Neanderthal hunting strategies inferred from mortality profiles within the Abric Romaní sequence

Ungulate mortality profiles are commonly used to study Neanderthal subsistence strategies. To assess the hunting strategies used by Neanderthals, we studied the ages at death of the cervids and equids found in levels E, H, I, Ja, Jb, K, L and M of the Abric Romaní sequence. These levels date between 43.2 ± 1.1 ka BP (14C AMS) and 54.5 ± 1.7 ka BP (U-series). The degree of eruption and development of the teeth and their wear stages were used to determine the ages of these animals at death, and mortality profiles were constructed using these data. The equids display prime dominated profiles in all of the analyzed levels, whereas the cervids display variable profiles. These results suggest that the Neanderthals of Abric Romaní employed both selective and non-selective hunting strategies. The selective strategy focused on the hunting of prime adults and generated prime dominated profiles. On the other hand, non-selective strategies, involved the consumption of animals of variable ages, resulting in catastrophic profiles. It is likely that in the selective hunting events were conducted using selective ambushes in which it was possible to select specific prey animals. On the other hand, encounter hunting or non-selective ambush hunting may have also been used at times, based on the abundances of prey animals and encounter rates. Specific hunting strategies would have been developed accordance with the taxa and the age of the individual to be hunted. The hunting groups most likely employed cooperative hunting techniques, especially in the capture of large animals. Thus, it is not possible to uniquely associate a single mortality profile with the predation tactics of Neanderthals at Abric Romaní.


Abstract
Ungulate mortality profiles are commonly used to study Neanderthal subsistence strategies. To assess the hunting strategies used by Neanderthals, we studied the ages at death of the cervids and equids found in levels E, H, I, Ja, Jb, K, L and M of the Abric Romaní sequence. These levels date between 43.2 ± 1.1 ka BP ( 14 C AMS) and 54.5 ± 1.7 ka BP (U-series). The degree of eruption and development of the teeth and their wear stages were used to determine the ages of these animals at death, and mortality profiles were constructed using these data. The equids display prime dominated profiles in all of the analyzed levels, whereas the cervids display variable profiles. These results suggest that the Neanderthals of Abric Romaní employed both selective and non-selective hunting strategies. The selective strategy focused on the hunting of prime adults and generated prime dominated profiles. On the other hand, non-selective strategies, involved the consumption of animals of variable ages, resulting in catastrophic profiles. It is likely that in the selective hunting events were conducted using selective ambushes in which it was possible to select specific prey animals. On the other hand, encounter hunting or non-selective ambush hunting may have also been used at times, based on the abundances of prey animals and encounter rates. Specific hunting strategies would have been developed accordance with the taxa and the age of the individual to be hunted. The hunting groups most likely employed cooperative hunting techniques, especially in the capture of large animals. Thus, it is not possible to uniquely associate a single mortality profile with the predation tactics of Neanderthals at Abric Romaní.

Introduction
Mortality profiles are an instrument traditionally used to infer the origins of fossil assemblages in archaeology and paleontology [1][2][3][4][5][6]. Kurten [4] and Voorhies [7] were pioneers in the study of paleontological assemblages. They applied principles developed by ecologists to fossil records (e.g. [8,9]), and established the ages at death of animals in order to construct life tables, which they used to infer the population dynamics represented by the fossil record. Mortality profiles were later employed at North American Paleoindian archaeological sites [10][11][12]. These studies focused on bison kill sites where mass predation events occurred in order to establish hunting seasonality [10][11][12].
Age at death research has conventionally focused on animal dentition. Non-destructive methods have been developed for these assessments, such as the measurement of the crown heights of teeth [3,13,14] and analyses of occlusal surface wear [15][16][17]. Both methods are based on comparing the assessed tooth wear stage with reference collections that include animals whose ages at death are known. Destructive analysis methods have also been used. Cementochronology is based on counting the layers of cyclically deposited cement, which alternate between relatively thin and relatively thick and reflect a growth periodicity that generally corresponds to an annual cycle [18][19][20].
Two types of plots are typically used in archaeological studies with this focus: 1) Mortality profiles show the relative or absolute frequency of faunal remains within each age range [3,5].
2) Survivorship curves show the surviving individuals in each interval, starting with the total number in the assemblage [3,5]. Age at death estimates are generally compiled in histograms and linear graphs (e.g. [21,22]). Greenfield [23] introduced the use of triangular graphs in the analysis of mortality profiles, although the works of Stiner [5,6] popularized the use of this type of diagram. The distribution of the three age groups (juvenile, prime adult and old adult) can be graphically represented to show the different mortality profiles. This makes it possible to compare the mortality profiles obtained from archaeological assemblages with profiles based on current observations. Two types of mortality profiles are characteristic of the populations of large mammals, and are commonly referred to as catastrophic (or living-structure) profiles and attritional profiles [2,5,7,24]. Catastrophic mortality profiles reflect the ranges of ages that can be observed in living groups of animals, which are present in direct relation to their abundance in the ecosystem. They are produced by ambush predators (e.g. lions, leopards and tigers), by communal hunting events carried out by hominins, and by massive mortality events caused by drowning, famines, etc [2,5,7,25]. Attritional mortality profiles are characterized by a high frequency of younger and older individuals, which are the weakest and most easily hunted animals, and such profiles are generally produced by cursorial hunters (e.g. cheetahs, wolves, spotted hyenas and lycaons) [2].
Using mortality profiles derived from Middle and Upper Paleolithic sites in Italy, Stiner [5] identified a progressive increase in the preference for hunting prime adult animals, an age range not exploited by other predators. This type of prey selection is ecologically complementary to cursorial predators and, to a lesser extent, ambush predators, and it allows different predators to exploit the same prey population while minimizing competition [2,5,6,26]. Stiner [5] defined this hunting behavior of hominin groups as the human predatory niche. In her earliest work, Stiner [5,6] indicated that the mortality profiles of the early Mousterian (pre-45,000 BP) are characterized by a non-selective pattern in the ages of prey. This pattern is in striking contrast to the selective pattern of the late Mousterian after 40,000 BP [5, 6], in which prime adults are more abundant. In her later work, she rejected differences between the periods mentioned above because she had observed mortality profiles that included the totality of the catastrophic and prime-dominated areas, which, on average, indicate a slight bias toward up of 20 m of well-stratified travertine sediments. Rock fragmentation and alluvial and biochemical sedimentary processes have generated beds of consolidated stones, gravels, calcarenites and calcilutites interspersed with very fine archaeological levels. Uranium-series and radiocarbon dates place the Abric Romaní deposit between 70 and 40 kyr ( Table 1). The sedimentation rate is estimated to have been approximately 0.6 m/kyr [67]. Excluding level A, all of the archaeological levels correspond to the Middle Paleolithic. The Abric Romaní sequence ranges from Marine Isotope Stage (MIS) 4 (sterile levels) to the first half of MIS 3 (archaeological levels). It includes the Dansgaard-Oeschger (D-O) cycles, which extend from 19 to 12, as well as the Heinrich Stadium (HS) from 6 to 5 [68] ( Table 1).
The site was discovered in 1909 by Amador Romaní and excavated during different periods throughout the 20th century. The current intervention, in which the full extent of the rock shelter (c. 300 m 2 ) has been excavated, began in 1983. Numerous well-preserved combustion structures have been identified [69][70][71] in addition to wood remains, including negatives and carbonized positives [72][73][74][75].
Faunal remains are very abundant in all levels of Abric Romaní. In total, 38,228 faunal remains belonging to 13 different taxa have been recovered. Among the ungulates, cervids (Cervus elaphus) and equids (Equus ferus/Equus hydruntinus) are the most abundant animals, according to the number of identified specimens (NISP), the minimum number of elements (MNE) and the minimum number of individuals (MNI). The remains of aurochs (Bos primigenius) are also common in the lower levels (I, J, K, L, M, O and P), and the chamois (Rupicapra pyrenaica) is common in the upper levels (A, B, D, E, F, G, H and J). The remains of rhinoceros (Stephanorhinos cf. hemitoechus) have been documented in levels H, J and M, although they are scarce. A proboscid femur was recovered from level E (Table 1). Zooarchaeological studies of each level indicate that the faunal record is the result of Neanderthal activity, which was characterized by primary access to animal carcasses and complete exploitation of their resources [86][87][88][89]. By means of the density of the faunal remains, as well as taphonomic analysis and anatomical refitting, it has been possible to establish areas where specific activities were performed. These analyses, along with the characterization of hearths, have resulted in the identification of sleeping areas, cleaning zones and animal processing areas in some of the levels [71,[86][87][88][89][90][91][92][93]. In all the levels, all of the activities associated with animal butchering have been documented, indicating the complete exploitation of animal resources. In addition, numerous thermoalterations have been documented. Evidence of carnivore activity is very scarce and is thought to have resulted from scavenging of the remains left by human groups [71,73,79,81,82,86,[88][89][90][91]94]. The faunal assemblage reflects the highly variable transport of animal carcasses. This transport ranged from complete animals to just a few elements. In general, anatomical elements that were high in unsaturated fat were preferentially transported to the rock shelter [94].
The occupation types that produced the archaeological record can be separated into two groups, long-term and short-term (and/or non-residential) occupation events [71,73,81,82,[90][91][92][93]95] (Table 1). In both of these models, all of the hearths were reused, especially during the long-term occupation events, and these structures were therefore preserved. Therefore, the materials present within the Abric Romaní correspond to the superposition of different occupational events [71,92].

Materials and methods Materials
To determine the mortality profiles of the equids and cervids found in the Abric Romaní sequence, 486 teeth from levels E, H, I, Ja, Jb, K, L and M were examined. Of these teeth, 259 were derived from equids, and 227 from cervids. The proportions of maxillary and mandible teeth, 126:133 for the equids and 101:126 for the cervids, are very similar for both groups of animals. A large fraction of the dental remains was found isolated within the site. In some cases, fragments of maxillae and mandibles were recovered with teeth anchored in their alveoli (62 dental remains in total). The number of analyzed teeth found in each level and their integrity (in situ or isolated teeth) is reported in Table 2. To establish dental age, various methods were combined depending on the type of tooth (superior or inferior and anterior or jugal) and the taxon being considered. All materials used in this study are deposited at the Institut Català

Determination of dental age
In the case of the equids, we looked at the eruption of deciduous teeth and their replacement by permanent teeth [14]. We used the nomenclature for the anatomical descriptions and for assessing the orientation of the jugal teeth established by Levine [14]. The dental ages of permanent superior and inferior premolars and molars with completely worn occlusal surfaces were calculated based on their crown heights. To estimate age, the parameters of a third-order polynomial regression that relates crown height to age were used [13]. The relevant equation is were a k is the regression coefficient [13,14,96]. The data used in the calculation of the polynomial regression are those provided by Fernandez and Legendre [13]. In keeping with the recommendations of Fernandez et al. [22], we estimated the individual average, minimum and maximum ages as a function of the minimum error associated with each prediction equation (E) [13,22,96]. Crown heights were measured from the cementum-enamel junction to the highest point of the occlusal surface along the labial face (Fig 1).
To estimate the dental ages of the cervids, the eruption, replacement and wear of the jugal teeth were analyzed [15]. The anatomical nomenclature of the teeth, the designation of the worn facets and the orientations of the teeth were based on the criteria given by Brown and Chapman [15] (Fig 2). The wear stage codes established by Mariezkurrena [97] and Azorit [98] were also used. These codes were modified for Cervus elaphus from Payne's [16] original work with sheep and goats. To estimate the degree of wear of the maxillary dentition, an approximation was made by analogy to the attrition described by Brown and Chapman [15] for mandibular dentition, assuming that the degree of wear for maxillae is similar [99]. As a complement, the quadratic crown height method (QCHM) was applied to the mandibular and maxillary teeth (dP 4 , M 1 , M 2 and M 3 / dP 4 , M 1 , M 2 and M 3 ) to estimate dental age. The crown height was measured on the labial faces of the lower teeth and on the lingual faces of the upper teeth, on the anterior lobe between the occlusal surface and the cement-enamel junction [3,99,100] (Fig 2). The prediction equations applied were: AGE ¼ AGESesð CHÀ CH 0 CH 0 Þ 2 for the fourth deciduous premolar (dP 4 / 4 ) and AGE ¼ AGEpel À AGEe ð Þð CHÀ CH 0 CH 0 Þ 2 þ AGEe for the permanent molars (M 1 / 1 , M 2 / 2 and M 3 / 3 ). Here, CH is the crown height of a tooth in mm; Ch 0 is the crown height of an unworn tooth in mm; AGEes is the age in months, in which dP 4 / 4 is replaced by P 4 / 4 ; AGEe is the eruption age in months of M 1 / 1 , M 2 / 2 and M 3 / 3 ; AGEpel is the potential ecological longevity (PEL) in months.

Estimation of the minimum number of individuals and the construction of age groups and size-weight categories
The MNE [101] for mandibles and maxillae was calculated by means of the construction of dental series, taking the side the tooth originated from (left and right) into account and estimating dental age by means of a combination of the methods described above. The sum of isolated teeth and those found anchored in alveoli with known ages provided a fairly complete dental series. Calculating the MNE can facilitate the estimation of MNI [102,103] for each of the taxa and levels studied. Individuals were assigned to age groups based on Bunn and Pickering's [61] method, which uses the potential ecological longevity (PEL), and is a modification of the model described by Stiner [5]. These authors suggest dividing the PEL into five groups: young juveniles and subadult juveniles (<20% PEL), early prime adults (20-50% PEL), late prime adults (50-75% PEL) and old adults (75%>PEL). This PEL sequence is intended to correctly establish the vulnerability of the ungulates to predators, which is not clearly predicted using the age classes established by Stiner [5]. The early and late prime adult individuals are physically robust, difficult to hunt or even dangerous, whereas the young juveniles, subadult juveniles and old adults are physically weak, more vulnerable and easily hunted. These authors restructure the limits between age groups through the progression of wear stages. Thus, the event that separates young juveniles from subadult juveniles is the eruption of M 1 and M 2 . They note that, unless extraordinary preservation conditions exist, young juveniles will be excluded from the analyses. The boundary between late prime adults and old adults is the loss of the mesial infundibulum of M1 in class 3 bovids (113-340 kg) and the loss of the mesial and distal infundibulum of M2 in class 1 and 2 bovids (4.5-22.5 kg and 22.5-113 kg), which occurs at a PEL of approximately 75%. However, the method proposed by Bunn and Pickering [61] focused on bovids. That proposal has been slightly modified in our work to adapt the old adult PEL to the taxa that are most frequently encountered at the Abric Romaní.
In this work, the PEL used for horses is 300 months (25 years) [35,104] and the PEL used for cervids is 192 months (16 years) [100].
The following age groups were established for the equids: 1) Young juveniles (12 months = 4% PEL) are identified by the presence of complete deciduous dentition, which erupts between 30-40 days after birth, until the eruption of M 1 . 2) Subadult juveniles (60 months = 20% PEL) are characterized by wear on M1 and limited by the total loss of deciduous dentition and the eruption and initial wear of P4 and M3 (Levine, 1983). Finally, individuals were assigned to the last three groups, 3) early prime adults (144 months = 50% PEL), 4) late prime adult (225 months = 75% PEL), and 5) old adults (more than 225 months = 75%! PEL) by measuring the crown heights of the teeth and assigning them to age groups as a function of the percentage of the PEL [14,22,96] (Table 3).
The cervids were divided into the following groups. 1) Young juveniles (5 months = 2.6% PEL) are identified by the presence of all deciduous dentition, which is present from birth, to the eruption and wear of M 1 . 2) Subadult juveniles (30 months = 20% PEL) are identified by the first wear on the mesial peak of M 1 to the start of wear on the distal peak of M 2 and the end of wear on dP 2 , dP 3 and dP 4 . 3) Early prime adults (78 months = 50% PEL) are identified by  [15,97,98]. At this time, the fecundity and body fat of females decreases enormously, and the males are not able to defend their harems due to a decline in their physical strength [105][106].
The two most common weight categories at Abric Romaní are large (equids) and mediumsized (cervids). However, the variation in the weight and size of animals over the course of their lives should not be ignored; taxa and weight categories are not equivalent. These variations in animal weight and size are the results of physiological changes that occur during the growth phase, which ends with sexual maturity and implies major changes in the social behavior of animals [35,[106][107][108][109].
In the case of equids, the weight of a two-year-old is equivalent to 75% of that of an adult individual, which is ultimately achieved at five years of age [108], the time at which sexual maturity is also reached by both males and females [35,104]. Therefore, the dividing line between subadult juveniles and prime adults is considered to correspond to the dividing line between medium and large individuals. The social organization of equids is highly hierarchical. A harem is made up of one stallion and several females with their foals. At around two years of age, the male foals abandon their maternal groups to join groups of single males. Unlike males, subadult females are not forced to leave their maternal group, but they often join other harems after two years [108]. Although this behavior highly influences the composition of social groups, individuals older than two years of age have not yet reached the weight or size of an adult individual. Since the prey choices made by predators, including hominins, are primarily based on feed return rates and the difficulty of capture [63,110,111], the fact that male juvenile horses leave their maternal groups and join groups of single males does not necessarily change the way they are considered by predators. Therefore, two-year-old individuals are still considered subadult juvenile individuals.
The size of Cervus elaphus is especially variable within single populations and between populations found in different environments. Individuals usually continue to grow until they have reached seven years of age. In Cervus elaphus, the boundary between subadult juveniles and prime adults (30 months) coincides with two important ethological events: the abandonment of the family group by young males, and the sexual maturity of females [104]. In Cervus elaphus sexual maturity also corresponds to the time at which they reach their full body weight [112]. Therefore, the dividing line between subadult juveniles and prime adults is considered to correspond to the dividing line between small and medium-sized individuals. As indicated above, significant fluctuations in the size of this species have been noted within the fossil record [113,114]. At present, it has been observed that Cervus elaphus in poor habitats are smaller. For example, adult males in northern France weight between 120 and 250 kg (150 kg on average) and females weight between 67 and 100 kg (80 kg on average) [115]. On the other hand, on the Iberian Peninsula, males can reach a body weight of approximately 160 kg, whereas females can reach a body weight of approximately 100 kg [106].

Mortality analyses
Once the age profiles had been obtained, we analyzed the profiles using triangular diagrams [5]. The individuals were assigned to three age groups (young, prime and old), and the proportion of each class was plotted on a triangular graph. To this end, the five groups were sorted into the three age categories mentioned above; Table 4 shows the correspondence between the age groups. Within the triangular graphs, the upper corner represents 100% old adults, the lower right corner indicates 100% prime adults, and the lower left corner indicates 100% young individuals. The areas representing catastrophic and attritional age structures, which occur to the left of the area indicating the dominance of prime adult and to the right of the area indicating the dominance of juveniles, respectively, are indicated on the graph. When a sample is plotted within a triangular graph, its position within one of these five zones is assumed to indicate a mortality profile [2,5,21,99].

Results
The MNI was determined to be 97 across the analyzed levels. Of these individuals, 47 are equids and 50 are cervids (Table 5).

Equids
Based on the MNI, the level with the highest number of individuals identified is level Ja with 15 individuals, followed by levels K, I, Jb, M, E, H and L, which contain 8, 7, 6, 4, 3, 2 and 2 identified individuals (Table 5). Tables 6 and 7 show the MNE of mandibles and maxillae. Figs 3-6 show the MNI of the equids by level (S1 Table). We were able to measure the crown heights of 33 mandible teeth and 54 maxillary teeth belonging to a total of 43 individuals, which allowed us to calculate the ages of these animals (Tables 6 and 7).
A subadult juvenile individual, an early prime adult, and a late prime adult were identified in level E based on the determined age at death. Within level H, two early prime adult individuals were identified. Within level I, a subadult juvenile individual and six early prime adults were identified. Within level Ja, four subadult juvenile individuals, four early prime adults, six late prime adults and one old adult individual were identified. Within level Jb, a subadult juvenile individual, three early prime adults, and two late prime adults were identified. Within level K five early prime adults, two late prime adults and one old individual were identified. In level L two late prime adult individuals were identified. Within level M, a one subadult juvenile, two early prime adults and one late prime adult were identified. In total, 9 juveniles, 37 prime adults and 2 old adult individuals were identified (Table 5). The equids show a mortality profile that is dominated by prime adults within all levels of the sequence (Fig 7). Table 5 shows the total NISP, MNE and MNI calculated for the cervids found in each of the levels studied. Tables 8 and 9 show the MNE of mandibles and maxillae in each level.  have been identified (S1 Table). Level K and level M have the most individuals, whereas level Jb contains of the fewest. It was possible to define the wear pattern of the occlusal surfaces of 63 mandible teeth belonging to 34 individuals. The identified wear patterns are indicated in Tables 8 and 9. Based on the determined ages at death a subadult juvenile individual and two early prime adults were identified within level E. Within level H, three late prime adults were identified. Within Level I, four young juveniles, a subadult juvenile, an early prime adult and an old adult individual were identified. Within level Ja, three young juveniles, two subadult juveniles, four early prime adults and one late prime adult were identified. Within level Jb, only one early prime adult has been identified. Within level K, two young juveniles, two subadult juveniles, four early prime adults, one late prime adult and two old adult individuals were identified. Within level L, two young juveniles, three early prime adults, one late prime adult and one old adult individual were identified. Within level M, a young juvenile, two subadult juveniles, two early prime adults, two late prime adults, and one old adult individual were identified. In total, 20 juveniles, 25 prime adults and 5 old adult individuals were identified (Table 5).

Cervids
For the cervids, the mortality profiles vary by level, and five different models have been identified (Fig 11): 1) a profile that is located at the intersection between the catastrophic mortality and the prime-dominated profiles in level E; 2) a profile that is dominated by prime adults within levels H and Jb; 3) a mortality profile that is dominated by juveniles in level I; 4) a profile that is positioned at the intersection between the attritional and catastrophic profiles at level Ja; 5) and a catastrophic mortality profile obtained from levels K, L and M (Fig 11).

Discussion
To date, the deposits uncovered at Abric Romaní have yielded a sequence of 15 levels formed during the interstadial MIS 3. An exclusively anthropogenic origin for the lithic and faunal remains has been inferred for all of these levels. The mortality profiles of the main taxa indicate the existence of marked differences between the ages of the equids and cervids. The equid Neanderthal hunting strategies at Abric Romaní Neanderthal hunting strategies at Abric Romaní Neanderthal hunting strategies at Abric Romaní accumulation is dominated by prime adults, whereas the cervid accumulation presents greater variability among the different levels (Figs 7, 11 and 12). None of the mortality profiles identified at Abric Romaní are unusual in the context of the European Middle Paleolithic record [5, 28,37,43,53,54,116,117]. Mortality profiles that differ according to the taxa hunted have been found at the Manie, Madonna and Lazaret sites [53]. At Manie and Madonna, aurochs present prime-dominated profiles, whereas red deer present a catastrophic profile; however, that difference only exists in one level at each site. At Lazaret, the profiles of deer and ibex change in each of the site's five levels, and neither species has a regular profile [53]. At Abric Romaní, two mortality profiles have been observed throughout the studied sequence. In addition, the atrophic character of the accumulation suggests that the mortality profiles could be considered a valid proxy that can be used to infer the hunting strategies used by the Neanderthals of Abric Romaní.
However, inferences made using mortality profiles should be taken with caution as they may not accurately reflect prey selections made by hominins [118]. Bone accumulations are affected by two key processes, specifically the possible enhanced destruction of young individual teeth and the differential transport of heads according to the weight of individual prey animals. These processes may affect the representation of juvenile individuals in a few different  Table 6.
https://doi.org/10.1371/journal.pone.0186970.g003 ways: a) because their teeth tend to disappear easily; and b) because their lower weight means that they are more likely to have been transported whole to the site [118].
Considering the differential preservation of individual as a function of age, the scarcity of juvenile equids (17% of the sample) seems to be due to differential destruction processes. However, the high abundance of juvenile cervids (40% of the sample), whose teeth have a lower density than those of equids, indicates that the bias toward adult horses is likely not the product of differential preservation processes, because it should also have altered the deer sample.
Regarding the possible differential transport of heads, Marean [118] indicated that three interrelated factors should be considered: 1) the weight of the animal; 2) the size of the hominin group; 3) the distance between the kill site and the reference site.
Depending on the weight of the animal, many ethnoarchaeological studies indicate that hunter-gatherers faced with equal conditions usually transport more complete small animals than large animals [119][120][121]. In addition, the variation in the weight of an animal species throughout its life must also be considered, as it affects the decision to transport the heads of that species or not, and therefore also the composition of the mortality profiles [118]. In equids, we documented a predominance of large adult individuals (MNI = 39) over mediumsized young individuals (MNI = 8) (Table 5), while among the cervids we found no great differences between the representation of small (MNI = 20) and medium-sized individuals (MNI = 30) (Table 5). Also, the anatomical representation indicates that the cranial elements of the large animals were transported in equal proportion, or even more, to the rock shelter than cranial elements of medium-sized animals (Table 10) (Fig 13). In levels Ja, Jb, K and M the value of %MAU for skulls is 100% and in levels H and L it is equal or superior to 50%. However, for the cervids the value is 100% only in level M, and in level L the value is even  Table 6. under 30% (Fig 13). MNE and %MAU data for level E are not available. Thus, the weight of the animals did not appear to be a major determining factor in the transport of their heads. However, is important to consider that this circumstance could have affected the transport events [122], because the superposition of diverse transportation strategies has been identified at Abric Romaní, due to multiple conditioning factors, such as the sizes of different animals and the food utility of the transported elements [94].
An ethnoarchaeological principle used in the study of mortality profiles indicates that, under equal conditions, within an average settling system of small groups, hunter-gatherers show greater selectivity in making transport decisions, and often abandon the heads of large animals [119,120,123]. On the other hand, in an average settlement system of large groups, the bias of the transport of these animals may be reduced [118][119][120][121]. The spatial patterns documented within the Abric Romaní correspond to bivouacs around isolated external hearths and to camps with many external hearths connected by reassembling, and resting and sleeping areas [78,92,124]. All the levels are the result of an unknown number of occupation events that occurring over hundreds to thousands of years. The levels originating from short-term or non-residential occupation events have been defined as the result of the activity of groups of hunter-gatherers immersed within a regional foraging mobility model, whereas the long-term  Table 7.
https://doi.org/10.1371/journal.pone.0186970.g005 events have been interpreted as a result of the activity of medium or large groups [70,81,82,84,91]. Levels H, I, J, K, L, and N have been defined as short-term occupation events associated with a highly mobile Neanderthal group. Analysis of the distribution and nature of the combustion structures associated with the sleeping and resting areas of level N, suggest that it could correspond to a Neanderthal group of between 8 and 12 individuals. Monahan [125] indicates that the Hadza require at least 10 to 12 porters to transport size 3 animals (113-340 kg) or greater. At Abric Romaní, the occasional transport of complete carcasses of mediumsized and large animals has been identified in all of the levels, regardless of the occupation model. This indicates that hunting parties may have been sufficiently numerous to transport a complete or almost complete large animal [94], at least occasionally.
In keeping with that suggested by Marean [118], differences in the transport of heads may be present within the sequence of Abric Romaní, depending on whether small or large groups occupied the rock shelter.
Level I is the result of short-term or non-residential occupations events, in which the mobility of the group was directly related to the exploitation of hunting resources [91]. The mortality profile of cervids shows a greater number of young individuals (MNI = 5) than adults or old individuals (MNI = 1/1). The predominance of young animals may be due to selective transport, as suggested by Marean [118], in which the transport of small individuals takes precedence over the transport of the heads of adult deer. However, at this level, we see that the transport of equids is not governed by this principle. The equids are represented by one young individual and six prime adult individuals; thus, the large animals are more abundant than    medium-sized animals. Therefore, the age representation of both species does not appear to be due to the selective transport of heads, according to the weight of the animal. On the other hand, level J has been identified as resulting from of long-term occupation by medium or large groups composed of more than 12 individuals, and may be related to the   Table 8. https://doi.org/10.1371/journal.pone.0186970.g008 aggregation of groups of hunter-gatherers [93]. We found a balance in the representation of small and medium-sized animals among the cervids in this level. Meanwhile, the equids are dominated by large prime adults, which is common to all the levels. Thus, in spite of differences in the settlement model, age selection does not seem to be conditioned by animal size. Finally, another factor that would condition the transport of the heads is the distance between the kill/butchering site and the reference site. The Abric Romaní is located inside an ecotone that is formed by the Anoia River, the plains that surround it, and the mountains that delimit the gorge. These features give the site strategic importance, and from it, the Neanderthals had immediate access to different biotopes and species [68,86,93,126]. Biotic resources, both faunal and vegetation-based, were locally exploited [86,89]. The selection of prey mainly focused on obtaining two types of animals, cervids and equids, which may indicate that the axis of mobility was oriented toward the open areas and the forests on the mountainsides that enclose the river valley [68,86,89,93,126]. In any case, assuming long distances from the kill/ butchering site to the reference site, would favor a bias toward large animals. However, as  Table 8.
https://doi.org/10.1371/journal.pone.0186970.g009 mentioned, the small and medium-sized cervids are represented approximately equally, whereas there is a dominance of large animals among the equids. Therefore, either the transport distance associated with the equids was shorter, or the transport of their heads did not constitute a problem.
Although the archaeological mortality profile does not necessarily correspond to the original mortality profile [118], it seems that the mortality profiles documented at Abric Romaní closely reflect the prey selection and hunting strategies and behaviors of the Neanderthals who occupied the shelter.
Prime-dominated profiles like those observed at Abric Romani have also been identified for different species and at numerous archaeological sites, such as equids at Cuesta de la Bajada, aurochs at Manie and Madonna, and cervids at Combe-Grenal, Lazaret E, Pech-de-l'Aze I level 7, Breuil and TD10.1 (Fig 14) [5, 27, 28, 52-54, 59, 60].  Table 9. https://doi.org/10.1371/journal.pone.0186970.g010 Neanderthal hunting strategies at Abric Romaní Considering the predominance of a particular age group among the equids, we argue that individual animals may have been selectively hunted. However, Bunn and Gurtov [127] have indicated that, when a group of hunter-gatherers has sufficiently advanced technology to hunt individuals of any age group, prime adults will be captured more frequently because they are the most numerous individuals in prey populations, and not because the hunters are selective.
If true, this principle should also govern the age selection of cervids. However, we observed that this premise is not fulfilled (Fig 6). Other studies have argued that the preference for prime adults reflects the objective of maximizing the return rates of predation events [63,128]. Thus, we assume that the ability to take game from any age group would result in a preference for larger, more profitable prime adults, rather than young animals [63]. The presence of a stable predation model for equids and a variable model for cervids throughout the sequence seems to be related to the intentional selection of prime adult equids. Neanderthal hunting strategies at Abric Romaní The identification of primaries access, the immediate and recurrent access to large carcasses (over 300 kg), has been interpreted as the result of cooperative hunting [27,28,58,59,125]. In addition, this group cooperation has already been demonstrated at Abric Romaní through an analysis of the transport strategies used for animal carcasses [94]. The mortality profiles of the equids, which are dominated by large animals, support cooperative and selective hunting as the social organization of the hunting groups.
Studies of the micro-wear of the teeth of equids from Abric Romaní indicate that they were hunted for short periods of time within the same season, although not synchronously [86,90]. Thus, we can reject the hypotheses of mass and multiple predations, and assume that simple predation events were most commonly carried out by the Neanderthal groups.  Several types of hunting techniques have been proposed that produce a prime-dominated profile. Prime-dominated profiles have been associated with selective ambush hunting in different ethnoarchaeological and archaeological studies where primary access to animal carcasses has been documented [1,5,24,28,61,129]. Bunn and Pickering [129] have proposed ambush hunting for the FLK Zinj assemblage (Bed I of Olduvai), where the mortality profile of the great ungulates is dominated by prime adults.

Equids Cervids Equids Cervids Equids Cervids Equids Cervids Equids Cervids Equids Cervids Equids Cervids Equids Cervids
Binford [24] observed that the Nunamiut generated prime-dominated profiles of caribou on certain occasions. They established hunting camps in narrow passages along the caribou migration routes, and captured the animals using bows and arrows or rifles. Thus, the hunting of prime adults was made possible by the planned use of space (i.e. the interception of prey in specific places) and by cooperative work during the acquisition of carcasses [24]. Therefore, we can argue that the hunting of horses by the Neanderthals of the Abric Romaní was accomplished by selective and cooperative hunting tactics, possibly by ambush hunting, although this conjecture cannot be demonstrated empirically.
The mortality profile of the cervids is characterized by great internal variability. Although it is widely assumed that Neanderthal activities generated prime dominated profiles, the mortality profiles of the cervids in the Abric Romaní sequence are not unusual within the Middle Paleolithic record (Fig 15). The mortality profile of the cervids shows both selective (levels E, H, I and Jb) and non-selective (levels Ja, K, L and M) patterns.
As indicated by Stiner [27], the mortality patterns generated by the hunting activities of human groups usually range from non-selective with an average towards the prime adults to heavily biased toward the prime adults. These patterns appear to be reflected in the cervid profile of level E, which is located between the intersection of the catastrophic and prime-dominated profiles (Fig 15). A prime-dominated profile has been identified at levels H and Jb (Fig  15). Although the cervids are only represented by three and one individual in these levels, respectively, which makes it difficult to defend a clear tendency in the selection of prey, primedominated profiles point to an age-selective hunting strategy [5].
Level I yielded a mortality profile that is dominated by young individuals (Fig 14). In this scenario, the Neanderthals would have deliberately hunted younger animals, which have the lowest economic return rates [130]. The preference for the young individuals of Cervus elaphus is also found in other Mousterian sites, such as Cova Beneito levels XII-X (Spain) [131] and Gabasa (Spain) [52]. In the latter case, once an increase in the presence of juvenile individuals by the action of other carnivores was ruled out, Steele [52] suggested that the greater presence of juvenile individuals is related to Neanderthal activity. The same behavior could account for the accumulation of cervids in level I, which includes four individuals with PEL values of less than 2.7% (less than 5 months of age). Measurement of the crown heights of these individuals indicates that two had not even passed the first month of age (Tables 8 and 9). This particular accumulation suggests a specialized seasonal hunt, perhaps during the same occupational event, indicating that hunting events were more common during cervids birthing season. Varin [132] has indicated that due to the presence of newborns, female groups display a pattern of low mobility with very specific habits in which they follow established paths at precise times of day. This behavior makes these animals extremely easy to track [54], so hunters may have waited for female groups to take their regular route and hunted them using ambushes or traps. The hunting of young individuals has been documented among current groups of  [53]; Md_Aur = Madonna aurochs [53]; Ma_Aur, Ma_Roe = Manie aurochs and roe deer, respectively [53]; Br = Breuil [53]; CG 50, 52 = Combe Grenal [52]; PA-7 = Pech-de-l'Aze I [54]; CB = Cuesta de la Bajada [60]. hunter-gatherers. During the dick-dick and steenbock birthing seasons the! Kung San of the Kalahari capture young individuals by chasing them on foot or throwing clubs [133].
Binford [24] observed hunting events in which the Nunamiut (Tulekana and Kakinya) exclusively hunt young reindeer in order to obtain soft leather for clothing. Lithic use-wear analyses at Abric Romaní show that worked skins existed within the sequence, with work on fresh leather being more common [134]. In addition, lithic functionality studies in level Ja relate denticulate and notch features to the hardening of hides [81]. In the Abric Romaní sequence, although young individuals have been identified in almost all of the studied levels, they do not reach 71% of the total, as in level I. Therefore, in this level, the hunting of cervids seems to have been specifically intended to obtain this prey of low economic return, possibly to obtain their hides. Gb e/f = Gabasa levels [52]; Ft_Rd = Fate red deer [27]; Lz A, B, C, D, E, 25 = Lazaret levels [53]; Md = Madonna [53]; M2-4, M6 = Moscerini levels [53]; Kv = Kevara [27]; Ma = Manie [53]; Br = Breuil [53]; CG 50, 52 = Combe Grenal levels [52]; PA 4, 6, 7 = Pech-de-l'Aze levels [54]; Hdz = Hadza assemblages; K = Kua [127]. The mortality profile of level Ja is located at the intersection between the attritional and catastrophic profiles, indicating equal proportions of juvenile and prime adults, but no old adult individuals (Fig 15). The occupations took place in different events throughout the autumn and early winter, generally during a whole season, as indicated by studies of the seasonality of the ungulates [93]. Thus, the mortality profile ranged from attritional to catastrophic, indicating that Neanderthals captured individuals ranging from the weakest to the strongest over long periods of time, thus reflecting non-selective hunting.
Catastrophic mortality profiles have been identified at levels K, L and M (Fig 15). Catastrophic mortality profiles have also been observed for Cervus elaphus at the Mousterian sites of Pech-de-l'Aze (France) in levels 4 and 6 [54], Manie (Italy), Madonna (Italy), Fate (Italy), Lazaret (France) [53], Kebara (Israel), Sant'Agostino 3 and Guattari G4-5 (Italy) [5,27]. Traditionally, catastrophic mortality profiles have been considered to illustrate mass communal hunting events or repeated episodes of non-selective hunting [2,3,5,25,127]. The faunal assemblages produced by mass communal hunting events are characterized by large accumulations made up of dozens of individuals of the same species, almost monospecific [25]. The systematic transport of high utility elements to the base camps has been documented in these assemblages [25]. The characteristic mortality profile of these sites is catastrophic, as determined by the age and sex of the individuals represented, with a strong trend toward seasonal or synchronous death [25]. In levels K, L and M more than three taxa are present (Table 1) with between 1 and 11 individuals of each species. The transport of the anatomical elements reveals a close correlation between the anatomical elements present and their richness in unsaturated marrow [94]. The cervids found in levels K and L were hunted over an estimated period of three months; those found in level M appear to have been hunted over an estimated period of about six months [86]. These analyses indicate that the cervids found in levels K, L and M were not hunted synchronously. Therefore, the catastrophic mortality profile of the cervids found in levels K, L, and M did not result from mass communal hunting events. As an alternative explanation for catastrophic profiles, Rendu [54] proposed that the lack of clear seasonality may be the result of a palimpsest, in which different seasonal events accumulate and finally generate a catastrophic mortality profile. However, if this explanation were to apply in the case of Abric Romaní, the equids should have catastrophic profiles as well as the cervids, since the sets are the result of multiple occupational events with a broad seasonal distribution that can occupy an entire season, as in levels Ja, K, L or M [86,91,93]. Thus, the differences between the cervids and equids in levels Ja, K, L and M tend to support the hypothesis of the non-selective hunting of cervids, suggesting that different tactics were used to hunt different taxa and were maintained over time, and they did not result from equifinal processes.
Thus, the mortality profile generated by the Neanderthals for the cervids indicates a less selective hunting strategy regarding the age of this taxon, and is more closely related to its abundance in the environment and therefore to the rates at which hunting groups would have encountered this animal [135,136]. In terms of the organization of hunting groups [65], once the possibility of communal hunting has been dismissed, it is impossible to determine whether the capture of the animals was accomplished by cooperative hunting parties or individuals. However, because cooperation has been established as necessary to hunt equids, it should not be ruled out for deer hunting. Unlike selective ambush hunting which would generate profiles dominated by prime adults [5,129], Bunn and Gurtov [127] point out that non-selective ambush hunting would generate catastrophic mortality profiles. Ethnoarchaeological observations indicate that other non-selective hunting strategies also generate catastrophic mortality profiles [120,121,127,137]. Middle Mississippi hunters during the Archaic period (1000-1550 BC) produced a catastrophic mortality profile of white-tailed deer in six sites through non-selective hunting by stalking [137]. The hunter groups of the Hadza and Kua generate catastrophic mortality profiles in their camps [127] (Fig 13).
Encounter hunting is the style most commonly practiced by the Hadza, although, during the dry season, they usually ambush hunt at night near water sources. In both cases, they use poisoned arrows to kill their prey [120,121]. The Kua mainly encounter hunt, and kill with bows and arrows, although they are also effective as hunters when stalking or using persistence running and traps [133]. The non-selective mortality profiles of levels K, L and M appear to be the product of non-selective stalking. However, it cannot be completely ruled out that this profile may have been produced by non-selective ambush hunting, as with the profiles generated by other non-human ambush hunters, such as leopards or lions [5].
The results of our analysis indicate that the differences in the mortality profiles within the Abric Romaní sequence are the result of various hunting strategies: 1) Selective hunting strategies were employed for equids in all levels and for cervids in levels E, H, I and Jb. 2) Non-selective hunting strategies were employed for cervids in levels Ja, K, L and M. Prime adults comprise the strongest and healthiest animals in a population. Moreover, the large size of equids implies that they pose a substantial risk to hunters. In non-human predators, this risk is reduced through cooperation, as seen in lions, wolves, spotted hyenas and lycaons. Cooperation makes it possible to hunt prey that considerably exceeds the weight of any individual predator [27]. A trend towards capturing prime and young individuals (juveniles = 2, prime adults = 5, old adult individuals = 3) has also been observed in the levels in which the other large ungulate found in this assemblage, Bos primigenius, is present (Ja, L, K and M). The two young individuals are subadult juveniles between 3 and 4 years of age and are very close to reaching the weight of an adult animal. Therefore, cooperative organization of hunting groups would have been an effective way to reduce risk in the capture of these large ungulates [27,28,58,59,138]. Cervid hunting may have also been cooperative, as with the equids, although it is possible that these animals were brought down by individual hunters, especially the young individuals [24,54,65,135]. Depending on the number of prey animals captured during each hunting episode, equid and deer remains may have accumulated over the course of various occupational events in which Neanderthals hunted. These episodes may have occurred over several months within the same season, during the same occupation events or at different times of the year, but not synchronously, so it seems that simple predation prevails over sequential predation. In the hunting of prime adults (equids and deer), it seems that the selective ambush tactic may have been the most commonly used [1,5,24,28,128]. Young cervids can be captured in multiple ways-by traps, projectiles or pursuit [24,54,138]-but would not have constituted a challenge to hunters. Finally, the catastrophic profiles noted for cervids most likely resulted from encounter hunting by stalking or non-selective ambush hunting [127,138].
The Neanderthals of the Abric Romaní developed specific hunting strategies based on two characteristics of their prey: the taxa and the age of the individual to be hunted. Each hunting strategy gave rise to specific mortality profiles.

Conclusions
The mortality profiles identified in the faunal assemblages of the Abric Romaní are useful for studying the hunting strategies of the Neanderthals who occupied the shelter. The ages at death of the equids and cervids reflect marked differences between the two taxa. The equids are always dominated by prime adults, whereas the cervids show great diversity in the sequence. This result indicates that the observed differences in mortality profiles resulted from the use of both selective and non-selective hunting strategies.
Selective hunting strategies were employed for the equids at all levels and for the cervids founds in levels E, H, I and Jb. Non-selective hunting strategies were only employed for the cervids of levels Ja, K, L and M. The capture of equids and cervids was probably carried out by various individuals cooperatively, although individual hunting cannot be ruled out in the case of cervids. Depending on the number of prey animals caught by the hunting groups, it seems that simple predation was the most common tactic. The prime adult may have been captured by means of selective ambush hunting. The selective capture of very young individuals may have been accomplished in multiple ways, without presenting a challenge for the hunters. The tactics that generated the catastrophic profiles may have included encounter hunting by stalking or non-selective ambush hunting. Therefore, different hunting strategies have been identified within the same Neanderthal group, some of which are repeated along the sequence, as seen in the case of the equids, and others that vary according to specific behaviors, as observed in the case of the cervids.
Supporting information S1 Table. Number of equids and cervids mandibles and maxillae at Abric Romaní, indicating level, archaeological reference, MNE, size, dental series, crown height of teeth and code of wear stage. (XLSX)