Peninsula: palaeoecological implications. Current Zoology 57:269-283.

Meloro, C., and P. O'Higgins. 2011. Ecological adaptations of mandibular form in Fissiped Carnivora. Journal of Mammalian Evolution 18:185-200.

Meloro, C., and P. Raia. 2010. Cats and dogs down the tree: the tempo and mode of evolution in the lower carnassial of fossil and living Carnivora. Evolutionary Biology 37:177-186.

Navarro, N., X. Zatarain, and S. Mountuire. 2004. Effects of morphometric descriptor changes on statistical classification and morphospaces. Biological Journal of the Linnean Society 83:243-260.

Nowell, K., and P. Jackson. 2006. Wild cats: status survey and conservation action plan. IUCN Produced Services Unit, Cambridge, U.K.

Olson, D., E. Dinerstein, E. D. Wikramanayake, N. D. Burgess, G. V. N. Powell, E. C. Underwood, J. A. D'amico, I. Itoua, H. E. Strand, J. C. Morrison, C. J. Loucks, T. F. Allnutt, T. H. Ricketts, Y. Kura, J. F. Lamoreux, W. W. Wettengel, P. Hedao, and K. R. Kassem. 2001. Terrestrial ecoregions of the world. Bioscience 51:933-938.

O'Regan, H. J., and A. Kitchener. 2005. The effects of captivity on the morphology of captive, domesticated and feral mammals. Mammal Review 34:215-230.

Ortolani, A., and T. M. Caro. 1996. The adaptive significance of color patterns in carnivores: Phylogenetic tests of classic hypotheses. Pp. 132-188 in J. L. Gittleman, ed. Carnivore behavior, ecology, and evolution, Vol. 2. Comstock, Ithaca, N.Y.

Palmqvist, P., V. Torregrosa, J. A. Perez-Claros, B. Martinez-Navarro, and A. Turner. 2007. A re-evaluation of the diversity of Megantereon (Mammalia, Carnivora, Machairodontinae) and the problem of species identification in extinct carnivores. Journal of Vertebrate Paleontology 27:160-175.

Plummer, T. W., and L. C. Bishop. 1994. Hominid palaeoecology at Olduvai Gorge, Tanzania as indicated by antelope remains. Journal of Human Evolution 27:47-75.

Plummer, T. W., L. C. Bishop, and F. Hertel. 2008. Habitat preference of extant African bovids based on astragalus

morphology: operationalizing ecomorphology for palaeoenvi-ronmental reconstruction. Journal of Archaeological Science 35:3016-3027.

Plummer, T. W., P. W. Ditchfield, L. C. Bishop, J. D. Kingston, J. V. Ferraro, D. R. Braun, F. Hertel, and R. Potts. 2009. Oldest evidence of toolmaking hominins in a grassland-dominated ecosystem. PLoS ONE 4:e7199.

Polly, P. D., and N. Macleod. 2008. Locomotion in fossil Carnivora: an application of eigensurface analysis for morphometric comparison of 3D surfaces. Palaeontologia Electronica 11.2.8A.

Polly, P. D. 2008. Adaptive zones and the pinniped ankle: a 3D quantitative analysis of carnivoran tarsal evolution. Pp. 165-194 in E. Sargis and M. Dagosto, eds. Mammalian evolutionary morphology: a tribute to Frederick S. Szalay. Springer, Dordrecht.

-. 2010. Tiptoeing through the trophics: geographic variation

in carnivoran locomotor ecomorphology in relation to environment. Pp. 374-410 in A. Goswami and A. Friscia, eds. Carnivoran evolution: new views on phylogeny, form and function. Cambridge University Press, Cambridge.

Reighard, J., and H. S. Jennings. 1901. Anatomy of the cat. Henry Holt, New York.

Salesa, M. J., M. Anton, A. Turner, and J. Morales. 2005. Aspects of the functional morphology in the cranial and cervical skeleton of the sabre-toothed cat Paramachairodus ogygia (Kaup, 1832) (Felidae, Machairodontinae) from the Late Miocene of Spain: implications for the origins of the machairodont killing bite. Zoological Journal of the Linnean Society 144:363-377.

-. 2006. Inferred behaviour and ecology of the primitive

sabretoothed cat Paramachairodus ogygia (Felidae, Machair-odontinae) from the Late Miocene of Spain. Journal of Zoology 268:243-254.

-. 2009. Functional anatomy of forelimb in Pristinosmilus

ogygia (Felidae, Machairodontinae, Smilodontini) from the Late Miocene of Spain and the origins of the sabre-toothed felid model. Journal of Anatomy 216:381-396.

Samuels, J. X., J. A. Meachen, and S. A. Sakai. 2012. Postcranial morphology and the locomotor habits of living and extinct carnivorans. Journal of Morphology 274:121-146.

Schutz, H., and R. P. Guralnik. 2007. Postcranial element shape and function: assessing locomotor mode in extant and extinct mustelid carnivorans. Zoological Journal of the Linnean Society 150:895-914.

Slater, J. G., and B. Van Valkenburgh. 2008. Long in the tooth: evolution of sabertooth cat cranial shape. Paleobiology 34:403419.

Sunquist, M., and F. Sunquist. 2002. Wild cats of the world. University of Chicago Press, Chicago.

Turner, A., and M. Anton. 1997. The big cats and their fossil relatives. Columbia University Press, New York.

Van Valkenburgh, B. 1985. Locomotor diversity between past and present guilds of large predatory mammals. Paleobiology 11:406-428.

--. 1987. Skeletal indicators of locomotor behavior in living

and extinct carnivores. Journal of Vertebrate Paleontology 7:62182.

--. 1988. Trophic diversity in past and present guilds of large

predatory mammals. Paleobiology 14:155-173.

--. 1989. Carnivore dental adaptations and diet: a study of

trophic diversity within guilds. Pp. 410-436 in Gittleman 1989.

--. 1999. Major patterns in the history of carnivorous

mammals. Annual Review of Earth and Planetary Sciences 27:463-93.

--. 2007. Deja vu: the evolution of feeding morphologies in

the Carnivora. Integrative and Comparative Biology 47:147-163.

Walmsley, A., S. Elton, J. Louys, L. C. Bishop, and C. Meloro. 2012. Humeral epiphyseal shape in the Felidae: the influence of phylogeny, allometry, and locomotion. Journal of Morphology 273:1424-1438.

Werdelin, L., and M. E. Lewis. 2001. A revision of the genus Dinofelis (Mammalia, Felidae). Zoological Journal of the Linnean Society 132:147-258.

Wroe, S., C. McHenry, and J. Thomason. 2005. Bite club: comparative bite force in big biting mammals and the prediction of predatory behaviour in fossil taxa. Proceeding of the Royal Society of London B 272:619-625.