Extinctions in near time: new radiocarbon dates point to a very recent disappearance of the South American fox Dusicyon avus (Carnivora: Canidae)

Almost all large carnivorans (Carnivora; > 20 kg) that inhabited South America became extinct around the Late Pleistocene – Early Holocene transition. Two exceptions were species of coyote-sized Dusicyon , one insular ( D. australis ) and one continental ( D. avus ). The extinction of the former is a resolved matter, but that of D. avus , found in the Patagonian and Pampean regions, is still poorly understood. Using the Gaussian-Resampled Inverse-Weighted McInerny method we present new radiocarbon evidence indicating that its disappearance occurred in very recent times (about 324 – 496 years cal BP ). We found no evidence to support a role for hybridization with domestic dogs in causing the extirpation of this fox. Climatic change may have reduced its distributional range, as has happened with other mammals, although not to the extent of explaining its extinction. Climatic change, however, coupled with increased anthropogenic impacts such as hunting, domestic dogs, and/or other aspects relating to the impact of European colonization in South America’s southern cone, were the probable main drivers of the recent extinction of D. avus . © 2015 The Linnean Society of London, Biological Journal of the Linnean Society , 2015, 116 , – 720.

INTRODUCTION are important in that they may herald the onset of a sixth mass extinction (Leakey & Lewin, 1992;Mac-Phee & Flemming, 1999;Ceballos & Ehrlich, 2002;Barnosky et al., 2011;Teta et al., 2014). These extinctions, which have impacted a wide array of organisms, principally affected small-sized, island-dwelling species (body mass < 10 kg) in the case of mammals, and were apparently not related to climatic change (MacPhee & Flemming, 1999). Continental South American mammals showed few extinctions after the Early Holocene (MacPhee & Flemming, 1999;Turvey, 2009;Turvey & Fritz, 2011). Since the arrival of Europeans, some ten speciesmostly small rodents and marsupials-are thought to have become extinct  although taxonomic uncertainties make the true number uncertain.
Here we describe new specimens of D. avus, dated between 700 and 400 cal years BP, from localities in the Pampean and Patagonian regions of Argentina and southern Chile. Using the Gaussian-Resampled Inverse-Weighted McInerny (GRIWM) method (Bradshaw et al., 2012), we estimated the time of extinction of this fox, and determined its occurrence in historical times. Based on this emerging evidence we revisited the potential factors that may have accounted for the extinction of D. avus.

TAXONOMIC IDENTIFICATION
Taxonomic assignation of the specimens was based upon a combination of qualitative and quantitative approaches (Kraglievich, 1930;Berman & Tonni, 1987;Trejo & Jackson, 1998;Amorosi & Prevosti, 2008;Prevosti et al., 2009Prevosti et al., , 2011. A list of specimens and definition of measurements utilized are provided in Supporting Information (Appendix S2). We performed a Principal Component Analysis (PCA) from the variance-covariance matrix of the ln transformed measurements following Prevosti et al. (2009Prevosti et al. ( , 2011 to evaluate the identification of the new specimens. A biplot between the length of first and second lower molar follows Prevosti et al. (2009Prevosti et al. ( , 2011 (Stuiver & Reimer, 1993;Reimer et al., 2009), using the SHCal13 curve (Hogg et al., 2013) and two sigma ranges. To estimate the age of extinction of D. avus we used the GRIWM method, with an R 3.0.1 script (R Core Team, 2013) obtained from C. Bradshaw and F. Saltre, which weighs observations inversely according to the temporal distance from the last confirmed observation of a species (Bradshaw et al., 2012). As some of the dates had large errors, and there were few dates between 4000 and 1000 years cal BP (Supporting Information, Appendix S3) that could bias the GRIWM results (F. Saltre, pers. comm.), we first run a model analysis reducing the error of four dates with values larger than 150 to 90 cal years. To address the second source of bias we used the younger eight dates, all of which are <4000 cal BP (Supporting Information, Appendix S4).
Stable isotopes (d 13 C, d 15 N) data were obtained from the dated specimens and compared with published data for D. avus to determine the potential existence of dietary variability (Supporting Information, Appendix S8; for the interpretation of dietary signals with these isotopes, see Bocherens & Drucker, 2003;Fox-Dobbs et al., 2007;Yeakel et al., 2013).

POTENTIAL DISTRIBUTION MAPS
To explore the role that climate change may have had in affecting the distribution of D. avus, we performed potential distribution models using MaxEnt v.3.3.e (Elith et al., 2006;, using specimens of different ages and two different climatic variable datasets. This software was chosen because it performs better than alternatives with presence-only data (Elith et al., 2006). We used the WorldClim database for present climatic conditions (Hijmans et al., 2005), with a spatial resolution of~1 km², and the CCSM4 database (Gent et al., 2011) for climatic information for the middle Holocene (i.e. 6000 years), with a spatial resolution of 2.5 km². Both environmental datasets contain monthly precipitation and temperature information, and 19 bioclimatic variables. We generated two potential distribution models: one including only middle Holocene localities and the 6000 years BP database (6 ka), and the other including only late Holocene localities and the 'actual' database (0 ka). Because there is an appreciable interval between the late Holocene fossils themselves and the modern environmental conditions in the WordClim database, there was a risk that the distribution of D. avus in the 0 k model would be biased by greater presentday humidity. To explore this alternative we ran an analysis using the middle Holocene database and the late Holocene localities of D. avus (6 kaB).Ten replicates were performed for each model, with 25% of localities used as test data, random seed and 10 000 background points. We used the cumulative output and assigned predictive values of 100-51, 50-26, 25-11, 10-2 and 1-0 (Fig. 6). Variable contributions were analyzed with jackknife tests and model predictions with a threshold independent measure, the area under the receiver operator curve (AUC) . The AUC is interpreted as the probability that a randomly chosen present location is ranked higher that a randomly chosen background point (Merow, Smith & Silander, 2013). It varies between 0 and 1, with an AUC of 0.5 meaning that the model does not perform better than would a random model (Hern andez et al., 2006).
To evaluate the effects of climatic change across time, we extracted climatic information (i.e. mean annual temperature, and annual precipitation) for each recorded in D. avus locality from three climatic databases, the two mentioned above as well as a third for the Last Glacial Maximum database (Collins et al., 2006

TAXONOMIC IDENTIFICATION
MCA 81-VI-1-1 is composed of an incomplete rostrum with part of the maxillary and premaxillary bones and left and right I3-M2; incomplete left and right mandibles with right c1-p1, broken p2 and p3-m1, and left c1-p2 and p4-m2 (m1 is broken; Figs 2-4, Supporting Information, Appendices S5, S6). The left p3 alveolus was reabsorbed, indicating that this tooth was lost while the animal was alive; the permanent dentition is in its final position and is moderately worn, suggesting an adult individual. CEHA 5131 is a nearly complete skull (right zygomatic arch and pterygoid bones are broken) with broken right P4, nearly complete left P4 and complete left M2 (Figs 2-4, Supporting Information, Appendices S5, S6). MC 787 and MPEF-PV 10887 are right and left mandibles, respectively. The dentition is nearly complete, but the i1-3, p1 and m3 were not preserved in the MC 787, while the m3 was missing in MPEF-PV 10887 (Figs 2-4, Supporting Information, Appendices S5, S6).
The new specimens are clearly separable from domestic dogs due to the presence of several cranial and dental traits that separate Dusicyon and Canis (see Pocock, 1913;Berta, 1989;Tedford, Taylor & Wang, 1995;Prevosti, 2010). These include: (1) the occipital forming a more rounded inion that is not so expanded posteriorly, while in Canis the inion is very pointed and posteriorly projecting; and (2) the jugal bone has a scar for the superficial masseter m. that is wider in Dusicyon than Canis, which instead exhibits a steeper forehead front with a more developed frontal sinus (Figs 2-4, Supporting Information, Appendices S5-S7). The dentition of Canis familiaris is more robust, with more bunodont teeth (e.g. molars, premolars), lower and more robust principal cusps in the premolars (especially the p4), shorter and more robust canines, and I3 with a strong mesiolingual cingulum. The M1 has a more reduced labial cingulum and a proportionally larger and more bunodont paracone, while the m1 has a stouter trigonid and a more reduced metaconid and entoconid (Figs 2-4, Supporting Information, Appendices S5-S7; Prates, Prevosti & Ber on, 2010;Prevosti, 2010).
Using size and evidence of a more carnivorous dentition, we were able to separate the new specimens from other South American living foxes (e.g. Cerdocyon thous, Lycalopex gymnocercus). We then differentially compared them with L. culpaeus, Dusicyon australis and D. avus. The new specimens present a set of characters that are diagnostic of D. avus. These include the presence of a well developed hypoconulid in the m1, a second accessory cusp and a narrowed distal cingulum in the p4 (in L. culpaeus is low and wide), and a protocone placed lingually in the P4 (in L. culpaeus is placed mesially or mesiolingually; Prevosti et al., 2011;Figs 2-4, Supporting Information, Appendices S5-S7). The new specimens have proportionally wider postorbital constrictions and processes, larger m1 in relation to m2 and a wider bulla, which are characters that are shared with D. australis (Prevosti et al., 2011;Figs 2-4, Supporting Information, Appendices S5-S7). D. australis has a more reduced protocone in the P4 and smaller metaconid in the m1 than D. avus. The principal cusps of the premolars are taller and more acute in D. australis (specially the p4), the distal cingulum of the p4 is narrower and more acute in D. australis. In some specimens the cingulum was raised and cusp-like, but a true second accessory distal cusp is not present (Prevosti et al., 2011;Figs 2-4, Supporting Information, Appendices S5-S7). The secondary palate is posteriorly extended just at the level of the distal border of the M2, while in D. avus it ends more anteriorly (at the level of the distal half of the M2; Figs 2-4, Supporting Information, Appendices S5-S7; Berta, 1989). The sagittal crest is more developed in several specimens of D. avus, while the studied D. australis presents a lyriform area delimited by the temporal crests (Figs 2-4, Supporting Information, Appendices S5-S7; Berta, 1989).
Morphometric analyses (i.e. Principal Component Analysis and biplot graph) were compatible with this comparison and supported the separation of D. avus from other similar sized South American wild canids (see Supporting Information, Appendix S7).
GRIWM analysis using all samples indicated that D. avus would have become extinct during the 20 th century (c. 1950), with a median date of 0 cal years BP. Such an analytical result theoretically implies that it might still be extant, since the upper confidence limit (2.5%) is 441 cal years after 1950. The lower confidence limit (CI; 97.5%) is 463 cal years BP. Additional GRIWM runs gave similar results (see Fig. 5; Supporting Information, Appendix S4).
Stable isotope ratios collected from R ıo Luj an 1 d 13 C: À12.8, d 15 N: 9.6, and Dinamarquero specimens exhibit values of d 13 C: À20.2 and d 15 N: 9.6. The d 13 C values for MC 787 and MPEF-PV 10,887 were À18.9 and À19.9, respectively. These values were within the previously published data range (see Supporting Information, Appendix S8), with the exception of the R ıo Luj an specimen which was more positive (d 13 C value, À16.8; Supporting Information, Appendix S8).

DISTRIBUTION AND POTENTIAL DISTRIBUTION MODELS
A potential distribution model 6 ka was performed with eight middle Holocene localities (Fig. 6, Supporting Information, Appendix S9). High predictive values were attained for southern and central Patagonia and western Argentina at 29°S, and mean predictive values along northern Patagonia and the Argentine Pampas (Fig. 6). Model 0 ka was performed with 13 late Holocene localities (Supporting Information, Appendix S9), and returned a southern displacement of high predictive values for Buenos Aires province, while southern Patagonia remained with high predictive values (Fig. 6). Model 6 kaB showed a pattern more similar to model 6 ka than to model 0 ka. Jackknife tests suggested that for the 6 ka model the most important environmental variables were temperature seasonality, isothermality and January maximum average temperature, whereas altitude and August minimum temperature were more relevant for models 0 ka and 6 kaB (Supporting Information, Appendix S9). All models presented better predictions that those randomly generated, with high AUC values: 6 ka, 0.997 AE 0.002; 0 ka, 0.957 AE 0.049; 6 kaB, 0.997 AE 0.002.
Considering all the Holocene localities studied, mean annual temperature during the Late Glacial Maximum (LGM) was 5.7°C (À3.3 to 12.8°C) and annual precipitation was 446 mm (152-1159 mm). Mean annual temperature of record localities in 6 ka was 9.8°C (4.4-16.4°C) and annual precipitation was 483 mm (186-977 mm; Supporting Information, Appendix S10). Mean annual temperature of record localities in the actual database was 10.5°C (5.3-16.8°C) and annual precipitation was 517 mm (197-1085 mm). The upper precipitation limit was smaller when only the localities used in the model 6 ka and model 0 ka were included: 765 mm for the middle Holocene, and 995 mm for the actual climatic database (Supporting Information, Appendix S10). Due to some uncertainty we excluded the R ıo Luj an specimen, that could have been collected elsewhere and brought in by humans (see below), and the northern Pampas specimens (Laguna El Doce and Estaci on J. M. Garc ıa), that lacked chronological data (they had a 'Holocene' age and the last one could be from the Early Holocene, see Prevosti & Pardiñas, 2001). Using this reduced sample we found a decrease of~1°C in the maximum of the mean annual temperature and >100 mm in the maximum annual precipitation.

AGE OF EXTINCTION
Based on an archaeological context a late Holocene extinction of Dusicyon avus was proposed in the 1970s and 1980s, (Caviglia, 1978(Caviglia, , 1986Berman & Tonni, 1987;Berman, 1994;Mansur, 2006Mansur, , 2007, but was only recently confirmed by direct taxon dates using 14 C. Absolute chronology places D. avus last occurrence at c. 3000 years 14 C BP (Prevosti et al., 2011). Martin (2013) BP) from material recovered in Negro Muerto 2 in northern Patagonia, Argentina (Fig. 1). Our new data extend D. avus biochron to <1000 years 14 C BP, with the last record in the Pampean region at c. 700 14 C years BP (AD 1232-1397 years), and the most recent record in southernmost Patagonia with an age of c. 400 years 14 C BP (AD 1454-1626 years). These dates would suggest that D. avus extinction could have occurred after the arrival of Europeans to South America ( Fig. 5; Supporting Information, Appendix S3). The extinction estimate using GRIWM overlaps with the younger 14 C, suggesting that the extinction could have happened since 463 cal years BP (AD 1482 years). Indeed, the species could even be living today, since the estimated age of extinction lays in the future, with a median value that falls in the middle of the 20 th century ( Fig. 5 Balmaceda, 1976) and might have been one of the last observations of D. avus. An unusual large fox was also mentioned for Tierra del Fuego (Lothrop, 1928). Together with 'Aguaras' reported in the 19 th century in northern Patagonia and the Pampean region (and uncritically assigned to Chrysocyon brachyurus; Prevosti et al., 2004), these could have been in fact D. avus specimens. However attractive this interpretation is, and bearing the support of the GRIWM results, the available information is not enough to corroborate this hypothesis (Prevosti et al., , 2011.
The climatic parameters obtained from the climatic databases and the distribution models are congruent with the interpretation that D. avus inhabited open areas (e.g. grass steppe, shrub steppes) under a wide range of climatic conditions (Fig. 5). The models also show that most recorded localities experienced arid to semiarid conditions (i.e. annual precipitation < 800 mm), with variable temperature and precipitation, and high seasonality and isothermality (Supporting Information, Appendix S9). This would support the hypothesis that at least in Buenos Aires province the distribution of D. avus was related to more arid conditions, as has been suggested for the middle and late Holocene (e.g. Tonni et al., 1999;Mancini et al., 2005). Several mammal species presently associated with dry environments disappeared from their Pampean ranges during the late Holocene, including ungulates (e.g. Lama guanicoe; Tonni & Politis, 1980;Politis & Pedrotta, 2006;Politis et al., 2011), mustelids (Lyncodon patagonicus; Prevosti & Pardiñas, 2001;Schiaffini et al., 2013), and marsupials (Lestodelphys halli; Prado, Goin & Tonni, 1985;Goin, 1995Goin, , 2001. A plausible explanation would be that these range retractions were triggered by a regional change to more humid conditions (Tonni et al., 1999;Stutz, Prieto & Isla, 2006;Tonello & Prieto, 2010;Del Puerto et al., 2011). We verified an additional regional extirpation of D. avus during the late Holocene for the northeastern portion of Chubut province, probably connected with the southward advance of the Monte shrubland during warmer and drier conditions (Medieval Warm Period and present). A similar situation involving small marsupials and rodents (e.g. Abrothrix olivacea, Lestodelphys halli) has been documented for eastern Patagonia (Pardiñas et al., 2012;Udrizar Sauthier & Pardiñas, 2014). However, although these climatic changes could be linked with local retractions, they were arguably not enough to trigger the extinction of a widespread fox species.
The presence of D. avus in a burial site in R ıo Luj an 1 around 700 14 C years BP contravenes the climate hypothesis, as warmer and wetter conditions related to the Medieval Warm Period were prevalent in the Pampean region at this time. This specimen might have been transported for ritual purposes from elsewhere, or might have been a tame animal (Prates, 2014). A similar explanation has been invoked for a late Holocene Lama guanicoe record in northeast Buenos Aires province (Politis & Pedrotta, 2006;Politis et al., 2011).
The second hypothesis advanced to explain the extinction of D. avus by Berman & Tonni (1987) through hybridization with domestic dogswas also discarded, as there is no evidence of hybridization in the morphology of the skull and dentition of D. avus nor in its mitochondrial DNA (Prevosti et al., 2011;Austin et al., 2013). We also failed to find any evidence of hybridization in the morphology of the new specimens, but this hypothesis remains unlikely, as hybridization with domestic dogs seems to be restricted only to the genus Canis (e.g. Canis lupus, C. simensis; Gottelli et al., 1994;Vil a & Wayne, 1999;Sillero-Zubiri, Hoffmann & Macdonald, 2004), and there are no records of such hybridization in living South American canids. More ancient DNA sequencing (including nuclear DNA) will be needed to further test this hypothesis.
Dusicyon avus were used by humans during the Holocene for ritual purposes, as suggested by its inclusion in burials, and its teeth were probably utilized in necklaces (Messineo & Politis, 2007;Prates et al., 2010;Prevosti et al., 2011;Laporte, 2014;Politis, Barrientos & Scabuzzo, 2014;Prates, 2014). Based on the frequency and the skeletal elements (mostly teeth) found in archaeological sites, Bonomo (2006) surmised that the species had a high symbolic value for aboriginal peoples. Prates (2014) inferred that a specimen from Loma de los Muertos was intentionally buried, and may have been tamed, a practice recorded for other fox species in historical times (Stahl, 2013). The increase of human populations during the late Holocene (since 3000 cal years BP; Borrero, 2008b;Mart ınez & Guti errez, 2008;Salemme & Miotti, 2008;Morales et al., 2009) may have intensified the use of this species with a consequent impact on its populations (cf. Prevosti et al., 2011). The new radiocarbon dates indicate that D. avus extinction happened very recently, perhaps as recently as the period following the arrival of Europeans. This opens the possibility that environmental modifications generated by European colonization could have caused the disappear-ance of this fox. Yet, the impact of pre-European populations cannot be discarded, and in fact both could be involved, as the introduction of cattle and horses changed the habits of hunter-gatherer societies in the Pampean and Patagonian regions (Huber & Markgraft, 2003;Garavaglia, 2012). Similarly, the new dates overlapped with the introduction of domestic dogs in these regions (first by early settlers around 1000 years BP, and later by Europeans; Cabrera, 1934;Prates et al., 2010;Fig. 5). Large packs of feral dogs were recorded in the Pampean region in the 18 th and 19 th centuries (Cabrera, 1934), and the influx of European dogs may have led to an extensive replacement of the native American dogs (Castroviejo-Fisher et al., 2011;but see Van Asch et al., 2013). Domestic dogs could have a negative impact on D. avus, as vectors of canid-related diseases such as rabies or canine distemper virus, and as competitors for similar prey species as it happens in extant wild canids , or through direct predation using dogs as hunting aids (see below).
Mid-sized canids are typically generalist and therefore it is intriguing to speculate as to why D. avus would have been so heavily affected by these potential threats, especially taking into account that it was a medium-sized canid with a generalized dentition and diet (Prevosti & Vizca ıno, 2006;Prevosti et al., 2011;. Some living canids, especially the larger, more carnivorous species (e.g. Canis lupus, Lycaon pictus) or those limited to small geographic areas (e.g. C. simensis), are more likely to be heavily affected by burgeoning human populations, and to suffer range and population reductions, but they still manage to subsist in more remote, wilderness areas (Mech & Boitani, 2003;Sillero-Zubiri et al., 2004). In contrast, other more generalist canids (e.g. Canis latrans, C. aureus, Lycalopex culpaeus) have experienced recent population or distribution expansions (Novaro, 1997;Sillero-Zubiri et al., 2004). Thus, it is difficult to understand why these threats could be the cause of the extinction of D. avus.
Stable isotopes suggest that D. avus in the Late Pleistocene of southern Patagonia had a more carnivorous diet than living foxes, and could scavenge large and mega-mammals . The d 13 C and d 15 N values are consistent with a carnivorous diet (Supporting Information, Appendix S8), but display some variability (especially in d 13 C values). R ıo Luj an 1 specimen has a much higher d 13 C values which indicate the consumption of C4 rather than C3 plant feeders). This could be a valid interpretation because C4 plant feeders (i.e. cervids) were present in the northeast of Buenos Aires province around the time of the D. avus burial (Loponte & Corriale, 2013), although we cannot discard an alternative explanation that this fox was kept as a pet by native peoples (Prates, 2014). More stable isotopic data are needed to test this hypothesis, but the variability observed in the limited isotopic data currently available is congruent with some degree of generalization in the fox's diet and large areal distribution (see above).

A COMBINATION OF CAUSAL FACTORS?
Changes in population densities and the behavior of early settlers in the southern cone of South America, followed by the impact of the European colonization and a climatic trend to more humid conditions, could have come together during the late Holocene in such a manner that there was a strong impact upon D. avus populations that had probably already suffered some reduction, as has been proposed for the Pleistocene-Holocene megafauna extinction (Cione et al., 2008). In this scenario D. avus would have been more heavily affected because it was more carnivorous and larger than other South American foxes (Prevosti & Vizca ıno, 2006;Prevosti et al., 2011;. Furthermore, the few genetic data available would indicate that the species had low genetic variability (Austin et al., 2013), something that could be related to low population densities or a population bottleneck (Chan et al., 2005;Chan, Anderson & Hadly, 2006). If this inference is correct, D. avus would have been more sensitive to external threats (e.g. human impact, climatic change, or in combination; Frankham, 2005;Jansson et al., 2012;Frankham, Bradshaw & Brook, 2014;Marris, 2014; but see Rodr ıguez et al., 2011) than other canids. Conversely, D. avus had a wide distribution, apparently restricted to open environments (e.g. Patagonian steppe, Pampas), while Lycalopex culpaeus is also present in forested region in Patagonia and widely distributed along the Andes to northern South America Sillero-Zubiri et al., 2004). Other living southern cone canids are also restricted to open environments and their distributions are not necessarily larger, but also occur in more closed environments such as the Monte or Chaco (e.g. Lycalopex gymnocercus; Sillero-Zubiri et al., 2004). The more open and nonforested environments would have had fewer areas offering refuges for this species, exposing it to human impact.
Summing up, the new data and the available information clearly support a 'multi-causal' process, in which environmental change and human impact interacted causing the recent extinction of D. avus. to improve the English. Silvina Rodr ıguez and Jos e Pavoni for assistance with fieldwork in Pen ınsula Vald es. Luis Borrero provided access to new D. avus specimens. Herv e Bocherens and Martin Cotte performed the isotopical analysis of MCA 81-VI-1-1. We thank two anonymous reviewers for their helpful comments. FMNH, AMNH, FLMNH and CONICET provided study collection grants. Thanks to Agencia (PICT 2008-0547 to UFJP; PICT 2011-309 to FJP), and CONICET (PIP 164 to FJP), UNLU (Disp. CDD-CB 328-14 to FJP), FONDECYT (1100822 to FMM) and UMAG (CD MAG0901 to FMM) for financial support. Thanks to Secretar ıa de Cultura, Turismo y Areas Naturales Protegidas of Chubut for permission to work in ANP Pen ınsula Vald es to DEU, and to Fundaci on Vida Silvestre Argentina (Alejandro Arias, Rafael Lorenzo and Esteban Bremer) for providing facilities and logistical support at wildlife reserve San Pablo de Vald es.