Bone modifications by the giant hyaena Pachycrocuta brevirostris on large-sized ungulate carcasses from the Lower Pleistocene site of Tsiotra Vryssi (Mygdonia Basin, Greece)

ABSTRACT The Early Pleistocene mammal communities of Europe are characterised by a great diversity of large carnivorans. Among them, the largest ever hyaenid, Pachycrocuta brevirostris, a fierce predator with great bone-cracking adaptations that has left its taphonomic signature on several fossiliferous sites. Here, we perform a rigorous taphonomic analysis focusing on bone surface modifications and damage patterns on large-sized ungulate bones from the site Tsiotra Vryssi (1.78 to ~1.5 Ma; Mygdonia Basin, Greece), aiming to identify the main biotic agent responsible for the modifications. Results reveal significant carnivore ravaging of the assemblage, and selective consumption of bones/bone portions related to nutrient value. Comparisons with modifications on similar-sized ungulate carcasses produced by extant and extinct predators, and similarities with Pachycrocuta-modified assemblages, indicate that Pachycrocuta was the principal agent of modification. Overall, this study not only provides taphonomic evidence for the interpretation of Tsiotra Vryssi, but also offers insights into the palaeobiology, and particularly carcass consumption behaviour of the giant hyaena. Hence, it advances our knowledge on carnivoran guild dynamics and prey–predator relationships during this epoch and has important implications for the investigations of the subsistence behaviour of the meat-eating hominins, who entered Eurasia at ~1.8 Ma, roughly synchronously with Pachycrocuta.


Introduction
The Early Pleistocene of Europe is characterised by a major reorganisation of the carnivoran guild, including the arrival of several canids, felids and hyaenids (Sardella and Palombo 2007;Konidaris 2022).Among the latter, the giant hyaena Pachycrocuta brevirostris is of particular importance because of its great impact on terrestrial ecosystems.This taxon represents the most important fossil accumulator during the Early Pleistocene of Eurasia (Martínez-Navarro 2010), leaving its taphonomic signature in several fossil sites geographically distributed from East to West (e.g.Saunders and Dawson 1998;Boaz et al. 2000;Mazza 2006;Dennell et al. 2008;Palmqvist et al. 2011;Madurell-Malapeira et al. 2017;Arnold 2022;Kahlke 2022).The species probably originated in eastern Africa and dispersed to South Africa, Asia, and finally Europe, where it entered at ~2.0 Ma ('Pachycrocuta brevirostris event'; Martínez-Navarro 2010) and persisted until ~0.8 Ma (Iannucci et al. 2021).Pachycrocuta is found in numerous sites, some of which with evidence of hominin presence, e.g.Fuente Nueva-3 and Barranco León in Spain, Pirro Nord in Italy, Dmanisi in Georgia, and Zhoukoudian in China (Boaz et al. 2000;Tappen et al. 2007;Arzarello et al. 2007;Pavia et al. 2012;Espigares et al. 2019), and it is regarded as the most direct competitor of hominins for access to large mammal carcasses (Martínez-Navarro 2010).Competition between Pachycrocuta and early humans for scavenging food resources is suggested for Fuente Nueva-3 (Espigares et al. 2013(Espigares et al. , 2021)), while consumption of humans by the giant hyaena is proposed for Zhoukoudian (Boaz et al. 2000).However, the exact relationships among hominins, Pachycrocuta and other large carnivorans are not totally clear, emphasising the need for detailed and multidisciplinary studies.
Taphonomy is at the centre of such investigations and can contribute to the research of the subsistence strategies and behaviour of extinct species, including interspecific relationships such as that between early hominins and other large predators.Although taphonomic processes such as transport, disarticulation and breakage result in loss of information about the initial thanatocoenoses, they leave taphonomic signatures that can provide valuable indirect insights into the palaeobiology of fossil species and the formation of fossiliferous sites (Behrensmeyer and Kidwell 1985;Wilson 1988).In particular, the study of bone modifications is an important part of taphonomic analyses and interpretations because it can provide information about the processes that lead to the formation of a bone assemblage, by identifying the agents that contributed to its creation (e.g.Fisher 1995).
The present study aims to contribute to the taphonomic interpretation of the Lower Pleistocene site Tsiotra Vryssi in the Mygdonia Basin (Greece).Previous studies have recognised that carnivores had an important role in the modification, and possibly the primary accumulation of the skeletal remains (Konidaris et al. 2015;Giusti et al. 2019).In the present study, we perform a multitude of taphonomic analyses in order to thoroughly investigate the extent to which carnivore alteration is present, as well as to identify the main biotic agent responsible for the bone modifications of large ungulate carcasses.To achieve this, we consider two taphonomic traces, bone breakage and bone surface modifications, and we perform diverse analytical methods, following the 'physical attribute approach' of Domínguez- Rodrigo et al. (2007).This approach is based on the notion that biotic agents produce specific bone breakages and bone surface modifications that are taphonomically detectable, and as such they should play a major role in taphonomic analyses and interpretations (Domínguez-Rodrigo et al. 2015b).
Two main depositional units have been identified at TSR, Geo 1 and Geo 2, of which the ~1 m thick fossiliferous Geo 2a consists of pale brown-to dark yellowish brown very fine sandy silts, locally intercalated by cm-thick lenses of medium-coarse grained sands (Giusti et al. 2019).Previous spatial taphonomic analyses of the vertebrate assemblage, in agreement with sedimentological and micromorphological observations, suggested that multiple dispersion events occurred in the formation of the deposit: the recurrent spatial re-arrangements of a lag, (peri)autochthonous assemblage being consistent with the cyclical lateral switching of a braided fluvial system (Giusti et al. 2019).Cosmogenic radionuclides, magnetostratigraphy and biochronological data indicate an age between 1.78 and ~1.5 Ma (late Villafranchian, Lower Pleistocene) for the site (Konidaris et al. 2021(Konidaris et al. , 2022)).

Materials
The study analyses all postcranial bones (n = 749) of ungulates belonging to size groups 4 (i.e.150-349 kg) and 5 (i.e.350-1000 kg), as defined by Palombo (2010); these groups roughly correspond to the size groups 3 and 4 of Bunn (1982).We focus on size groups 4 and 5 due to the abundance of specimens compared to the other size groups represented in the assemblage.The studied size group 4 includes both the medium-and the large-sized Equus species (n = 436), while size group 5 includes the Bovini Bison cf.degiulli and Leptobos sp.(n = 220), the giraffid Palaeotragus sp.(n = 9) and the deer Praemegaceros sp.(n = 7).Specimens that could not be identified taxonomically but belong to the aforementioned size groups are also included in some of the analyses (n = 77).Concerning the collection protocol of fossils during the excavation, collected were all diagnostic (to skeletal element) bones and teeth, as well as carnivore modified bones, regardless of their preserved size; not collected, yet recorded (location, dimensions) with the total station were non-diagnostic bone fragments with length of ≥50 mm, while non-diagnostic and noncarnivore modified fragments ≤50 mm were not recorded (Giusti et al. 2019).Although this sampling protocol introduces few limitations in some of the analyses included in this study, we believe that these are only minor and that the studied sample is sufficient for a safe interpretation of the site.The studied material includes specimens collected during the excavation seasons of 2014-2019 and is stored at the Museum of Geology, Palaeontology and Palaeoanthropology of the Aristotle University of Thessaloniki (LGPUT).

Frequency and spatial distribution of bones with evidence of carnivore gnawing.
In order to estimate the role of carnivorans in the primary accumulation of the bone deposit, we examined the frequency and spatial distribution of carnivore alterations in our sample of ungulate postcranial bones.Specifically, the relative spatial distribution of modified (with clear evidence of gnawing in the form of tooth marks, furrowing and notches) versus non-modified bones allowed us to verify the spatial extent of our sample and the presence of spatial structures ascribable to particularly intense and/ or better preserved carnivore activities.Spatial distribution maps were produced using a subset of the studied sample, including only specimens whose position was recorded during 2015-2019 with a total station.

Body mass estimations.
In order to classify the different taxa into the size groups of Palombo (2010), their body mass was calculated.The body mass of the Equus individuals was calculated according to the method of Eisenmann and Sondaar (1998) for estimating weight using metapodial variables.Only metapodials that belong to complete/partially complete front and hind limbs of subadult juvenile and adult individuals were taken into consideration.These individuals were identified following the epiphyseal bone fusing sequence in horses (Budras et al. 2009).More specifically, for the front limb, the criterion used was the fusion of the distal radial epiphysis that occurs between 20 and 24 months of age.For the hind limb, the criterion used was the fusion of the tibial distal epiphysis that occurs between 17 and 24 months of age.The equations for weight estimation proposed by Eisenmann and Sondaar (1998)  where MC10 and MT10 refer to the distal supra-articular width of the third metacarpal and metatarsal, respectively, while MC13 and MT13 refer to the distal minimal depth of the medial condyle of the third metacarpal and metatarsal, respectively (Eisenmann et al. 1988).
The body mass of Palaeotragus sp. and Praemegaceros sp. was calculated according to the prediction equations of Damuth and MacFadden (1990).The equations used regard both dental and postcranial measurements.More specifically, for Palaeotragus dental elements, the equations used were those regarding all selenodonts, based on the length of the first, second and third lower molars.For postcranial elements, the equations regarding artiodactyls were applied, based on metatarsal measurements (M1-M7).For Praemegaceros dental elements, the equations for all selenodonts were applied for the length of first and second upper molars.For postcranial elements, the equations used regard cervids only, for measurements of the distal tibia (T4, T5).For each genus, the mean value was calculated for the mass values resulting from dental and postcranial measurements separately, and then the total mean from those two results was computed.This last mean is the value used as the estimated body mass.The body mass of 600 kg of Bison cf.degiulli was taken from Kostopoulos et al. (2018), and the body mass of 426 kg of L. etruscus by Maniakas (2019) is used for Leptobos sp.

Skeletal part representation.
The representation of skeletal parts is provided in terms of number of identified specimens (NISP), minimum number of elements (MNE), minimum number of individuals (MNI), minimum animal units (MAU), and standardised minimum animal units (%MAU) (Binford 1984;Lyman 1994).The MNE was calculated considering anatomical overlap between fragments by determining their location on the complete bone, as well as the size of individuals and their ontogenetic age based on the fusion of epiphyses.Additionally, the percentage of articulated and isolated elements was calculated for all long limb bones of Equus and Bovini.As articulated, we regard two or more adjoining elements that appear connected (Fourvel and Mwebi 2011).

Bone breakage patterns.
Fractures in diaphyses are examined for all limb long bones following the methodology proposed by Villa and Mahieu (1991).The following features are considered: fracture angle, fracture outline and fracture edge.All studied fractures refer to those defining the limits of each specimen.Fracture angle refers to the angle between the fracture surface and the bone cortical surface and is classified as follows: 1, oblique; 2, right; and 3, mixed (when the fracture exhibits both oblique and right angles).Fracture outline considers only the proximal and distal fractures of shaft fragments, either splinters or more complete, and is classified as follows: 1, transverse (for straight fractures that appear transverse with respect to the long axis of the specimen); 2, curved (for spiral or V-shaped fractures); and 3, intermediate (for straight diagonal and stepped fractures).Fracture edge refers to the texture of the fracture and can be either smooth or jagged.
Notches were classified and measured according to the criteria and methodology of Capaldo and Blumenschine (1994); only normal notches were studied.These are defined as 'semicircular to arcuate indentations on the fracture edge of a long bone that are produced by loading on cortical surface that removes a single or a nested series of bone flakes and leaves a negative flake scar that extends through the entire thickness onto the medullary surface' (Capaldo and Blumenschine 1994).The notch (cortical surface) and negative flake scar (medullary surface) breadth and depth were measured for all complete notches.Additionally, the ratios of double opposing:complete notches and double overlapping:complete notches were computed and plotted along data from Domínguez- Rodrigo et al. (2007Rodrigo et al. ( , 2015b)).
Bone damage patterns are examined and quantified for each long limb bone element using two different methodologies: 1, by recording the surviving bone portions (proximal epiphysis, proximal shaft, middle shaft, distal shaft, and distal epiphysis) according to Marean and Spencer (1991); 2, by classifying each long limb bone element to one of the 16 taphotypes as defined by Domínguez- Rodrigo et al. (2015b).Furthermore, the possible correlation of abundance of long bone epiphyses with mineral density, marrow cavity volume, and fat weight was examined through regression analysis following the methodology of Palmqvist and Arribas (2001).The bone mineral density values are expressed in grams per cubic centimetre (g/cm 3 ) and the marrow content in grams (g).Data for mineral density are taken from Lam et al. (1999), and for marrow content from Outram and Rowley-Conwy (1998), both regarding modern horses; data on the fat weight were taken from Brink (1997) for modern American bison.Additionally, the correlation between %MAU and the standardised food utility index (S) FUI proposed by Metcalfe and Jones (1988) was examined, with data for horses from Outram and Rowley-Conwy (1998) and for bovids from Emerson (1993).Lastly, the transport strategy/selectivity of equid and bovid carcasses was investigated by computing the Shannon evenness index as defined by Faith and Gordon (2007): evenness = -Σp i * lnp i /lnS, where S is the number of bone types and p i the standardised proportion of specimens of the i-th element.
Carnivore tooth mark analysis.All specimens were examined for the presence of carnivore modifications under strong light and magnification.Tooth marks were identified and classified according to the criteria proposed by Pobiner (2007) and Binford (1981).Tooth pits and punctures have a circular, oval, or polygonal shape and a bowl-shaped cross section.Their long axis is at most three times the length of their short axis.They are formed as pressure is directly applied from teeth on the surface of bones.Pits are relatively shallow deformations of histological structure that do not penetrate the whole thickness of compact cortical bone, while punctures are generally larger and are formed when the bone collapses under the tooth, and may penetrate all layers of compact cortical bone (Binford 1981, p. 45, figs. 3.01, 3.02).Tooth scores and furrows are linear with U-shaped cross sections that usually have a smooth base.Their length can vary as well as their orientation, which is usually more or less perpendicular or transverse in relation to the long axis of long bones.Scores and furrows are formed when teeth are dragged across the bone surface.Tooth scores are shallower and smaller than furrows and, unlike the latter, they do not penetrate the cortical surface of compact bone.Furrowing, defined as the deletion caused by the gnawing of cancellous bone (Haynes 1980;Domínguez-Rodrigo et al. 2012), was also recorded.
The percentage of elements displaying direct carnivore modification (tooth marks, furrowing and/or notches) was calculated separately in Equus and Bovini for cervical vertebrae, scapulae, pelves, humeri, radio-ulnae, metacarpals, femora, tibiae, metatarsals, calcanei, and phalanges.Additionally, for all long limb bones, the presence of tooth marks was recorded for each bone portion.The spatial distribution of tooth marks is visualised by employing GIS image analysis, as proposed by Parkinson et al. (2014), with the use of the QGIS software (https://www.qgis.org/en/site/).The exact location of each tooth mark was plotted for the anterior, posterior, lateral, medial, proximal, and distal view of each bone.All marks were plotted on elements of the right side.A different layer was created for each tooth mark category (pits, scores, and punctures), where information about the dimensions (length and width), bone portion and specimen ID were also recorded.Pits and punctures were plotted in point feature layer files, while scores were plotted in linear feature layer files.Furthermore, an additional point feature layer file was created where all types of tooth marks were plotted.The scores were represented by points in the middle of each linear feature that corresponded to the initial score.This layer file was used to conduct a spatial analysis for all tooth marks.For each element, a heatmap of tooth mark density (Kernel Density Estimation) was created, for all views of the bone that exhibited tooth marks.The GIS method was applied to equid and bovid specimens separately, for each different type of long bone.Cervid, giraffid and taxonomically unidentifiable bones were excluded, as their number was very small (in many cases only one specimen for each element was available) for this kind of analysis.
All identified pits, scores and punctures were measured with the use of a digital calliper at 0.01 mm precision directly on the bones, under strong light and magnification.The measured length refers to the maximum dimension of the mark, while breadth refers to the maximum dimension transversal to length (Andrés et al. 2012).Tooth marks with unclear borders due to extensive (overlapping) pitting or scoring were not measured.The mean length and width values of the tooth marks were plotted along with data from actualistic studies conducted by Domıńguez- Rodrigo and Piqueras (2003), Delaney-Rivera et al. (2009), Sala et al. (2012Sala et al. ( , 2014)), Andrés et al. (2012), andYravedra et al. (2014).

Frequency and spatial distribution of bones with evidence of carnivore gnawing
When considering all remains included in this study, bones with clear evidence of carnivore modifications (in the form of tooth marks, furrowing and notches, i.e. not including those showing only fresh bone breakage) account for 16.3% of the studied sample.When we additionally include green breakage of long limb bones as evidence of carnivore alteration, the percentage increases to 17.8%.Visual examination of the horizontal and vertical distributions of the sample subset (Figure 1) suggests that specimens preserving clear evidence of carnivore modifications appear to be homogeneously distributed in the three-dimensional space, forming few clusters where the rest of the assemblage is also denser.Therefore, there is no evidence of spatial segregation of specimens showing carnivore modifications versus non-modified ones, or spatial structures ascribable to particularly intense and/or better preserved carnivore activities.

Body mass estimations
The mean body mass (subadult juvenile and adult individuals, i.e. excluding young juveniles) of the medium-sized Equus species is estimated at 257 kg, and for the large-sized one at 343 kg.They both fall in size group 4 (150-349 kg) of Palombo (2010), and therefore, they are treated together in the analyses regardless of their specific attribution.For Praemegaceros, the mean body mass, estimated from both cranial and postcranial variables, is 478 kg, classifying it in size group 5 of Palombo (2010) (350-1000 kg).For Palaeotragus, the estimate for the mean body mass is 896 kg, classifying it in size group 5, as well.Considering also a mean body mass of 426 kg and 600 kg for Leptobos and Bison, respectively (Kostopoulos et al. 2018;Maniakas 2019), the entire Bovini sample is included in size group 5.

Skeletal part representation
The representation of skeletal parts is provided in terms of NISP, MNE, MAU, and %MAU (Table 1; Figure 2A).All values are provided separately for Equus, Bovini, Palaeotragus, Praemegaceros and taxonomically unidentified size group 4/5 elements.MNE values show that metapodials are abundant in both Equus and Bovini.Equus tibiae are very abundant as well.The femur is the least represented element in both.Palaeotragus and Praemegaceros are represented only by few specimens.MAU values clearly suggest a preferential preservation of tibiae and metapodials, and an underrepresentation of axial elements (even when considering the 12 unidentified vertebrae at taxon level, the MAU of which was not calculated).MNI was calculated from the MNE values of tibiae in Equus and metapodials in Bovini.Equus (medium-and large-sized species together) is represented by a minimum of 15 individuals (13 of which are classified as subadult juveniles/adults) and Bovini (Leptobos and Bison together) by 7.
The percentage of articulated elements has been correlated, among others, with prey availability (e.g.Fourvel and Mwebi 2011) and predator species (e.g.Haynes 1982).Figure 2B1, B2 displays the proportions of articulated and isolated elements for Equus and Bovini, respectively.The percentages of articulation were computed by dividing the number of articulated elements by the NISP for each element.Anterior and posterior phalanges, carpals and tarsals are grouped together, with the exception of the cuboscaphoid and malleolus in Bovini.Phalanges, carpals, tarsals, and sesamoids are usually articulated, with relatively lower percentages for Equus calcanei and astragali.In Equus isolated elements are prevalent with the exception of radio-ulnae, metatarsals (articulation with the lateral metatarsals is included) and cervical vertebrae, where articulated elements surpass 50%.The percentage of articulated metapodials in Equus is actually slightly lower, due to the presence of isolated metapodials (n = 6) that cannot be classified either to the front or hind limb due to their lowgrade/incomplete preservation and were therefore excluded from this analysis.In contrast, most elements in Bovini appear articulated more often, with the exception of metapodials, scapulae, and pelves.Notable is also the number of complete/partially complete articulated limbs: three front and seven hind limbs in Equus and three front and four hind limbs in Bovini.These are associated with green fractures that often display crenulated edges of the most proximal element and/or with tooth marks (90.0% of equid and 85.7% of bovid limbs) (Figure 2C, D).

Pattern
Percentages of fracture variables are depicted in Figure 3.The majority of outlines are curved (81.8%), followed by intermediate (14.6%), while transverse outlines are scarce (3.6%).Fracture edges appear mostly smooth (80.3%) versus jagged (19.7%).Angles appear oblique in 54.7% of fractures, right in only 8.8%, while mixed angles are also common (36.5%).The high percentages of curved outlines, smooth edges and oblique angles (see Figure 12B,  C) suggest that most studied fractures are green, indicating that they occurred when the bones were still in fresh condition (Villa and Mahieu 1991), i.e. retaining the edible tissues during the nutritive phase (Domínguez- Rodrigo et al. 2022).
A total of 29 complete, opposing and Type C notches were identified.Measurements of all needed variables (breadth and depth for both notch and flake scar) were possible for 27 of them.The mean, standard deviation, minimum, and maximum values for breadth and depth of notches are presented in Table 2.The ratios of double opposing:complete notches and double overlapping:complete notches are plotted in Figure 4, along with data from the Maasai Mara hyaena den sample and experimental human-butchered bone assemblages (Domínguez- Rodrigo et al. 2007Rodrigo et al. , 2015b)).The TSR sample appears closer to the Maasai Mara den, with very close ratios regarding double overlapping:complete notches (1.02 in Maasai Mara and 1.0 in TSR) (see Figure 12B for overlapping notches).
In order to quantify bone destruction in the TSR long bone sample and identify whether any specific pattern exists, bones were classified into taphotypes (T) (Domínguez- Rodrigo et al. 2015b).The taphotypes for Equus and Bovini are shown in Figure 5 and expanded upon below.
Humerus: In equids humeri display intense deletion of both ends, with a high percentage of isolated shafts (T15: 36.4%) and a significant percentage of complete elements (T0: 18.2%).The proximal epiphysis is always absent in partially deleted bones, while the majority preserve the distal end (T1, T3-T5).In bovids, there is a similar percentage of complete bones (14.3%).The deletion is limited to the proximal half (T1, T3-T6).
Radius: In equids a significant number of radii appear complete (T0: 27.3%).A considerable number retain only the distal half (T4 and T5: 27.3% and 18.2%, respectively).9.1% of the elements retain only the distal epiphyses (T6), while another 9.1% appear almost complete, lacking the distal end.The percentage of complete radii in the bovid sample is substantially higher (T0: 71.4%).The remaining radii present deletion of the distal end, with 14.3% belonging to T9, while another 14.3% presents more intense deletion (T12).
Since the great majority of the breakages occurred in the fresh stage during the nutritive phase, the classification into taphotypes can reveal the consumption sequence for each element.Humeri and femora display a proximodistal consumption sequence, with both epiphyses frequently deleted in advanced stages of consumption in the case of equids (Figure 6A, C, F, H).For radii, the consumption sequence is usually proximodistal in equids and always distoproximal in bovids (Figure 6B, G), while in tibiae it is proximodistal for both groups, with a very high survival rate of distal epiphyses (Figure 6D, I).Metapodials usually appear complete, less so in Bovini, where partial deletion is more frequent, usually in the distal half (Figure 6E, J).Long bones are frequently reduced to shaft fragments that often display tooth-marks and notches (Figures 6K  and 10F, G, H).

Selection
Since the abundance of elements and bone portions appears to be non-random, further analysis is performed in order to determine which variables can explain the selection of specific parts.For this purpose, the possible correlation of mineral density, marrow cavity volume, fat weight and abundance of long limb bone epiphyses was examined for Equus (Figure 9A, B) and Bovini (Figure 9C).
Figure 9A shows a statistically significant positive correlation between epiphysis abundance and bone mineral density, described by the following equation:   In Bovini there is some correlation between the abundance of epiphyses in the Bovini sample and their fat weight (Figure 9C).They demonstrate a negative correlation described by the following equation: Abundance = -0.0267(fat weight) + 8.7877, R 2 = 0.53 The regression analysis results show correlation between nutrient content of bones andtheir deletion.Bone portions high in nutrients (high fat and marrow yields) are present in lower frequencies, whereas those with low nutritive value (high mineral density) are more abundant.This correlation suggests selective bone portion survivorship.
Furthermore, the standard food utility index shows some correlation, with both the %MAU of equid long bone epiphyses (Figure 10A) and that of bovid whole-limb bone elements (Figure 10B).
Shannon's evenness index values are 0.933 for equids and 0.904 for bovids.They both fall inside the bulk strategy range (0.927-0.989 for equids where MNE = 100, and 0.899-0.987for bovids where MNE = 51; Faith and Gordon 2007).This suggests a moderately selective transport, where the quantity of all elements is maximised, except in the case of those with the lowest utility.This moderately high degree of evenness indicates either a selective transport to a certain extent, transport distance may have been relatively short or long limb bones were acquired at rather short distance away from their final accumulation site.Such a degree of evenness is compatible with the presence of complete/partially complete articulated limbs and, overall, the abundance of articulated specimens.However, we must consider that relatively small sample sizes, such as in our case, are prone to give false correlations in the case of the bulk strategy model (Faith and Gordon 2007).

Anatomical and spatial distribution
Tooth marks appear frequently in the assemblage.The most abundant types are pits (n = 475), followed by scores (n = 201), whereas punctures are rare (n = 11).The percentages of gnawed elements for the Equus and Bovini samples are presented in Figure 11A.Only the elements that bear carnivore modifications in either group are portrayed.In the case of vertebrae only the cervical ones are considered.Furrowing is present in almost all long limb bones (45.0% of humeri, 5.0% of radii, 66.7% of ulnae, 80% of femora, 10.3% of tibiae, and 2.8% of metatarsals), as well as in cervical vertebrae (7.7%), scapulae (6.7%), pelves (40.0%), and notably a few calcanei that exhibit intense deletion (16.7%).Some examples of carnivore modification are shown in Figure 12.
Upper limb bones appear more frequently gnawed in general (presenting tooth marks, furrowing and/or notches) in both Equus and Bovini.The femur is the most gnawed element in both categories (85.7% and 100%, respectively), while the percentage of modified humeri is considerably higher in Equus (72.7% as opposed to 57.1% in Bovini).Intermediate limb bones show significant percentages of gnawing (36.4% for radio-ulnae in Equus and 50.0% in Bovini tibiae as well as radio-ulnae).Notably, 69.2% of Equus and 50% of Bovini tibiae appear gnawed.Lower limb bones exhibit considerably lower percentages.Metapodials, in general, appear gnawed in moderate percentages that do not surpass 40.0%.Equus scapulae are moderately gnawed (18.2%), while in Bovini none show gnaw marks.This could be attributed to the low number of scapulae in the latter (two specimens).In the Bovini sample, the pelvis appears 100% gnawed, but it is represented only by a single specimen.
Table 3 presents the frequency of tooth-marked bone portions per long bone element in Equus and Bovini.The anatomical distribution of tooth marks can indicate the predator taxon, since different carnivores modify bone portions in diverse frequencies (Haynes 1983;Parkinson et al. 2015).It is important to note that even some areas that are represented by a very low number of specimens, such as femora and humeri epiphyses in both taxa and tibiae proximal shafts in Equus, exhibit high frequencies.
In order to acquire a more comprehensive view of the spatial distribution of tooth marks (pits, scores, and punctures), a GIS spatial analysis was conducted for long limb bones of Equus and Bovini (Figure 13).The number of elements, as well as the number of tooth marks, is lower in Bovini (tooth mark n = 345 for Equus, and n = 102 for Bovini).Tooth mark clusters are visible in both groups and for all elements.However, their distribution varies among different elements, while differences are also evident between Equus and Bovini.Equus elements are tooth-marked on all aspects, while Bovini on all except posterior and lateral aspects of humeri, posterior tibiae and lateral metatarsals.Humeri show clusters that are limited in the distal half in Bovini, while in Equus significant clustering is visible across the entire diaphysis and distal epiphysis.Radii appear predominately tooth-marked on middle shafts, in both groups.Metacarpals exhibit substantially lighter tooth marking, with the presence of small clusters.

Metric analysis
The mean, standard deviation, minimum, and maximum values for length and breadth of scores, pits and punctures are presented in Table 2.The results are categorised by bone portion, separately for Equus, Bovini and collectively for the size groups 4 and 5 (besides Bovini including also Praemegaceros and Palaeotragus) specimens.It is notable that dimensions of pits and scores on the epiphysis do not always exceed those on the diaphysis.It can be hypothesised that once pits and scores reach the higher range of dimensions, they penetrate all layers of cortical bone and are presented as punctures and furrows, respectively.Punctures are substantially larger in cancellous bone with respect to the thin cortical bone of the near-epiphyses.
A comparison of mean length and width values was conducted, in order to gain a perspective on how the TSR sample compares to assemblages modified by extant species.Figure 14 presents a comparison between diaphyseal pit and score dimensions in TSR, and data from actualistic studies.The TSR sample in all three groups (Equus, Bovini and total size groups 4-5), parallels with data from extant large predators such as Panthera leo, Crocuta crocuta and Canis lupus.Notable is the wider range of dimensions in the TSR sample, especially compared to relatively small predators.Furthermore, the upper range of pit dimensions (on cortical bone) in the TSR sample seems to exceed that of most modern large predators.It is important to note that these comparisons can only provide a rough size distinction (small-/large-sized predator) because there are several variables besides the inflictor taxon that can influence the size of tooth marks, such as the size and ontogenetic age of both prey and predator.

Discussion
Several carnivoran taxa are present at TSR, belonging to the families Canidae, Ursidae, Hyaenidae and Felidae.These include mediumand large-sized canids (Canis spp.including the hunting dog Canis (Xenocyon) sp.), the bear Ursus etruscus, the hyaena Pachycrocuta brevirostris and the sabre-toothed cat Megantereon sp.(Konidaris et al. 2015(Konidaris et al. , 2021;;Koufos et al. 2018;Karakosta et al. accepted).In addition to these taxa, the large carnivoran fauna from the late Villafranchian of Mygdonia Basin also includes the lynx Lynx issiodorensis, the pantherine felid Panthera gombaszoegensis and the sabre-toothed felid Homotherium latidens (Koufos 2014(Koufos , 2018 and references therein); thus, most members of the large carnivoran guild of the European late Villafranchian (Konidaris and Tourloukis 2021; Konidaris 2022) are recorded in the basin.Each of these large carnivorans was equipped with great hunting, killing, or scavenging capabilities and dental specialisations related to their dietary preferences (hypocarnivorous, carnivorous, bone/meat, hypercarnivorous) (Konidaris and Tourloukis 2021; Konidaris 2022 and references therein) and could potentially modify and consume the TSR ungulate bones and their nutrient content.
In order to assess the involvement of extinct carnivores in the formation and alteration of fossil assemblages, numerous studies have utilised comparisons with modern analogous species (e.g.Blumenschine 1988;Arribas and Palmqvist 1998;Capaldo 1998;Domínguez-Rodrigo et al. 2007;Sala et al. 2012;Arilla et al. 2014).We follow the suggestion of Gidna et al. (2013), who propose focusing on actualist studies performed with wild carnivores when comparing gross bone damage, and tooth mark frequencies and  distribution.Bone modification in captive conditions has been reported to reach higher stages than those observed in the wild, due to the altered behaviour of animals in confined, drastically different living environments (see Pobiner 2007;Gidna et al. 2013;Sala et al. 2014).Moreover, we concentrate on limb bones, not only due to their abundance in relation to axial elements but also due to their capacity to better preserve taphonomic information (Cleghorn and Marean 2007).Several researchers have emphasised the need of a multivariate approach in order to make valid assumptions about bone accumulators/modificators at fossil sites (e.g.Domínguez Rodrigo and Pickering 2010; Domínguez- Rodrigo et al. 2012;Saladié et al. 2019).This study follows this notion, considering multiple aspects of carnivore alteration of skeletal assemblages, such as skeletal element survivorship, gross limb bone damage, and tooth mark dimensions, frequency and anatomical distribution.

Tooth mark dimensions
The value of tooth mark dimensions alone as indicator of the species responsible for bone modification has been debated.Overall, researchers agree that tooth mark dimensions can be used as an indicator of carnivore size and should be utilised in a multivariate context in order to lead to more specific conclusions regarding carnivore taxa.
In the TSR sample, pits on cancellous bone have a mean length of 3.38, 4.04 and 3.60 mm in the Equus, Bovini and total size 4-5 sample, respectively.For cortical bone, the mean length/breadth values for the same groups are 3.43/2.15,3.52/2.42,and 3.49/ 2.18 mm.The comparison of TSR pit dimensions on shafts (Figure 14) with measurements from modern experiments also suggests that the carnivoran responsible for the alterations was of large body size, analogue to modern brown bears, lions, spotted hyaenas and wolves.Notably, the upper range of pit dimensions on cortical bone seems to exceed that of most modern large predators.This is particularly important because in contrast to small tooth marks, the large ones can be inflicted only by large predators  (Fernández-Jalvo and Andrews 2016).It is important to note the dimensions of punctures on cancellous bone; mean lengths of 10.71, 6.60 and 8.66 mm and mean breadths of 7.14, 5.16 and 6.15 mm for Equus, Bovini and total size 4-5 sample, respectively, suggest the involvement of carnivores that possessed teeth of proportional size in order to produce such impressions.We also have to mention that since a juvenile hyaena mandible was found at the site (Konidaris et al. 2015) it is probable that at least part of the observed tooth marks were produced by juvenile individuals, particularly those marks that are in the lower size range.

Gross limb bone damage
Although light gnaw damage is not useful in the identification of the acting carnivore taxa, medium and heavy stages of bone modification can be discriminative (Haynes 1983).The consumption sequence for each long limb bone is presented in Figure 6.The upper limb bones are the most heavily modified elements in both Equus and Bovini, with high percentages of isolated shafts in the former (Figure 6A, C, F, H).Radii appear complete frequently, more so in Bovini.Tibiae, on the contrary, appear heavily modified with a very low percentage of complete elements.In both groups (equids and bovids) the majority of the tibiae lack the proximal epiphysis (Figure 6D, I).Metapodials appear mostly complete.The fracturing of midshafts (Τ3-5 and Τ10-12) is common in the sample.Furthermore, although not quantified in this study, the numerous indeterminate, often tooth-marked, shaft fragments present at the site indicate intense ravaging (Figure 6K).The majority of the studied fractures possess characteristics that classify them as green, indicating that they occurred when the bones were still fresh, i.e. during the nutritive phase.This, in addition to the numerous concurring tooth marks and notches, suggests that most studied fractures are a result of carnivore activity.
Concerning the bear Ursus, Medin et al. (2017Medin et al. ( , 2019) ) performed morphological and dental microwear analyses on teeth from Dmanisi (Georgia) and the Orce sites (Guadix-Baza basin, Spain), and concluded that this ursid followed an omnivorous diet, consuming both plant material and vertebrate flesh depending on availability, similar to the extant Ursus arctos.Observations of brown bears in the wild showed that they usually do not fracture ungulate long bones (Haynes 1983;Sala and Arsuaga 2013).Saladié et al. (2013) report fracturing by captive and semi-captive brown bears, although limited to small-sized and juvenile carcasses.Arilla et al. (2014) noted that the intensity of other types of modification (tooth marks, crushing and furrowing) is moderate, and not significantly affected by prey size.Although modern brown bears have been observed modifying bones, albeit only moderately, the isotopic analyses of Medin et al. (2017) for U. etruscus teeth from the Orce sites showed a significant contribution of fish and plant tissues in its diet.Therefore, the intensity of gross bone damage observed in the TSR sample and the dietary preferences of U. etruscus, disqualifies this carnivore taxon from being the primary agent of bone modification at the site.
Concerning felids, studies on wild lions reported that they do not inflict substantial gross damage on bones of medium-and large-sized ungulates, only causing moderate furrowing that is usually restricted to the proximal epiphysis of the humerus, the olecranon process, both femur epiphyses, and the proximal epiphysis of the tibia (Domıńguez-Rodrigo 1999; Pobiner and Blumenschine 2003;Pobiner 2007;Domínguez-Rodrigo et al. 2012;Pobiner et al. 2020).Like extant felids, sabretooths were also flesh specialists [hypercarnivorous, see e.g.Palombo (2016), i.e. with >70% meat in their diet ( Van Valkenburgh 1988)], as suggested by the lack of bone crunching dental adaptations (Hartstone-Rose and Wahl 2008; Hartstone-Rose 2011).Moreover, some researchers argue that they avoided tooth contact with bone, in order to prevent damage to their remarkably long canines (Emerson and Radinsky 1980;Valkenburgh and Ruff 1987;Valkenburgh et al. 1990;Arribas and Palmqvist 1999).The craniodental features of Megantereon suggest the exclusive consumption of soft tissues (Palmqvist et al. 2007).Its enlarged canines, in combination with a significant reduction of the premolars and the shortening of the coronoid process, as well as the possession of powerful forelimbs, suggest an increased ability in the killing of larger prey, paired with the consumption of exclusively soft tissues; therefore, leaving considerable amounts of flesh and bones for scavengers, like the hyaena Pachycrocuta (Martínez-Navarro and Palmqvist 1996; Palmqvist et al. 2007;Espigares et al. 2021).However, there also exists evidence of bone modifications by sabretooths.For instance, the study of a Homotherium serum den (Friesenhahn Cave, U.S.A.) revealed that this predator generated relatively high frequencies of tooth-marked elements (over 50% for Mammuthus columbi bone fragments) and was probably capable of consuming the unfused epiphyses of juvenile mammoth remains, while additional bone deletion was limited to small, nearepiphyseal bone potions (Marean and Ehrhardt 1995).Dental microwear analysis on Homotherium serum remains of the aforementioned locality suggested that this species consumed mainly tough and soft tissues and not bone (DeSantis et al. 2021).The Haile 21A accumulation (Florida, USA), modified by Xenosmilus hodsonae, revealed that this sabretooth was capable of considerable furrowing of upper limb bone and tibiae epiphyses of the peccary Platygonus vetus [50-150 kg, size class 3 of Palombo (2010), size class 2 of Bunn (1982) (see also Domínguez-Rodrigo et al. 2022, Table 1)].Besides sabretooths, the other large and hypercarnivorous felid of this period, Panthera gombaszoegensis, possessed a robust dentition that probably reflects its prey preferences (Jiangzuo and Liu 2020).Studies of bone damage by modern jaguars in captive conditions show that they are more capable than lions in furrowing epiphyses, nevertheless not reaching the levels of damage observed in hyaena modified assemblages (Domínguez- Rodrigo et al. 2015b;Rodríguez-Alba et al. 2019).A smaller-sized felid present in the late Villafranchian carnivoran fauna from Mygdonia Basin is Lynx issiodorensis.The diet of its putative descendant Lynx pardinus (Iberian lynx) consists mainly of rabbits, supplemented with other small mammals and birds and rarely small-sized ungulates (Rodríguez-Hidalgo et al. 2020).Lynx issiodorensis was larger than its modern counterpart (Mecozzi et al. 2021), and this was possibly reflected in the size of its prey.However, it is highly unlikely that it was able to hunt prey as large as the ungulates concerning this study, as well as to produce the intensity of carnivore damage observed at TSR.
Modern wolves (Canis lupus) have been observed fracturing ungulate bones to access marrow (Sala et al. 2014), as well as consuming parts of the bone itself, evident by the presence of bone fragments in their faeces (Esteban-Nadal et al. 2010;Fosse et al. 2012).Campmas and Beauval (2008) noted that the damage patterns produced by captive wolves and wild hyaenas on large ungulate carcasses, are indistinguishable.Although both hyaena and wolf modified assemblages can exhibit heavy modification of bones, hyaenas are generally more destructive (Haynes, 1983;Domínguez-Rodrigo et al. 2015b).Fosse et al. (2012) mention that gross bone damage caused by wild wolves on bison long limb bones is rather limited and concentrated on upper limb bone epiphyses.Carcasses altered by wild wolves have been shown to reach high degrees of modification (collapsed epiphyses) only in small-sized mammals, or juveniles of medium-sized species, as observed by Yravedra et al. (2011), where horse carcasses consumed by wolves showed significant bone modification in the range of 120-150 kg, and a very low degree of bone deletion on adult carcasses (350-420 kg).A 'kennel pattern' of more intense modification of large ungulate bones has also been observed when wolves repeatedly access carcasses in the wild, although it is rather uncommon (Haynes 1982).A similar pattern can perhaps be assumed for the several Canis species known from the late Villafranchian-Epivillafranchian of Europe: C. arnensis, C. etruscus, C. borjgali, C. orcensis, C. appolloniensis and C. mosbachensis, although their smaller size, closer to modern coyotes and jackals than wolves (Brugal and Boudadi-Maligne 2011) could suggest lesser bone processing capabilities and/or smaller prey size preferences compared to modern wolves.Pobiner (2007) observed that captive jackals caused very limited gross bone damage, and never fully defleshed carcasses, another fact perhaps pointing towards a more limited bone modification capacity of Early Pleistocene canids, in relation to modern wolves.
Not all modern large canids modify bones to the same degree.The African wild dog Lycaon pictus, which is characterised by hypercarnivory (Hartstone-Rose and Wahl 2008) has been shown to modify bones only moderately (Yravedra et al. 2014).The diet of the late Villafranchian-Epivillafranchian Canis (Xenocyon) lycaonoides consisted of ungulates with a body mass larger than its own (Palmqvist et al. 1999).Dental microwear analysis shows that it mainly consumed meat (Medin et al. 2017).Its forerunner, Canis (Xenocyon) falconeri is also considered to have been hypercarnivorous, since its craniodental morphology is similar to C. (X.) lycaonoides and to that seen in extant hypercarnivorous canids (Palmqvist et al. 1999).Nevertheless, Bartolini-Lucenti and Spassov (2022) support that Canis (Xenocyon) spp.possessed a more robust mandible compared to modern wild dogs, a fact that could reflect different feeding strategies (struggling of larger prey) or the consumption of tougher tissues, possibly even bone fragments.
Extant hyaenas vary in body mass and craniodental morphology, and these differences are reflected in their ability to modify carcasses.The larger body size and more specialised premolars of Crocuta crocuta (spotted hyaena) suggest a greater capability in bone cracking of large bones, compared to Hyaena hyaena (stripped hyaena) and Parahyaena brunnea (brown hyaena) (Ewer 1973;Mills 1990; Van Valkenburgh and Binder 2000).This is further supported by dental microwear analysis that reveals extreme bone cracking abilities for Crocuta crocuta, and a meat and bone diet for Hyaena hyaena (Bastl et al. 2012).A characteristic of hyaena modified assemblages is the predominance of limb bones, as they consume axial elements at kill sites (Binford 1981;Capaldo 1998).Haynes (1983) observed highly destructive behaviour in spotted hyaenas, where in the final stages of utilisation the femur was either reduced to short shaft fragments or completely consumed, and the tibia was proximally reduced until less than a third of the shaft survived.Hill (1989) reported that spotted hyaena modified assemblages display a high percentage of indeterminate fragments, with the majority of bones (especially limb bones) being fractured, and a very high proportion of complete metapodials.The recurring breakage of dense elements through their midshafts is another characteristic that suggests the involvement of hyaenas or possibly canids (Domínguez- Rodrigo et al. 2007).Bones modified by hyaenas display considerably heavier damage than those modified by felids.This motif is clearly observed in tibiae fed to captive lions, jaguars, and hyenas, where modification was limited to the proximal epiphysis and proximal shaft in the two felids (taphotypes 0-1 in lions, and 1-3 in jaguars), while extensive furrowing of both epiphyses and shaft breakage is observed in the hyaena sample (see Domínguez-Rodrigo et al. 2015b, Figure 8) .
During the late Villafranchian-Epivillafranchian, the sole hyaenid that appears in the European fossil assemblages is Pachycrocuta brevirostris (with the exception of Chasmaporthetes which is still present during the earliest late Villafranchian, and the appearance of Crocuta at the end of the Epivillafranchian) (Palombo 2014;Iannucci et al. 2021).This hyaenid had an average estimate of body mass of 110 kg (nearly twice that of the spotted hyaena) and unique craniodental adaptations, which indicate great bone fracturing capability (Palmqvist et al. 2011).Specialisations include robust premolars that along with the carnassials are positioned posteriorly, thus allowing the manipulation of larger bones, a higher, more resistant mandibular corpus, and a well-developed angular process that suggests the possession of large masseter and pterygoid muscles.These adaptations, along with a deeper mandibular corpus and a more developed symphysis compared to extant hyaenas, suggest that the extinct hyaenid possessed drastically higher bone-cracking capabilities than its extant counterparts (Palmqvist et al. 2011).Its postcranial skeleton also shows adaptations for scavenging, with a robust body and shortened distal limb bones, less suited for a cursorial lifestyle, but offering strength and stability for the dismembering and transport of carcass parts (Turner and Antón 1996).Nevertheless, a more opportunistic behaviour, which besides scavenging included also hunting, is also proposed for Pachycrocuta brevirostris (Turner and Antón 1996;Galobart et al. 2003;Dennell et al. 2008;Iannucci et al. 2021).The faunal and skeletal representation of Venta Micena (Spain), a site interpreted as a P. brevirostris accumulation, indicates that this hyaena selectively scavenged prey hunted by sabretooths, as well as the canid Canis (Xenoxyon) lycaonoides (Palmqvist et al. 1996).A pattern of selective deletion according to higher nutritional value has been observed in ungulate carcasses, something also seen in extant hyaenas (e.g.Palmqvist et al. 1996;Arribas and Palmqvist 1998;Leakey et al. 1999;Mazza 2006).The same can be observed in the TSR sample, where the preferential deletion of specific long limb bone epiphyses appears to be correlated with their nutritional value and mineral density.More specifically, epiphyses with low mineral density and high marrow yield appear less frequently in the Equus sample.Similarly, in the Bovini sample, the investigation of a possible correlation between fat content and epiphyses survivorship showed that epiphyses richer in fat were deleted more frequently.The preferential deletion according to nutritional value is reflected in the sequence of consumption of long bones that appears proximodistal in the humerus, femur, and tibia, mainly distoproximal in metapodials, and without a clear direction in the radius.In Venta Micena, a proximodistal direction in the humerus and tibia, distoproximal in metapodials, and without a clear direction in the femur and radius, but involving the deletion of both epiphyses, is observed (Palmqvist et al. 2011).The site of Vallparadís (Spain), layers of which are also interpreted as a Pachycrocuta accumulation, also shows similarities with TSR.Humeri and tibiae are usually preserved as distal fragments, radii as proximal, femora as cylinders lacking both epiphyses, while metapodials and compact bones are usually complete (Madurell-Malapeira et al. 2017).Remains of the cervid Eucladoceros dicranios in Poggio Rosso (Italy) also reveal a correlation of nutritive value and abundance of epiphyses, with an underrepresentation of the proximal epiphyses in humeri, metacarpals and tibiae, distal epiphyses in the radii and metatarsals, and both epiphyses in femora (Mazza 2006).Bison menneri remains from Untermassfeld display a consumption sequence very similar to TSR, with a proximodistal direction in humeri and tibiae, as well as in femora, where isolated shafts are also very abundant, and distoproximal in radio-ulnae and metapodials; the olecranon and tubera of the calcaneum are frequently deleted (Arnold 2022).Saunders and Dawson (1998) studied the Haro River Quarry (Pakistan), another assemblage accumulated by Pachycrocuta, and reported high levels of modification, some of which are similar to the 'kennel pattern' proposed by Haynes (1982), referring to the high level of damage caused by modern wolves in captivity.The intensity of the damage caused by Pachycrocuta in the Haro River Quarry is manifested in a variety of bone modifications that include deletion of epiphyses, bone shaft fragments, furrowing, pitting, scoring, punctures, notches, crenulated edges, and acid corrosion (Saunders and Dawson 1998).

Skeletal element representation and bone articulations
Although ribs cannot be taken into consideration, since their collection was not systematic during the early years of excavation, MAU values of vertebrae and long bones (Table 1, Figure 2A) show that axial elements are underrepresented in the assemblage, while limb bones are abundant (see also Giusti et al. 2019).These values could suggest hyaena involvement, as it has been observed that spotted hyaenas consume the axial skeleton at kill sites, thus creating assemblages where limb bones are predominant (e.g.Skinner et al. 1986;Capaldo 1998), whereas felids have been mostly shown to modify axial bones without deleting them (e.g.Domıńguez-Rodrigo 1999).However, it is also possible that the under-representation of axial elements is, at least in part, a result of fluvial sieve and transport, as suggested by Giusti et al. (2019).
The ratio of articulated to isolated elements has been correlated with accumulator type, where primary assemblages collected by predators comprise an abundance of articulated elements, while in secondary assemblages collected by scavengers only metapodials, phalanges and vertebrae appear articulated often (Palmqvist and Arribas 2001 and references therein).As an example, in Venta Micena only 20% of elements appear articulated (all size classes are considered) (Palmqvist and Arribas 2001).The prevalence of isolated elements has also been reported in hyaena den accumulations where prey was scarce, thus the consumption of carcasses reached higher stages (Fourvel and Mwebi 2011).In the case of TSR, considering all studied skeletal elements, 59.4% appear articulated in Equus and 77.5% in Bovini.However, among articulated elements, the majority are carpals, tarsals, as well as lateral metapodials in the case of Equus.Regarding long limb bones, in the Equus sample most appear isolated, with the exception of radii and metatarsals, where 54.5% and 56.5%, respectively, are articulated, in addition to ulnae that always appear articulated.On the contrary, most Bovini long limb bones are articulated, but notably 61.5% of metacarpals and 69.2% metatarsals appear isolated.Taking into account both the Equus and Bovini samples, isolated long bone elements predominate slightly, with a percentage of 52.2%.This does not give a clear signal as to the type of accumulator, but it could suggest an abundance in prey/scavengable resources and/or perhaps reduced competition among large carnivorans for carcasses of this size group.The latter is also supported by the presence of complete and almost complete elements (taphotypes 0 and 1).High percentages of articulation can also suggest ephemeral exposure before burial (Mazza et al. 2004).The relatively higher percentage of articulated elements in Bovini can be attributed to their larger size, since size is positively correlated with abundance of articulated elements, as a result of increased stability of muscles and tendons (Kahlke and Gaudzinski 2005).It is important to note that the proximal element of articulated legs often exhibits green fractures, in many cases associated with tooth marks.

Tooth mark frequency and anatomical distribution
The study of the frequency of tooth-marked bone portions and the analysis of spatial distribution of tooth marks revealed relatively intense tooth marking across all long limb bones.The femur is the most frequently tooth-marked element in both examined size groups, followed by the humerus and tibia.Notably, half of radioulnae are also tooth-marked.Metapodials exhibit noticeably lower frequencies, metacarpals in particular.Middle shafts are the most tooth-marked bone portions in all elements, except the femur, tibia, and metatarsal in Equus, where the distal, proximal shaft and distal end, respectively, present higher frequencies.When taking into consideration both Equus and Bovini, the only bone portions that appear free of tooth marks are the distal end of radii and metacarpals and the proximal end in tibiae and metatarsals.Concerning the spatial distribution of tooth marks, all limb bones display visible clusters.The diaphyses exhibit higher concentrations of tooth marks, nevertheless clusters are visible in both femoral epiphyses and the distal epiphyses of humeri and metatarsals.Tooth pits (n = 471) are more abundant than scores (n = 205) across all bone portions.
The most frequently tooth-marked limb bone elements in assemblages modified by modern wild lions are the upper limb bones, with most tooth marks occurring on long limb bone ends (Domıńguez-Rodrigo 1999;Gidna et al. 2014;Pobiner et al. 2020).Additionally, shafts bear more tooth scores than pits, in contrast to hyaena and wolf modified limb bones (Egeland 2008;Domínguez-Rodrigo et al. 2012;Gidna et al. 2014).A similar pattern has been observed in long bones modified by captive jaguars (Rodríguez-Alba et al. 2019).Concerning Homotherium, tooth marks have been observed in juvenile mammoth remains of the Friesenhahn Cave Homotherium den with frequencies similar to modern hyaenids and canids, although punctures and notches were scarce (Marean and Ehrhardt 1995).The study of the Xenosmilus hodsonae modified assemblage at Haile 21A showed that this sabretooth generated relatively low long bone tooth mark frequencies, with only a few tooth marks per element (Domínguez-Rodrigo et al. 2022).Compared to the TSR sample, the percentages of tooth-marked elements (displaying at least one pit, score, and/or puncture) are consistently lower across all long limb bones (Haile 21A (peccary)/ TSR (equids): 17.2/72.7%for humeri, 11.0/.25.0 for radio-ulnae, 3.9/16.0for metacarpals, 28.9/85.7 for femora, 11.7/61.5 for tibiae, and 3.4/30.4for metatarsals).Experiments with captive wolves showed heavy tooth-marking on all long bones, with lower degrees in midshafts, with the exception of radii (Parkinson 2013).Modern hyaenas show high variability in the frequency of tooth-marked bone portions, not only among different species but also between different assemblages created by the same species (see Faith 2007;Kuhn et al. 2009).In spotted hyaenas, all bone portions across all long bones appear tooth-marked in the experiments of Faith (2007), with higher frequencies for humeri proximal and femoral distal ends (100%), midshafts (82.4% for femora and 80.65 for humeri), and tibiae middle shafts (77.0%).The middle shaft consistently displays very high frequencies across all long limb bones (71.6-82.4%).Kuhn et al. (2009), considering all three species of hyaena studied (stripped, brown, and spotted), report the presence of tooth marks on all bone portions except proximal epiphyses in tibiae, and middle shafts in humeri and femora; humeri and femora are toothmarked in higher frequencies, with the highest value of 66.7% for humeri distal ends and both end of femora for elements modified by the brown hyaena.Blumenschine (1988) reported high frequencies (>75%) of tooth-marked mid-shaft fragments in assemblages where spotted hyaenas had primary access to bovid limb bones.In Untermassfeld, where P. brevirostris is considered the main agent of carnivore modification, 12.6% of Bison menneri remains display modifications, with long bones, mandibulae, pelves, and calcanei being the most affected (Arnold 2022).As is the case with gross bone damage, tooth mark clustering patterns can be interpreted as a spectrum of bone processing intensity, with large felids at the lower, large canids at the middle, and hyaenids at the higher end of the spectrum (Parkinson 2013).

Conclusions
The TSR assemblage exhibits high degrees of modification.Anatomical profile analyses reveal an underrepresentation of axial elements and of nutrient-dense epiphyses.Levels of articulation are relatively high, yet articulated elements are frequently gnawed.Fracture patterns reveal that fractures occurred mostly during the nutritive phase, while notch-type ratios are close to carnivore modified assemblages.The deletion of epiphyses and fracture of the diaphyses is common in the sample, with the severity of modification decreasing from upper, to intermediate, and finally lower long limb bones, and exhibiting great variety, with the presence of pits, scores, punctures, furrows, furrowing, and crenulated edges.Upper limb bones are the most frequently tooth-marked, with tooth mark clusters across almost all element bone portions, mainly on the diaphysis, and dimensions of pits, scores, and punctures that suggest infliction by a large carnivore.The level of modification is higher on Equus than in Bovini, something expected from the greater body mass of the latter group.
All comparisons regarding skeletal part representation, gross bone damage, and tooth mark dimensions, frequency, and spatial distribution indicate that a large carnivoran was the primary biotic agent of bone modification at TSR.The intensity of modification further points to a carnivoran with systematic bone cracking behaviour in order to gain access to the bone marrow content.All things considered indicate the action of a hyaenid (i.e.Pachycrocuta brevirostris, the sole hyaenid present at the site) as the main candidate.The similarities of TSR with Venta Micena and other sites where Pachycrocuta is considered the main biotic taphonomic agent, further point to this taxon as the main agent of bone modification at the site.Although not examined in the present study, the presence of bone alterations (tooth marks, furrowing, bone fracturing) in megafauna (>1000 kg; Stephanorhinus spp., and Mammuthus meridionalis) also suggests an agent with formidable bone cracking capabilities, such as the giant short-faced hyaena.It is important to note that although the principal taphonomic signal points towards the short-faced hyaena, other carnivorans may have probably also contributed to the alteration of bones, but to a lesser degree.This is perhaps expected, considering that Pachycrocuta is believed to have mainly acted as a scavenger of carcasses abandoned by flesh-eating carnivorans, such as the sabretooths Megantereon and Homotherium (Palmqvist et al. 1996;Arribas and Palmqvist 1998;Palmqvist and Arribas 2001).Whether TSR can be ascribed to a hyaena denning site (like Venta Micena), before being eventually subject to multiple dispersion events, needs further analyses (e.g.study of the other size-groups recorded at the site, comparison with modern dens, mortality profiles), but the presence of juvenile remains of Pachycrocuta (Konidaris et al. 2015) and of several coprolites (preserving small bone fragments) at the site (Figure 1), both regarded as some of the criteria for the identification of hyaena dens (Kruuk 1972;Kuhn et al. 2010), are indications in support.
The findings of the present study have to be seen in light of certain limitations regarding the inclusion of a particular body size range in the analyses, as well as the excavation's collection protocol.
Nevertheless, we believe that the herein studied sample is a representative subset of the whole assemblage and provides secure data for the safe interpretation of the site as a Pachycrocuta-modified assemblage.Future work will aim to address some of these limitations by studying the remains of all body size groups, as well as the cranial elements.
Overall, this study not only provides taphonomic evidence for the interpretation of TSR but also offers insights into the palaeobiology, and particularly carcass consumption behaviour of Pachycrocuta, a fierce predator of the late Early Pleistocene terrestrial ecosystems of Europe.As such, it contributes to the investigations of carnivoran guild dynamics and of the prey-predator relationships during this period, as well as of the subsistence behaviour of the meat-eating hominins, who entered Eurasia at ~1.8 Ma, roughly synchronously with Pachycrocuta.

Figure 1 .
Figure 1.Horizontal (A) and vertical (B) distributions of a subset of the studied sample (2015-2019).Grey dots indicate non-modified bones; red dots mark carnivore modified bones (tooth marks, furrowing and/or notches); Orange squares denote carnivore coprolites.Dashed squares represent 1 × 1 m; the grey continuous line delimits the 2021 excavation area.

Figure 3 .
Figure 3. Percentages of fracture outline, edge, and angle (according to Villa and Mahieu 1991) for size group 4-5 long bones, and size group 4/5 shaft fragments.
Figure 9B shows that there is negative correlation between the abundance of epiphyses in Equus and their bone marrow content, described by the following equation: Log (abundance of epiphyses) = −0.831[Log (bone marrow content)] + 1.776, R 2 = 0.61

Figure 4 .
Figure 4. Ratios of double opposing:complete notches and double overlapping: complete notches of the TSR sample plotted along with data from the Maasai Mara hyaena den sample and experimental human-butchered bone assemblages (Domínguez-Rodrigo et al. 2007, 2015b).

Figure 5 .
Figure 5. Schematic drawing of taphotypes on a horse tibia, as defined by Domínguez-Rodrigo et al. (2015b) (top), and graphs displaying taphotype percentages separately for Equus and Bovini long limb bones (bottom).

Figure 7 .
Figure 7. Percentages of survival of bone portions (in colour) for Equus front (A) and hind (C), and Bovini front (B) and hind (D) limbs.Percentages of preservation were computed by dividing the raw abundances of bone portions of long limb bone elements by the minimum number of front/hind limbs of Equus and Bovini.Percentages of tooth-marked bone portions are also displayed for each bone portion, with the highest values for each element in red.Drawings modified from M. Coutureau, available at www.archeozoo.org.

Figure 9 .
Figure 9. Equus: Regression analysis between Log (abundance of epiphyses) and Log (bone marrow content) (A), Regression analysis between abundance of epiphyses and bone mineral density (B); Bovini: Regression analysis between abundance of epiphyses and bone mineral density (C).Data for mineral density from Lam et al. (1999), for marrow content from Outram and Rowley-Conwy (1998), and for fat weight from Brink (1997).

Figure 10 .
Figure 10.A, Regression analysis between %MAU of epiphyses and standard food utility index [S(FUI)] for Equus, data from Outram and Rowley-Conwy (1998); B, Regression analysis between %MAU of whole long limb bone elements and averaged standard food utility index [S(AVGFUI)], data from Emerson (1993).

Figure 11 .
Figure 11.Percentages of gnawed elements for Equus and Bovini (A).Raw values are also given for each column, and percentages of frequency of elements among gnawed specimens for Equus and Bovini (B).

Figure 13 .
Figure 13.Heatmap of tooth marks (pits, scores, punctures) of posterior, medial, anterior, and lateral views of humerus, radio-ulna, metacarpal, femur, tibia and metatarsal of Equus front (A1) and hind (B1), and Bovini front (A2) and hind (B2) limbs.Distal and proximal views are depicted only when tooth marks are present.Numbers on colour scale refer to raw number of tooth marks.

Table 1 .
Number of identified specimens (NISP), minimum number of elements (MNE), minimum animal units (MAU), and standardised minimum animal units (%MAU) values for the studied sample.

Table 2 .
Descriptive statistics for measurements (in mm) of scores, pits, punctures, and notches in the sample of Equus, Bovini, and total size group 4-5 sample.

Table 3 .
Frequency of tooth-marked bone portions, per element, for Equus and Bovini, presented in raw abundances (TM) and percentages (%).