Department of Geological Sciences and Museum of Paleontology, University of Michigan, Ann Arbor, MI 48109, USA

Transcription

1 Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) Dietary reconstruction of Miocene Gomphotherium (Mammalia, Proboscidea) from the Great Plains region, USA, based on the carbon isotope composition of tusk and molar enamel David L. Fox*, Daniel C. Fisher 1 Department of Geological Sciences and Museum of Paleontology, University of Michigan, Ann Arbor, MI 48109, USA Received 22 October 2003; accepted 17 December 2003 Abstract The Miocene of the Great Plains of North America has long been recognized as an interval of major ecological reorganization. To reconstruct the dietary response of the proboscidean Gomphotherium to Miocene ecosystem change in the Great Plains, we analyzed the carbon isotope composition of 185 serial samples of tusk enamel from 17 individuals and bulk samples of posterior molars from 15 individuals of Gomphotherium from localities in the Great Plains ranging in age from the Early Barstovian land mammal age (ca. 15 Ma) to the Early Hemphillian land mammal age (ca. 8 Ma). Sets of samples from each tusk were designed to encompass about 1 year of tusk growth. Based on cheek tooth morphology, Gomphotherium is thought to be a browser with a diet primarily of dicots. The mean d 13 C of all samples is 9.8 F 1.2x, indicating that the diet of Gomphotherium was dominated by C 3 biomass. If Gomphotherium habitually foraged on water-stressed plants in arid habitats, which would have d 13 C values higher than the average composition for C 3 and C 4 plants, then all but one individual in this study consumed less than 50% C 4 biomass. The maximum percentage of C 4 in the diet would be much lower if food plants had average d 13 C values or if Gomphotherium foraged in closedcanopy habitats. Differences in d 13 C values between specimens from the southwestern US and the Great Plains, as well as between some coeval specimens from Nebraska, suggest geographic differences in either diet or typical foraging habitat. The data do not indicate a trend toward inclusion of more C 4 vegetation in the diet of Gomphotherium during the Miocene, and none of the serially sampled tusks exhibit seasonally varying d 13 C profiles. For most of its history in North America, Gomphotherium was a C 3 browser or a mixed feeder with a preference for browse. Our results indicate that areas of wooded habitat sufficient to support herds of large-bodied herbivores remained available in the Great Plains through the Late Miocene. D 2004 Elsevier B.V. All rights reserved. Keywords: Miocene; Gomphotherium; Paleodiet; Carbon isotopes; Enamel * Corresponding author. Current address: Department of Geology and Geophysics, University of Minnesota, Minneapolis, MN 55455, USA. Tel.: ; fax: addresses: dlfox@umn.edu (D.L. Fox), dcfisher@umich.edu (D.C. Fisher). 1 Tel.: ; fax: Introduction Reconstructing the diet of extinct species is a fundamental goal in vertebrate paleobiology. The nutritional quality, spatial and temporal distributions, /$ - see front matter D 2004 Elsevier B.V. All rights reserved. doi: /j.palaeo

2 312 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) and relative abundances of dietary constituents are major determinants of many aspects of the physiology, behavior, and ecology of animals (Damuth et al., 1992). The range of food resources available to a species is also an indication of the physical and ecological characteristics of the habitats to which a species is adapted. Changes in the diet of single lineages or entire communities through time can identify times of intensified or accelerated ecological and evolutionary change; when such a period is characterized by heightened extinction rates, reconstructions of diet can be used to infer or test mechanisms of extinction. For example, a number of hypotheses have proposed nutritional stress associated with ecosystem reorganization during the late Pleistocene deglaciation as a cause for megafaunal extinctions in North America (Graham and Lundelius, 1984; Guthrie, 1984; King and Saunders, 1984). Conversely, periods of relative stasis in dietary behavior through intervals independently identified as times of climatic or ecological change can indicate the tolerance of a species or community to such change. Traditional methods of reconstructing the diet of extinct mammalian herbivores rely heavily on extant analogues and include measures of molar occlusal morphology (Kay, 1975) and crown height (Webb, 1983; Janis, 1984; Janis et al., 2000), evidence from tooth enamel microwear (Walker et al., 1978), measures of cranial shape and masticatory functional anatomy that correlate with dietary preferences (Solounias et al., 1988; Janis, 1990), and phylogenetic affinity (Vrba, 1985). Recently, geochemical measurements of mineralized tissues of fossil mammalian herbivores have provided a new source of data on diet that is effectively independent of comparisons with extant analogues (van der Merwe, 1982; Sillen, 1986; Lee- Thorp and van der Merwe, 1987; Sillen and Lee- Thorp, 1994; MacFadden et al., 1999; Passey et al., 2002). In this paper, we use one such geochemical tracer, the stable carbon isotope ratio of enamel hydroxyapatite, to examine diet in the proboscidean Gomphotherium productus over the course of the Middle to Late Miocene in the Great Plains region, USA. Gomphotherium first appeared in North America during the Barstovian North American land mammal age (or NALMA; ca Ma; NALMA ages based on Tedford et al., 1987) and persisted through the Clarendonian NALMA (ca Ma) until the end of the Hemphillian NALMA (Lambert, 1996). Various lines of evidence indicate substantial ecological changes took place in the Great Plains during this interval, although details of both the timing and causes of these changes are still being investigated. The traditional view of the evolution of the Great Plains, based primarily on data from plant macrofossils (Axelrod, 1985; Leopold and Denton, 1987; Jacobs et al., 1999) and mammalian faunas (Webb, 1977, 1983; Janis et al., 2000, 2002), is that closed forest habitats that had dominated the mid-continent since the early Cenozoic gave way to savanna and more open-country faunas over the course of Middle and Late Miocene. This view is also supported by the temporal distribution of paleosols and soil structures characteristic of grasslands (Retallack, 1997), the evolution of locomotor adaptations for open habitats in several mammalian lineages during the Middle and Late Miocene (Camp and Smith, 1942; Webb, 1972), and previous carbon isotope studies of mammalian herbivores, which have identified a shift toward inclusion of more isotopically distinct tropical grasses in the diets of some lineages, particularly equids, beginning in the Hemphillian (Wang et al., 1994; Latorre et al., 1997; Passey et al., 2002). However, unlike these diverse lines of evidence for environmental changes in the Great Plains during the Miocene, the carbon isotope composition of Neogene paleosol carbonates in the region does not vary during the Miocene (Fox and Koch, 2003). Instead, this record indicates that tropical grasses constituted a substantial proportion (12 34%) of plant biomass that did not change much throughout the Miocene and only increased in abundance during the Plio- Pleistocene. Although the paleosol carbonate record precludes a long-term expansion of tropical grasses during the Miocene, because soil carbonates integrate soil biomass on timescales of years, the record is consistent with a complex mosaic of grasslands and forests prior to the major Plio-Pleistocene expansion of tropical grasslands. Additionally, grasslands dominated by grasses that are isotopically distinct from forest plants may have increased in area during the Miocene, as suggested by plant macrofossils and the ecological structure of mammalian faunas.

3 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) The proximate cause of these changes in terrestrial ecosystems in North America during the Late Miocene is generally thought to be global cooling and drying and an increase in the seasonality, or intraannual variability, of temperature and/or precipitation (Janis, 1989, 1993). This global explanation is consistent with carbon isotopic evidence for ecological changes among mammals on several continents during the Late Miocene, including North America (Cerling et al., 1997). Based on differences in paleosol carbonate records from various continents, Fox and Koch (2003) suggest that regional explanations focused on local climatic factors and ecological interactions within ecosystems are more likely than global climatic or atmospheric changes. Detailed studies of individual lineages, such as the genus Gomphotherium, can help resolve both the pattern and the causes of ecological changes. Treating North American Gomphotherium as a single species is not universally accepted, but here we follow the most recent species-level systematic treatment of North American gomphotheres (Tobien, 1972, 1973) and consider the genus to be monospecific. Despite the abundant evidence for environmental and faunal change during the Miocene, Gomphotherium was morphologically conservative throughout its history in North America (Lambert, 1996). The cheek teeth remained relatively simple and primitive with bunodont crowns, three transverse loph(id)s on the intermediate molars, and generally no more than four loph(id)s on M3. Accessory conules on the molars are absent or only weakly developed, and as a consequence, the complex trefoil wear figures characteristic of more derived North American gomphotheres such as Stegomastodon are absent or only weakly developed. The simple character of the molar crown morphology, in conjunction with their brachydont or low-crowned condition, suggests that Gomphotherium was primarily a browser. This interpretation is in good agreement with carbon isotope measurements of a population of Gomphotherium from at ca. 7.5 Ma (Fox and Fisher, 2001), and also with carbon isotope measurements of bulk samples of seven specimens from Arizona and New Mexico from localities that date to 6.9, 9.2, and 10.7 Ma (Latorre et al., 1997), all of which indicate a predominantly browsing to mixed feeding habit. However, two reasons suggest caution is warranted in assigning a diet to North American Gomphotherium for its entire history based only on molar morphology and these few isotope data. First, several recent studies have found surprising contrasts between herbivore molar morphology and tooth carbon isotope composition. For example, several specimens of the gomphothere Rhynchotherium from Bone Valley, Florida (4.5 Ma) have carbon isotope compositions of grazers, although the teeth are morphologically similar to Gomphotherium (MacFadden and Cerling, 1996). In contrast, two species of horse from Bone Valley with hypsodont cheek teeth typical of grazers (Astrohippus stocki and Dinohippus mexicanus) have carbon isotope compositions consistent with browsing (MacFadden et al., 1999). Similarly, specimens of the elephantid Mammuthus (mammoth) from Late Pleistocene sites in Florida have d 13 C values consistent with C 4 grazing, but co-occurring specimens of Equus with high-crowned teeth have d 13 C values consistent with C 3 consumption (Koch et al., 1998). A second reason for caution is that lineages can experience significant shifts in diet without a concomitant change in morphology, as observed from carbon isotope measurements of the teeth of the rhinoceros Teleoceras during the Late Miocene in Florida, which indicate a shift from browsing to mixed feeding (MacFadden, 1998). In this study, we use carbon isotope analyses of serial samples of tusk enamel and bulk samples of molar enamel to reconstruct the diet of Gomphotherium from the Great Plains through the Late Miocene. Proboscidean tusks are highly modified lateral incisors and are ideal for high-resolution stable isotope studies. The large size and simple accretionary growth geometry of proboscidean tusks (Fisher, 1987; Fox, 2000) facilitates recovery of multiple, temporally discrete samples with a resolution of samples per year of tusk growth. Unlike the tusks of most elephantids and mammutids, which lack enamel as adults, Gomphotherium tusks have a lateral enamel band along the length of the tusk that persists throughout ontogeny. Although our emphasis is on serial samples of tusk enamel, we also present additional measurements of bulk samples of molar enamel to increase the spatial and temporal coverage of the data set. Both tusks and molars are mostly composed of dentin, which allows much greater temporal resolution

4 314 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) of serial samples due to the orientation of incremental growth lines in both tissues (Fox, 2000). However, we focused our sampling on enamel, which has larger and more densely packed apatite crystallites than dentin, cementum, or bone. Early isotopic studies on fossils demonstrated that enamel can retain a biologically meaningful carbon isotope composition far longer than other tissues (Land et al., 1980; Schoeninger and DeNiro, 1982; Lee-Thorp and van der Merwe, 1987), increasing the likelihood of measuring biologically meaningful isotope ratios. These findings have been corroborated by modeling of isotope diagenesis in biogenic apatites (Wang and Cerling, 1994). However, no reliable, independent means of checking biogenic apatites for post-mortem carbon isotope alteration is known. The basic goal of this study is to determine how diet in Gomphotherium responded to the ecological and environmental changes in North America during the Late Miocene. By utilizing serial samples with sub-annual resolution, we can examine changes both in average or bulk diet and also in the seasonal variability of diet within individuals. The bulk samples from molars can indicate whether mean tusk values are consistent with values from molar enamel. The large body size of mammalian herbivores such as late Cenozoic proboscideans allows tolerance of diets with a wide range in nutritional quality, due to the allometric scaling of many aspects of digestive physiology and specializations such as hind-gut fermentation (Owen-Smith, 1988). This is certainly the case with the distant cousins of gomphotheres, the living Asian elephant Elephas maximus and the African elephant Loxodonta africana (Barnes, 1982; Owen- Smith, 1988; Ruggiero, 1992). Thus, if grasslands occurred locally as part of a spatially complex mosaic of habitats during the Miocene, Gomphotherium would have been able to subsist by increasing the amount of grass in the diet in rough proportion to its abundance on the local landscape. Such a dietary pattern should be recorded in the carbon isotope composition of enamel samples if the local grassland were dominated by isotopically distinct tropical grasses, as discussed below. Similarly, if increasing seasonality led to significant seasonal variation in the abundance of suitable browse, gomphotheres could subsist by seasonally switching between browse-dominated and grass-dominated diets, which should be reflected in the sub-annual patterns of carbon isotope composition of tusk enamel. However, if more nutritious browse plants remained abundant at sufficient levels throughout the Miocene without seasonal variation, presumably herds of Gomphotherium would focus their dietary strategy on these plants, rather than increase consumption of grasses that are often less nutritious and more difficult to digest (Demment and Van Soest, 1985; Ehleringer and Monson, 1993). 2. Carbon isotopes and dietary reconstruction Carbon has two naturally abundant stable isotopes, 12 C and 13 C, the latter of which is much less abundant, comprising only 1.1% of terrestrial carbon. Light stable isotope compositions are conventionally expressed in d notation, which is the permil (x, parts per thousand) difference between the ratio of rare to abundant isotope (R) in a sample and the same ratio in a standard material, normalized to the ratio in the standard. Thus, d 13 C=((R sample /R standard ) 1) The standard used with carbonate analyses, VPDB, is based on belemnite shells from the Cretaceous Pee Dee Formation in South Carolina. VPDB has a d 13 C of 0.0xby definition. The basis of using carbon isotopes to reconstruct the diet of extinct mammalian herbivores is the degree to which food plants that form the primary source of carbon for a herbivore fractionate carbon isotopes (discriminate against 13 C) during photosynthesis. Plants can be broadly divided into three categories based on differences in the photosynthetic pathways used for fixing atmospheric CO 2. The two most common photosynthetic pathways, the Calvin cycle and the Hatch Slack cycle, result in plant tissues with distinct and non-overlapping carbon isotope compositions (Smith and Epstein, 1971; O Leary, 1981; Farquhar et al., 1989). Plants using the Calvin cycle, or C 3 plants, include trees, shrubs, and cool-climate grasses and have mean d 13 C values of ca F 3x(all means reported F 1 S.D.) with a range of 21xto 35x(O Leary, 1988; Tieszen and Boutton, 1989). Plants using the Hatch Slack cycle, or C 4 plants, include sedges, some herbs, rare shrubs, and most temperate and tropical-climate grasses and have average d 13 C values of about 13 F 2x, with a range of 9xto 19x(O Leary, 1988; Tieszen

5 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) and Boutton, 1989). The third pathway, crassulacean acid metabolism or CAM, is uncommon and found mostly in succulents adapted to arid conditions. CAM plants are not thought to have ever been an important food source for mammalian herbivores and will not be considered further here. The variation in d 13 C in plant tissues, particularly for C 3 plants, is largely a function of environmental conditions such as light and moisture stress, nutrient availability, and temperature. Water-stressed plants in arid environments typically have d 13 C values that are shifted in a positive direction and closed-canopy forest plants generally have lower d 13 C values relative to open-habitat plants (Ehleringer and Monson, 1993). Plant carbon ingested by herbivores gets incorporated into mineralized tissues and passes the characteristic carbon isotope signature of the food on to carbonate that substitutes naturally into the structure of the mineral phase of the tissue. In mammals, mineralized tissues have a hydroxyapatite mineralogy, with an ideal formula of Ca 10 (PO 4 ) 6 (OH) 2, although several distinct apatite species are present; up to 5% (by weight) carbonate is found in the phosphate and hydroxyl sites in the enamel hydroxyapatite structure (LeGeros, 1991). Metabolism and biomineralization fractionate ingested carbon relative to the source plants so that enamel is enriched in 13 C relative to the composition of bulk diet. Although some controversy persists over the exact value of the enrichment factor between diet and enamel, here we follow the results of the most recent and detailed field study and assume an isotope enrichment for enamel of x (Cerling and Harris, 1999). Thus, enamel from a modern herbivore with a pure C 3 diet of average carbon isotope composition would have a d 13 C value of 12.9x, and enamel from a modern herbivore with a pure C 4 diet of average composition would have a d 13 C value of + 1.1x. Intermediate values would indicate a mixed diet, and the d 13 C value could be used to estimate the relative proportions of C 3 and C 4 plants, with some uncertainty in the exact C 3 /C 4 ratio corresponding to a given d 13 C value due to environment-specific variation in plant compositions. Dietary interpretation of d 13 C values from fossil herbivores must also account for the 1.5xdecrease in the d 13 C of atmospheric CO 2 due to fossil fuel burning by humans (Friedli et al., 1986). Generally speaking, C 4 plants are less nutritious and more difficult to digest than C 3 plants (Demment and Van Soest, 1985; Ehleringer and Monson, 1993), so changes in the relative proportion of these two types of plants within an ecosystem during the year would be an important component of the ecology of mammalian communities. Additionally, as C 4 grasses are characteristic of drier and warmer habitats (Teeri and Stowe, 1976; Hattersley, 1983), seasonal utilization of C 4 plants by Gomphotherium could imply seasonal variation in both temperature and moisture stress. Isotope analysis of the structural carbonate in apatite also yields the oxygen isotope composition of the carbonate (d 18 O c ). Although carbonate oxygen is theoretically more susceptible to isotopic alteration than either carbon or phosphate oxygen, the d 18 O c of pristine samples is a paleoclimate proxy that is related to both temperature and humidity (Longinelli, 1984; Bryant et al., 1996; Koch, 1998). In the context of this study, this allows comparison of the history of dietary behavior in Gomphotherium with the history of Late Miocene climate change, as recorded in the d 18 O c of Gomphotherium tusks. 3. Materials and methods 3.1. Specimens sampled A total of 185 serial samples of enamel from 17 tusks and bulk samples from 15 M2s or M3s were analyzed in this study. The specimens are in collections at the American Museum of Natural History (New York, NY), the University of Nebraska State Museum (Lincoln, ), and the Vertebrate Paleontology Laboratory, University of Texas at Austin. The specimens are from localities in New Mexico, Texas, Oklahoma, Kansas, and Nebraska (Figs. 1 and 2). The specimens from New Mexico are from sites in the Tesuque Formation, which is composed of sediments shed off of the Sangre de Cristo Range during the Barstovian (Tedford, 1981; Barghoorn, 1981). The remaining tusks come from localities in the Ogallala Group (equivalent to the undifferentiated Ogallala Formation in Texas), which is composed of fluvial, lacustrine, and eolian sediments shed from the Laramie, Front, and Sangre de Cristo Ranges and

6 316 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) molars. As far as we can tell from collection records and the specimens themselves, we have not sampled multiple elements from the same individual at any locality, but the taphonomy of these localities has not been examined in detail Sampling strategy For serial sampling, tusks were chosen that preserved a lateral enamel band of sufficient width and Fig. 1. Location map of localities from which tusks and molars were sampled. EB, Española Basin localities; CHP, Cole Highway Pit; VVP, V.V. Parker Pit; PEP, Pit; JSQ, Quarry; MF, Myers Farm; LM, Lake McConnaughy Shoreline locality; HR, Hottell Ranch; EQ, Ewert Quarry; MQ, Megabelodon Quarry; BQ, Burge Quarry; RLMQ, Rock Ledge Mastodon Quarry; DG, Devil s Gulch localities; GEP, George Elliott Place. the Hartville Uplift during the Middle Miocene to earliest Pliocene (Stanley, 1976; Skinner and Johnson, 1984; Gustavson and Winkler, 1988). Only stratigraphic information is available for the tusk Stratigraphic order and age assignments of the localities (Fig. 2) are based primarily on mammalian biostratigraphy and follow Tedford et al. (1987), Tedford (1981), Voorhies et al. (1987), Schultz (1990), Leite (1990), and Voorhies (1990a,b). Use of medial Barstovian, medial Clarendonian, and medial Hemphillian are based on recent biostratigraphic data from the Ogallala Group in Nebraska (Voorhies, 1990a,b). Tedford et al. (1987) divide these land mammal ages only into early and late intervals. Samples from several localities represent multiple tusks and/or molars, and at two localities samples represent one tusk and one molar. Most of these specimens are isolated teeth, whether tusks or Fig. 2. Stratigraphic position of localities from which tusks and molars were sampled, arranged by North American Land Mammal Age (NALMA) with absolute ages indicated in millions of years (Ma). Localities are not ordered within sub-divisions of land mammal ages, except for the Late Clarendonian. Numbers in parentheses indicate numbers of tusks and molars for localities with multiple specimens. Abbreviations same as in Fig. 1. CBM is Crookston Bridge Member of the Valentine Formation and corresponds to 1673, for which only stratigraphic provenance is known.

7 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) length for sampling and that appeared relatively pristine based on color and texture of the surface of the enamel. Sample series from each tusk were designed based on the geometry of tusk growth in Gomphotherium (Fox, 2000; Fox and Fisher, 2001). Tusks are ever growing, and new mineralized tissue (enamel, dentin, cementum) is added proximally within the tusk alveolus. Increase in the length of the tusk during growth is accommodated by continued eruption distally, so that the distal part of a tusk is chronologically oldest and the tissues are progressively younger toward the growing margin within the alveolus. The approximate orientation of the enamel appositional surface is inferred from incremental growth lines in the enamel, which (when adequately preserved) extend more or less directly across the width of the enamel band and dip proximally at a low angle (ca. 7j) in longitudinal section. As the growth lines are effectively discrete time lines, their orientation provides a framework for removing a time series of samples from the lateral enamel band. The surface of the enamel of several specimens ( 1393, 1673, 1949) had repeated, topographic depressions that were roughly perpendicular to the enamel band but had a subtly sigmoid course. These appear to be perikymata (see review in Dean, 1987), the topographic expression of outcropping enamel increments, and were used as guides in sampling those specimens. The outer surface of the enamel in most other tusks had no clear indication of growth lines, so individual samples were removed perpendicular to the long axis of the enamel band. Tusks were serially sampled in two modes. For most specimens, complete sample series consisted of 27 samples. However, to conserve on costs we initially analyzed only the first 10 samples from each series for d 13 C and d 18 O c. When these results showed no significant variation in d 13 C values, we decided against additional analyses. Based on an estimate of tusk eruption rate derived from measurements of growth increments in Gomphotherium tusk dentin (Fox, 2000), the complete sample series from tusks sampled in this mode were designed to cover approximately 3 years of tusk growth. Thus, the series of carbonate analyses from these tusks should each span approximately 1 year of growth. Samples were removed with a 1.0-cm diamond bit in a handheld flexible-shaft grinder. The samples were shallow grooves 1.5 mm wide in the growth direction and 1.5 cm across the enamel band, spanning the central 1/3 of the band width. Successive samples were separated by about 0.5 cm. The five tusks from (Figs. 1 and 2) were part of an earlier study of carbon and oxygen isotope variation in a population sample (Fox and Fisher, 2001) and were sampled in a slightly different manner. Sample series from these tusks were designed to encompass only 1 year of tusk growth, but included 18 samples. The samples, which were removed with a smaller diamond bit (0.4 mm), spanned the full width of the enamel band and were spaced on 0.25-cm centers. Several samples from Pit specimens were too small for isotope analysis of both structural carbonate and phosphate. Because Fox and Fisher (2001) gave priority to phosphate analyses, some sample series include fewer than 18 results. Bulk samples of molars were drilled either opportunistically on broken surfaces or along the labial or lingual surface of a loph(id) perpendicular to the occlusal plane, so as to cross growth lines in the enamel and increase the amount of time (hence, feeding behavior) represented in each sample. As we chose the locations of our bulk samples to minimize impact on the fossils, we did not necessarily sample all specimens in the same manner and we did not necessarily sample all of the growth history preserved in each molar. Bulk samples were typically about 5 mg of powdered enamel Sample preparation and isotope analysis All samples were pretreated to remove residual organic matter and sedimentary carbonates following a method suggested by Koch et al. (1997). Each sample was soaked in 2 3% NaOCl (0.04 ml/mg sample) for ca. 20 h to oxidize organic matter and then rinsed five times in an excess of distilled water. Next, samples were soaked in 1 M acetic acid buffered with calcium acetate (0.04 ml/mg sample) for another 20 h and again rinsed five times in an excess of distilled water and then thoroughly dried. Serial samples were reacted with 100% phosphoric acid at 72 jc in a Finnigan Kiel-I automatic carbonate extraction system, and the isotope composition of the resulting CO 2 was measured with a Finnigan MAT 251 isotope ratio mass spectrometer in

8 318 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) the University of Michigan Stable Isotope Laboratory. Bulk samples were reacted with 100% phosphoric acid at 90 jc in a Micromass Isocarb automatic carbonate extraction system, and the isotope composition of the resulting CO 2 was measured using a Micromass Optima isotope ratio mass spectrometer in the Stable Isotope Laboratory at the University of California, Santa Cruz. Raw isotope ratios were normalized either to international standards (NBS-18, NBS-19, NBS-20) or to a laboratory standard (Carrera marble). Analytical precision of both d 13 C and d 18 O c values is better than F 0.1x(1 S.D.). 4. Carbon isotope composition of Gomphotherium tusk and molar enamel The mean d 13 C value of all 200 serial and bulk samples is 9.8 F 1.2x(1 S.D.) and the values range from 13.3xto 4.8x(Fig. 3A and B). Interestingly, the mean d 13 C values for all of the serial samples of tusk enamel ( 9.9 F 1.1x, n = 185, Fig. 3A) and the bulk samples of molar enamel ( 8.3 F 1.2x, n = 15, Fig. 3B) are statistically distinct based on a two sample F-test for variances and a two-tailed t-test assuming equal variances (t = 5.44, df = 198, pb0.001). The result is the same if mean values are used for each tusk ( 9.9 F 1.2x; Fig. 3B) rather than all of the values (t = 3.94, df = 30, p < 0.001), which is more meaningful statistically as repeated observations from the same individual are not strictly independent. The basis for the difference between tusk and molar samples is not clear. The two sets of samples were analyzed in different labs (University of Michigan and University of California, Santa Cruz), but all sample processing protocols were otherwise the same, and both labs regularly analyze the carbonate standard NBS-19. One possible explanation is that the tusk enamel and molar enamel represent ontogenetic differences in diet, although this would require that all of the tusk samples came from a stage of ontogeny other than the age at which M2s and M3s mineralize. Another possible explanation is that enamel from tusks and molars may not be equally resistant to alteration as a result of differences in typical thickness (1 3 mm for tusk enamel, up to about 10 mm for molar enamel). The data in hand do not allow us to test this hypothesis, but similarities in Fig. 3. Histograms of d 13 C values from Gomphotherium. Labels indicate the lower, inclusive boundary of each bin. (A) All data from serial samples of tusk enamel (n = 185). (B) Mean d 13 C values for individual tusks (gray, n = 17) and d 13 C values of bulk samples of molar enamel (white, n = 15). the preservation and outward appearance of the specimens sampled do not suggest a consistent difference in preservation between tusks and molars. An intriguing alternative is that tusk and molar enamel carbonate have different isotope enrichment factors relative to

9 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) bulk diet, although the physiological basis for this is not apparent. Regardless of cause, the difference between tusk and molar d 13 C values is slight and does not substantially alter our interpretations of diet in Gomphotherium. A large part of the range in the whole data set (8.5x) is due to two individuals is a Late Barstovian tusk from Nebraska with unusually low d 13 C values in comparison to all other specimens in this study (mean: 13.0x, range: 13.3xto 12.7x). Esp is an upper M3 from the Ojo Caliente Member of the Tesuque Formation in the Española Basin, New Mexico and is Late Barstovian Early Clarendonian in age. It has a d 13 C value of 4.8x, which is 2xhigher than any other sample of Gomphotherium in this study. Excluding the data from these outliers increases the mean d 13 C value by only 0.2xto 9.6 F 0.9x, but decreases the range of the remaining data to 4.4x. We suspect this slightly reduced data set, which indicates that most Fig. 4. Variation in enamel carbon isotope composition in serial samples of Barstovian aged tusks. Specimens and lettering arranged with stratigraphically lowest specimen at bottom. In each plot, distances of samples along lateral enamel band (LEB) are plotted in negative centimeters so that time runs from oldest samples on left to youngest samples on right. Mean d 13 C values and ranges are indicated in each plot. (A) , Aqiquiu Rio del Oso sites, Tesuque formation, NM, Early Barstovian. (B) 1673, Valentine Formation, Ogallala Group,, medial Barstovian. (C) 1951, Hottell Ranch quarries, undifferentiated Ogallala Group,, medial Barstovian. (D) 99073, George Elliott Place, Valentine Formation, Ogallala Group,, Late Barstovian. (E) 1950, Myers Farm, Ogallala Group,, Late Barstovian.

10 320 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) specimens fall in a relatively narrow range of values, is a better representation of the typical carbon isotope composition of Great Plains Gomphotherium enamel during the Miocene. Below, the results from each interval are briefly described. All data are presented in Appendices A and B Early Barstovian is the oldest tusk sampled and is also from the locality that is furthest to the west, the Rio del Oso Abiquiu sites in the Española Basin, New Mexico. It has a mean d 13 C value of 7.8x, with a range of 8.3xto 7.5x(Fig. 4A). These are the most enriched values from any Gomphotherium prior to the Late Clarendonian Medial Barstovian Overall, the two tusks from the medial Barstovian have a mean d 13 C value of 9.3x. However, the ranges of values for the two specimens sampled from this interval do not quite overlap (Fig. 4B) has a mean of 9.8xand ranges from 10.1x to 9.5x (Fig. 4C) is slightly more enriched, with a mean d 13 Cof 8.5xand a range of 9.1xto 8.5x. However, the differences between the two specimens are not so great that the mean value for all 20 samples is misleading. The single bulk sample from the medial Barstovian, from the Tesuque Formation, Española Basin, New Mexico, is more enriched in 13 C than either tusk from this interval and has a d 13 C value of 7.6x. Fig. 5. Variation in enamel carbon isotope composition in serial samples of Clarendonian aged tusks. Specimens and lettering arranged with stratigraphically lowest specimen at bottom. In each plot, distances of samples along lateral enamel band (LEB) are plotted in negative centimeters so that time runs from oldest samples on left to youngest samples on right. Mean d 13 C values and ranges are indicated in each plot. (A) 1393, Megabelodon Quarry, Valentine Formation, Ogallala Group,, Early Clarendonian. (B) 25711, Rock Ledge Mastodon Quarry, Ash Hollow Formation, medial Clarendonian. (C) 8-31, Cole Highway Pit, Ogallala Formation, TX, Late Clarendonian. (D) 1949, Lake McConnaughy Shoreline localities, Ash Hollow Formation, Ogallala Group,, latest Clarendonian.

11 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) Late Barstovian The mean d 13 C value for Late Barstovian tusk serial samples is 11.2x. However, unlike the medial Barstovian tusks, the two tusks from Late Barstovian localities in Nebraska are quite distinct from each other isotopically. The samples from (Fig. 4D) have a mean carbon isotope composition of 9.5xand range from 10.0x to 8.9x (Fig. 4E) is the negative end-member in the entire data set, with a mean d 13 C of 13.0x and a range from 13.3x to 12.7x. Three bulk samples of molars from Late Barstovian localities in Nebraska were analyzed. One Fig. 6. Variation in enamel carbon isotope composition in serial samples of Early Hemphillian aged tusks. Specimens from different localities not arranged stratigraphically. In each plot, distances of samples along lateral enamel band (LEB) are plotted in negative centimeters so that time runs from oldest samples on left to youngest samples on right. All specimens from Ogallala Group in Nebraska or Ogallala Formation elsewhere. Mean d 13 C values and ranges are indicated in each plot. (A) 38257,. (B) 38258, Pit, OK. (C) 38269,. (D) 38270,. (E) 38259,. (F) 1843, V.V. Parker Pits, TX. (G) ,. (H) 86-9,.

12 322 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) of these, Ains from George Elliott Place, has a d 13 C value of 9.2x, which falls within the range of The other two, Ains and Ains 76, both from Devil s Gulch Quarry, have d 13 C values of 8.8xand 8.4x, respectively, and they are not substantially different from Early Clarendonian The single tusk sampled from the Early Clarendonian, 1393, has a mean d 13 Cof 10.4xand ranges from 10.8xto 10.0x(Fig. 5A). Five Early Clarendonian molars were sampled. Three molars from Burge Quarry are somewhat enriched in 13 C relative to the samples from 1393 ( 9.7x, 9.0x, and 8.4x). The single molar from Ewert Quarry, Nebraska is similarly enriched in 13 C and has a d 13 C value of 9.2x. The molar from New Mexico has the highest d 13 C value ( 4.8x) of any sample of Gomphotherium in this study Middle Clarendonian The single tusk from the Middle Clarendonian, 25711, has a mean d 13 Cof 8.8xand ranges from 9.2xto 8.4x(Fig. 5B). Both the carbonate and phosphate oxygen isotope composition of this specimen (discussed in Fox and Fisher, in preparation) indicate that the 10 samples from this tusk span 1 year of tusk growth (see below) Late Clarendonian The Late Clarendonian tusk specimen, 8-31 from Cole Highway Pit, Texas (Fig. 5C), has a mean d 13 Cof 9.6x. With d 13 C values that range from 10.7xto 8.2x, 8-31 is one of the specimens with the greatest variability. However, this variability is mostly due to one extreme value ( 8.2x). Excluding this sample reduces the range to 1.5x, which is closer to the mean individual range for all tusks (1.0x) Latest Clarendonian 1949 (Fig. 5D) has a mean d 13 C of 9.5xand ranges from 10.0xto 9.0x Early Hemphillian The largest sample from a single time period is the Early Hemphillian sample, which includes eight tusks (Fig. 6) and six molars. Data from five of the tusks ( Pit tusks, Fig. 6A E) have been presented and discussed in detail elsewhere (Fox and Fisher, 2001). The serial samples of Early Hemphillian tusks have a mean d 13 Cof 10.1xand range from 11.3xto 7.8x. Overall, the Early Hemphillian tusk values are the most variable. However, as with the single Late Clarendonian tusk, the higher variability is due to a single, relatively extreme value ( 7.8x, 1843, Fig. 6F). If this sample is excluded, the range of values is reduced to 2.4x. Nevertheless, the only interval with a higher range is the Late Barstovian. As with other intervals represented by both tusk and molar data, the molars are more enriched in 13 C than the tusks. The mean for the six molars ( 8.3 F 0.8x) is almost 2xhigher than the mean for tusks. 5. Discussion 5.1. Dietary reconstruction of Gomphotherium Taken as a group, the carbon isotope measurements of Gomphotherium tusks and molars are variable, but not extremely so (Figs. 3 6). The d 13 C values indicate that Gomphotherium generally had a diet dominated by C 3 vegetation. This conclusion is consistent with the bunodont and brachydont morphology of the cheek teeth, which suggests that Gomphotherium was a browser rather than a grazer. Taken at face value, the carbon isotope variability in Gomphotherium could imply that a single feeding strategy such as strict browser does not adequately describe this genus across its geographic and stratigraphic range, which would not necessarily be surprising given the dietary flexibility available to a mammal at large body size (Owen-Smith, 1988). However, assigning specific feeding strategies and converting the d 13 C values into quantitative estimates of proportions of C 3 and C 4 biomass in the diet are complicated by both uncertainty about the d 13 C value of atmospheric CO 2 through the Miocene and the environmentally controlled variability in the d 13 C of the diet.

13 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) Passey et al. (2002) calculated atmospheric CO 2 d 13 C values for the Miocene based on time series of d 13 C values from planktonic foraminifera and used the calculated values to predict past plant tissue and herbivore enamel d 13 C values. Their predictions for enamel d 13 C values of herbivores consuming only C 3 vegetation with average carbon isotope composition range from about 10.5xto about 11.8xover the Miocene, with most predictions being slightly less than 11.0x. As a simplification, we assume that atmospheric CO 2 during the Miocene had a d 13 C value of 6.5x(Friedli et al., 1986). This implies average C 3 and C 4 carbon isotope compositions of 25.5xand 11.5x, respectively; hence, an enamel d 13 C value of 11.4xfor a herbivore consuming C 3 biomass with average carbon isotope composition, which is in good agreement with the approach of Passey et al. (2002). To control for environmental effects on plant compositions, we use the standard deviation of mean C 3 and C 4 d 13 C values (O Leary, 1988; Tieszen and Boutton, 1989) and assume that water stress in arid environments would increase the average d 13 C values of C 3 and C 4 by 3xand 2x, respectively, and that the canopy effect due to recycled CO 2 in closedcanopy habitats would decrease d 13 C values to a similar degree. To calculate percentages of C 3 and C 4 plants in diet, we assume that Gomphotherium d 13 C values result from simple linear mixing of dietary inputs with the specified C 3 and C 4 d 13 C values. Dietary reconstructions for tusks are based on the mean d 13 C values. Considering the implications of average plant compositions for dietary reconstruction (Fig. 7A), the d 13 C values imply that the typical Gomphotherium diet included 84% C 3 biomass, and the average diet of individuals ranged from about 53% C 3 to exclusively C 3. The lowest percentage of C 3 consumption corresponds to the high d 13 C value of Esp ; if this extreme value is excluded, the mean does not change, but the minimum percentage of C 3 increases to 67%. If all of the individuals in this study foraged only in arid habitats dominated by water-stressed vegetation (Fig. 7B), plant compositions would be shifted positively and the enamel d 13 C values would imply almost exclusive reliance on C 3 biomass by Gomphotherium. Excluding Esp would imply that Gomphotherium diets included a minimum of 88% C 3 plants. Assuming Gomphotherium foraged in closed-canopy habitats (Fig. 7C) permits the greatest amount of C 4 biomass in the diet. The highest percentage C 3 in this case corresponds to 1950, the specimen with unusually low d 13 C values, and the lowest value is, again, for Esp Excluding both of the outliers (Esp , 1950) from the closed-canopy interpretation does not change the mean diet (65% C 3 biomass) but decreases the range to 50 76% C 3 in the diet. The shapes of the distributions for average and closed-canopy environments are different because of the difference in the responses of C 3 and C 4 plants to environmental conditions. The interpretations for average plant compositions and closedcanopy habitat would indicate that the feeding strategy of different populations of Gomphotherium ranged from strict browsing (>90% dicot plant parts, particularly leaves; Janis et al., 2000) to mixed feeding with a preference for browse; the interpretation for arid foraging habitats implies that Gomphotherium was a strict browser (with the exception of Esp from New Mexico) Spatial variation in Gomphotherium d 13 C values and diet Two geographic patterns are notable among the results. First, the one tusk and two molars from New Mexico in this study have higher d 13 C values than specimens from the Great Plains, regardless of age , from the Early Barstovian of New Mexico, has the most enriched d 13 C values of any tusk measured in this study (Figs. 4A and 8). Based on a two sample F-test of variances and a one-tailed t- test assuming unequal variances, the mean value for the serial samples from is greater than the mean of all of the mean d 13 C values of serial samples from Great Plains specimens to a statistically significant degree ( 7.8xvs. 10.0x, respectively; t = 19.1, df = 25, pb0.001), implying that had a diet that was isotopically distinct from Gomphotherium in the Great Plains. The two bulk samples of molars from the Española Basin reported here also indicate higher d 13 C values for Gomphotherium in New Mexico in comparison with specimens from the Great Plains. Esp has

14 324 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) the highest d 13 C value of any specimen in this study ( 4.8x) and has the third highest d 13 C value, after a molar from that is unusual in comparison to values from other Gomphotherium specimens at that site (this study; Fox and Fisher, 2001). The sample size of molars from New Mexico presented in this study (two) is too small for statistical analysis, but including seven published d 13 C values for Gomphotherium from Arizona and New Mexico (Latorre et al., 1997) does allow for statistical comparisons, albeit still with relatively small sample sizes. An F-test and a onetailed t-test assuming unequal variance indicates that the mean d 13 C value of bulk samples of molars from the southwestern US cannot be distinguished statistically from the mean value of teeth from the Great Plains ( 7.9x vs. 8.6x, respectively; t = 1.13, df = 10, p = 0.14). Pooling the mean d 13 C value from each tusk and the values from bulk samples of molars from each region, however, indicates that Gomphotherium from the southwestern US does, in fact, have significantly higher d 13 C values on average ( 7.9xvs. 9.4x, respectively; F- test and one-tailed t-test assuming equal variance, t = 3.20, df = 37, p = 0.001). This final analysis is probably the most robust given the larger sample size achieved by including both mean tusk values and bulk sample d 13 C values. The apparent contrast between d 13 C values in Gomphotherium from the southwestern US and the Great Plains suggests one of two interpretations. First, southwestern populations may have relied Fig. 7. Histograms of the estimated proportions of C 3 biomass in the diet of Gomphotherium during the Miocene. Percentages assume enamel d 13 C values result from linear mixing of C 3 and C 4 biomass in the diet, an apparent fractionation between diet and enamel of x, and a d 13 C value for atmospheric CO 2 of 6.5x. Labels indicate the lower, inclusive boundary of each bin. Grey bars indicate reconstructions for tusk means and molar bulk samples from the Great Plains (n = 29); white bars indicate reconstructions for tusk means and molar bulk samples from Arizona and New Mexico (n = 10), including data from Latorre et al. (1997). Mean dietary percentages of C 3 ( F one standard deviation) are indicated for Great Plains (GP) and southwestern USA (SW) for the assumptions in each panel. (A) Diet with average plant compositions (C 3 = 25.5x, C 4 = 11.5x). (B) Diet from arid, water-stressed environment (C 3 = 22.5x, C 4 = 9.5x). (C) Diet from forested, closedcanopy environment (C 3 = 28.5x, C 4 = 13.5x).

15 D.L. Fox, D.C. Fisher / Palaeogeography, Palaeoclimatology, Palaeoecology 206 (2004) somewhat more heavily on C 4 plants than populations in the Great Plains. Regardless of the foraging habitat, the relatively high d 13 C value for Esp indicates that at least this individual consumed considerable amounts of C 4 biomass (28 64%). If the relative abundance of C 3 and C 4 biomass in the two regions were the same, then regional variations in dietary preferences are indicated. Alternatively, gomphotheres may have fed indiscriminately so that individuals consumed plants in proportion to their abundance on the local landscape, which would imply a greater abundance of C 4 vegetation in the southwestern USA. A test of this explanation would be a comparison of d 13 C values of coeval paleosol carbonates from the two regions, which could indicate regional differences in the proportions of C 3 and C 4 plants growing in ancient soils. The carbon isotope composition of paleosol carbonates in the Great Plains indicates the persistent presence of some C 4 biomass but predominantly (66 88%) C 3 biomass throughout the entire Miocene in that region (Fox and Koch, 2003). That reconstruction of Great Plains biomes is in good agreement with the dietary reconstructions for Great Plains gomphotheres regardless of environmental assumptions (Fig. 7), suggesting that gomphothere diets, at least in the Great Plains, do reflect the local mix of C 3 and C 4 biomass. We do not know of any published d 13 C values for Miocene paleosol carbonates from the southwestern USA. A second interpretation is that southwestern populations of Gomphotherium fed in more open habitats on water-stressed C 3 plants, which would have had higher d 13 C values. This interpretation would imply regional differences in overall habitat structure. A possible test of this explanation would be a comparison of locomotor adaptations in faunas of the two regions under the assumption that more open habitats would have a greater proportion of cursorial species. A second geographic pattern in the results presented here is apparent in the Late Barstovian sample (a tusk) is from the Myers Farm locality near the Republican River in southernmost central Nebraska; (also a tusk) and Ains (a molar) are from George Elliott Place near the Niobrara River, roughly 300 km north of the Myers Farm locality. The most enriched value in the sample series from 1950 ( 12.7x) is 2.7xlower than the lowest value in the series from ( 10x). The mean values for the two tusk specimens ( 13.0xand 9.5x, respectively) are even more disparate. The bulk sample from molar Ains is within the range of the sample series from Within the temporal resolution of this study, these data imply a considerable degree of variation in the feeding behavior of Gomphotherium on a relatively small spatial scale. The basic interpretations are similar to those for the contrast between southwestern and Great Plains samples: inter-populational differences in feeding behavior despite overall similarities in habitats, indiscriminate feeding and small-scale variation in the abundance of C 3 and C 4 plant biomass, or inter-populational differences in habitat utilization while foraging. Data from more Late Barstovian specimens from Nebraska are necessary to distinguish between these interpretations Temporal variation in Gomphotherium d 13 C values As the tusk means and molar bulk samples have statistically different mean d 13 C values, their patterns of change over time must be treated separately. Over the course of the Middle to Late Miocene, the mean tusk values exhibit a slight trend toward lower d 13 C values (Fig. 8), which would be consistent either with increasing utilization of C 3 plants by Gomphotherium over time or a shift in foraging habitats to more closed-canopy environments. However, time explains only a small portion of the variation in the tusk data (r 2 = 0.09), the standard error of the regression is relatively high (1.08), and the regression is not statistically significant ( F = 1.43, p = 0.25). If the outlier value for 1950 is not included in the analysis, the variation explained increases (r 2 = 0.44), the standard error is lower (0.606), and the regression is statistically significant ( F = 10.79, p = 0.005). The trend in d 13 C, such as it is, appears unrelated to paleoclimate as recorded by d 18 O c : a least squares linear regression for the tusk means with d 18 O c as the independent variable and d 13 C as the dependent variable (Fig. 9) is not statistically significant ( F = 1.18, p = 0.03) and the proportion of the variation in the d 13 C values explained by the d 18 O values is low (r 2 = 0.09). The molar d 13 C values (including the seven