Chimpanzee accumulative stone throwing

The study of the archaeological remains of fossil hominins must rely on reconstructions to elucidate the behaviour that may have resulted in particular stone tools and their accumulation. Comparatively, stone tool use among living primates has illuminated behaviours that are also amenable to archaeological examination, permitting direct observations of the behaviour leading to artefacts and their assemblages to be incorporated. Here, we describe newly discovered stone tool-use behaviour and stone accumulation sites in wild chimpanzees reminiscent of human cairns. In addition to data from 17 mid- to long-term chimpanzee research sites, we sampled a further 34 Pan troglodytes communities. We found four populations in West Africa where chimpanzees habitually bang and throw rocks against trees, or toss them into tree cavities, resulting in conspicuous stone accumulations at these sites. This represents the first record of repeated observations of individual chimpanzees exhibiting stone tool use for a purpose other than extractive foraging at what appear to be targeted trees. The ritualized behavioural display and collection of artefacts at particular locations observed in chimpanzee accumulative stone throwing may have implications for the inferences that can be drawn from archaeological stone assemblages and the origins of ritual sites.

stone accumulation sites in wild chimpanzees reminiscent of human cairns. In addition to data from 17 mid-to long-term chimpanzee research sites, we sampled a further 34 Pan troglodytes communities. We found four populations in West Africa where chimpanzees habitually bang and throw rocks against trees, or toss them into tree cavities, resulting in conspicuous stone accumulations at these sites. This represents the first record of repeated observations of individual chimpanzees exhibiting stone tool use for a purpose other than extractive foraging at what appear to be targeted trees. The ritualized behavioural display and collection of artefacts at particular locations observed in chimpanzee accumulative stone throwing may have implications for the inferences that can be drawn from archaeological stone assemblages and the origins of ritual sites.
In both contemporary and ancient human societies, stone piles are often used to mark natural cavities in the landscape for caching food, as well as paths and important places 1 , and can hold a more symbolic meaning for burials 2 , ceremonial counting 3 , and the establishment of shrines 4 . Through archaeology, analyses of stone assemblages have provided us with insight into the technological and cognitive abilities of ancestral hominins 5 . It is therefore notable that the use of stone tools has also been observed in wild populations of nonhuman primates, including chimpanzees, one of our closest living relatives 6,7 . Primate archaeology has therefore emerged as a new field of research where archaeological evidence from nonhuman primates can be compared to our own 8,9 . Any similarities may not only challenge, but may also illuminate the interpretations of stone accumulations in human prehistory.
Along with chimpanzees 7 , bearded capuchin monkeys in Brazil 10 and long-tailed macaques in Thailand 11 are also known to use stone hammers to crack open food encased by an outer shell. The use of stone tools in an extractive foraging context provides individuals with an immediately observable benefit, unlike other forms of stone tool use observed in nonhuman primates, such as stone throwing. Stone throwing was first described by Goodall who documented aimed throwing of sticks and rocks by male chimpanzees during agonistic displays 12 and was later described for other nonhuman primates, particularly captive Japanese macaques 13 , wild baboons 14 and capuchins 15 . Bearded capuchins have also been documented to bang stones in the wild, presumably to deter predators 16 , and females of this same species have been observed to throw stones during courtship contexts 15 . Nonetheless, previous examples of nonhuman primates throwing stones in the wild were limited to a single group 15 or anecdotal observations 12,14 . As such, stone throwing and banging has not been described as a customary behaviour of any nonhuman primate species. Stone handling, however, is a behavioural tradition found uniquely among Japanese macaques and has been observed in multiple groups 13,17 . Although stone handling is not a form of tool use 18 , the behaviour occurs frequently enough to result in recognizable use-wear patterns 19 . Of particular interest is the accumulation of stones at 'play stations' which has the potential to leave behind archaeological evidence of the behaviour 19 , similar to chimpanzee and capuchin nut-cracking sites 8 .
Chimpanzees exhibit the greatest variation in tool-use behaviours of any animal, second only to humans 7,18 . While all studied chimpanzee populations utilize leaves for the acquisition of food, some additionally use sticks, twigs, and even spears 7,18,20 . The majority of stone tool use has been observed in Western chimpanzee communities, where stone hammers and anvils are used to crack open nuts 20,21 , and stone cleavers are used to cut up large Treculia fruits 22 . The diversity and variation in the behavioural repertoire of chimpanzee populations across Africa strongly supports group-specific and socially-learned cultural traditions in our closest living relatives 7,20 . However, our current knowledge about the repertoire of chimpanzee tool use and inferred cultural transmission stems from a very limited number of long-term field sites, patchily distributed across the entire range of wild chimpanzees in Africa 20 . This limitation becomes particularly apparent when compared to the large number of human populations studied for similar questions on the evolution of culture 7 . The regular discoveries of additional tool-use behaviours in previously unstudied chimpanzee populations, or variants of behaviours already described, suggest that the true spectrum of natural chimpanzee behaviour is likely to be much broader than what is currently known 7 .
In an effort to overcome this limitation, we sampled a large number of previously unstudied and/or unhabituated chimpanzee groups across the species' range as part of the Pan African Programme: The Cultured Chimpanzee (henceforth "PanAf ") 23 , to study the influence of ecological conditions on the observed spatial variation in tool-use behaviours and inferred cultural patterns. At each PanAf research site, a broad spectrum of non-invasive sampling methods were applied for a 14-17 months period (see Supplementary Table 1), according to a standard protocol 24 . Here we present the first evidence of a previously undocumented, chimpanzee stone tool-use behaviour arising from the data collection of the PanAf.

Results
Prior to our study, there were reports and anecdotes of wild chimpanzees throwing and banging stones in Liberia, Guinea, and Guinea-Bissau (A.G. pers. obs.) in addition to occasional observations of habituated chimpanzees throwing rocks and other objects during male threat displays 12 . The first direct observations of stone throwing behaviour in association with the presence of accumulated stones at specific trees (henceforth "accumulative stone throwing") were recorded by camera traps on March 24, 2011, at the Sangaredi PanAf temporary research site (TRS) in Guinea ( Fig. 1; TRS #2). Following this observation, additional data collection procedures particular to this behaviour were incorporated into the PanAf protocol and administered to all TRSs across Africa to ensure comparable data across research sites (see Methods). Chimpanzee accumulative stone throwing behaviour. Thirty-one TRSs located within the Pan troglodytes range were sampled between 2011 and 2015 for a period of 14-17 months. An additional three TRSs were on-going and studied for less than 14 months, for a total of 34 (see Supplementary Table 1). At four TRSs: (Boé,  Guinea-Bissau; Sangaredi, Guinea; Mt. Nimba, Liberia and Comoé GEPRENAF, Côte d'Ivoire; Fig. 1) we found multiple hollow and/or buttressed trees exhibiting clear signs of wear with an accumulation of rocks at their base or inside the tree (Fig. 2). Using remote video camera traps, we subsequently filmed chimpanzees at each of these four TRSs approaching focal trees with a stone in their hand, or grabbing a stone from the base or from inside the tree's hollow cavities, and then proceeding to throw it (N = 64 total stone throwing events; Table 1). We observed three particular variants of the behaviour: the rock was thrown at the tree using one or both hands ('hurl'); hit repeatedly against the tree while the chimpanzee held it ('bang'); or thrown into the hollow tree or a hollow groove formed by large buttress roots ('toss'; Table 1; Supplementary Movies 1-7). The individuals observed in the camera trap footage were mainly adult males, but we also observed an adult female and a juvenile exhibiting the behaviour ( Supplementary Movies 1 and 6). Common to all accumulative stone throwing observations exhibited by adults (N = 63) was the pant hoot vocalization, in particular the introduction and/or build-up phase 25 , which occurred after or while the individual picked up and handled the rock (Fig. 3). The pant hoot is a characteristic feature of the ritualized agonistic displays of adult male chimpanzees, which typically also involves piloerection, bipedal stance, hand and feet drumming on buttress roots of trees and, in some populations, is preceded by leaf-clipping [25][26][27] . Unfortunately, audio was recorded for only 50 of the 64 events captured on camera traps, so we may underestimate the variation in vocal behaviour accompanying accumulative chimpanzee stone throwing. We further observed that rock handling and throwing was sometimes accompanied by the individual swaying back and forth while bipedal and piloerect, and even leaf-clipping (Supplementary Movie 4; Fig. 3), all behaviours associated with a typical chimpanzee display 27 . When the rock was thrown, this was often, but not always, accompanied by the climax phase of the pant hoot consisting of scream elements and drumming with the hands or feet on the tree 25,26 . In some cases we do not have footage of the full series of behaviours since camera trap videos are limited in length (60 seconds), and cameras were triggered at varying times for each accumulative stone throwing event captured.
The accumulative stone throwing behaviour was only observed in Western chimpanzees, Pan troglodytes verus (Fig. 1). Moreover, to our knowledge, the behaviour has not been observed at any of the 17 existing mid-to long-term chimpanzee research sites (LRS) across Africa 20 (Fig. 1). The wear observed on the trunks and buttress roots of trees targeted by the accumulative stone throwing indicated that all of the active sites had been in use for some time. Stone tools appeared to be regularly reused: in 57 of 64 stone throws filmed, the individual picked up a rock from the base of the tree, and once from inside the tree. We also observed the same individual at the same tree repeatedly engaging in accumulative stone throwing (Table 1), suggesting individuals frequently revisit sites.

Raw material accumulation and availability.
At all four TRSs where the behaviour was observed, stones had accrued around the base of the tree at each accumulative stone throwing site. In the few cases when rocks were found piled inside the hollows of trees, or nestled between the grooves of buttress roots, we counted between four and 37 stored rocks (Boé (Fig. 3), Comoé GEPRENAF, and Mt. Nimba). The average weight of the individual stones at Boé was 3.6 kg (range: 0.5-17 kg), at Mt. Nimba 2.06 kg (0.2-7.1 kg) and at Comoé GEPRENAF 0.98 kg (0.8-7 kg).
In order to determine if inter-site variation in the presence or absence of accumulative stone throwing could be explained by ecological factors, we examined the availability of stones and the availability of hollow trees. Using habitat plots and strip transects at 11 West African TRSs, including the four TRSs exhibiting accumulative stone throwing (Table 1), we calculated the density of hollow trees, or trees with hollow grooves between roots, and found no significant difference in hollow tree density between TRSs where accumulative stone throwing had been observed relative to TRSs where it had not been observed (Mann Whitney U test: N present = 3, N absent = 4, U = 4, P = 0.63). While hollow tree density was quite high at Boé, other TRSs also exhibited a high hollow tree density but did not reveal evidence for accumulative stone throwing (e.g., Bakoun and Djouroutou, Table 1). Similarly, a high availability of stones did not appear to be a necessary determinant for the behaviour. The densities of loose rocks calculated along strip transects at each TRS (Table 1), indicated that although the accumulative stone throwing behaviour occurred in habitats with relatively high stone availability, this factor did not guarantee the occurrence of the behaviour. Furthermore, we found no significant difference in rock density between West African TRSs (Table 1), where the behaviour was observed compared with West African TRSs where the behaviour was not observed (N present = 4, N absent = 7, U = 5, P = 0.11). Moreover, no observations of accumulative stone throwing were made at Bakoun, Sobeya or Kayan, although these TRSs also exhibited relatively high rock densities.

Discussion
The accumulative stone throwing behaviour described here seems to be rare in chimpanzees as it has not been observed during decades of research at long-term field sites across Africa and has not been observed at other PanAf sites ( Fig. 1; Supplementary Table 1). This suggests that these initial observations may underestimate the behavioural complexity characteristic of accumulative stone throwing in chimpanzees. The habitual incorporation of stone tools into the ritualized display of chimpanzees is a novel discovery, and according to our data, may be a socially-learned cultural tradition found in limited populations of West African chimpanzees. It remains to be tested whether fine-scale genetic differences may influence the observed distribution of stone tool-use behaviours in this subspecies. However, a genetic explanation seems unlikely to account for the patchy distribution of this behaviour within West Africa (see Fig. 1). We expect that the concentrated accumulation of stone tools at specific tree locations will facilitate further study of this behaviour, for example through determining site fidelity using excavations and other archaeological methods. If sites are found to be long-lived, it would represent another type of stone tool-use behaviour in chimpanzees, in addition to nut-cracking, that leaves behind an archaeological record 8,9 . Our observations raise numerous questions about the potential interpretations of this behaviour. There are at least two contrasting hypotheses that need to be evaluated with long-term studies to elucidate the broader context in which this behaviour is shown. First, the accumulative stone throwing behaviour could be a modification of the male chimpanzee display, since the action bears a close resemblance to hand and feet drumming 26 , a ritualized behaviour found in all known chimpanzee populations. The incorporation of stone tools into this display may serve to enhance the sound propagation properties in more open savannah-woodland landscapes. In this case there should be a clear functional advantage for sound produced by stone throwing compared to hand and feet drumming. For example, fundamental frequencies produced by throwing stones at particular trees may be higher and would thus travel further in a more open environment 28 . Thus, one may hypothesize that stone throwing, specifically the 'bang' variant of the behaviour, initially emerged as a variant of hand and feet drumming. The continuous practice of stone throwing and subsequent collection of artefacts at specific trees then led to modification of stone tool availability at these selected trees. This in turn may have influenced chimpanzee behaviour with a preference for site re-visitation, similar to a local stimulus enhancement mechanism proposed for stone-handling behaviour in Japanese macaques 17 . Over time chimpanzee accumulative stone throwing would therefore have become more and more independent of its original trigger (hand and feet drumming) and, as some of our recordings suggest, is now occasionally practiced independently of it (e.g. Supplementary Movies 2 and 4). As such, the accumulation of stones at particular trees may have originated as a by-product of modified display behaviours occurring at fixed locations.
Alternatively, chimpanzee accumulative stone throwing could have emerged from a motivation other than the enhancement of ritualized male displays. In this case one would not necessarily expect a major functional advantage of sound propagation compared to hand and feet drumming. If this were true, these rock accumulation sites may need to be considered in a more symbolic context 1,29 . Analyses of the behavioural responses of conspecifics, present or nearby, to stone throwing versus hand and feet drumming may provide crucial insight into the meaning of this behaviour. Additionally, the spatial patterning of nesting and home range use in relation to stone accumulation sites may reveal whether they are centrally, peripherally or randomly located in the territory of a chimpanzee community. In fact, marking territorial boundaries and pathways with cairns has been an important practice of many historical human societies 1 . For example, stone accumulation shrines at 'sacred' trees are well described for indigenous West African peoples 4 . Superficially, these cairns appear very similar to what has been described here for chimpanzee accumulative stone throwing sites, thus it would be interesting to explore whether there are any parallels between chimpanzee accumulative stone throwing and human cairn building, especially in regions of West Africa where the local environment is similar.
Incidentally, chimpanzee accumulative stone throwing and the subsequent aggregation of tools at particular trees shares two important features with human ritual practices: the strong association to a particular location or site with a collection of artefacts over time, and ritualized behaviour patterns 30,31 . Although there is no overarching, agreed-upon definition of ritual 30 , similarities between ritualized animal behaviour and the repeated, stereotyped behaviours commonly observed during human rituals have already been proposed by anthropologists 31 and ethologists 32 as having a common origin, as well as sharing similar neurological pathways 30,33,34 . From a phylogenetic perspective, we suggest that further research on naturally occurring stone tool-use traditions in nonhuman primates, especially those whose occurrence appears to be tied to specific locations with no immediate functional benefit to the individual, namely accumulative stone throwing in chimpanzees and stone-handling behaviour in Japanese macaques, could serve to enlighten our understanding of the origins of ritual sites. At the very least, stone accumulation sites produced by extant nonhuman primates have the potential to challenge and refine our interpretations of hominin archaeological assemblages in similar contexts 5

Methods
We sampled 29 lesser-known chimpanzee communities as part of the Pan African Programme: The Cultured Chimpanzee (PanAf) 23 to complement the published data already available from 17 mid-to long-term chimpanzee research sites. PanAf data collection was conducted for 14-17 months at 26 TRSs with on-going data collection at an additional three TRSs (ranging from 4-10 months duration; Supplementary Table 1). A further five PanAf TRSs were carried out at established long-term research sites (LRS) on unhabituated chimpanzee communities, also for a 14-17 months duration. At each PanAf TRS a broad spectrum of non-invasive sampling methods was applied following a uniform sampling protocol 24 . One TRS in West Africa, Grebo in Liberia, was in the final months of completion when the discovery of accumulative stone throwing behaviour was made; therefore, stone and hollow tree density were not available for this site.
Stone banging was first heard in Boé, Guinea-Bissau in 2010 by an independent team of field workers. The first confirmed observation of the accumulative stone throwing behaviour occurred by chance at Sangaredi in Guinea in March 2011 after discovering unusual markings on a hollow tree. When investigating the area further, we found conspicuous accumulations of rocks next to trees with clear indications of wear. The subsequent placement of remote video camera traps (Bushnell Trophy Cameras) at these locations revealed the chimpanzees' accumulative stone throwing behaviour. Following this observation we systematically collected data on this behaviour across all 34 PanAf TRSs. For this we used both recce methodology and opportunistic sampling to try to locate conspicuous accumulation of stones in close proximity to trees. At each TRS we sampled 15-44 km of recces or line transects (Table 1), and transects were walked three times during a one-year period, during which any signs of stone throwing and piling behaviour were sought. In case such locations were found, we then placed one or more video camera traps at a distance of a few metres from the target trees and stone piles. Video cameras were strapped to trees at a height of 1 m and were operational 24 hours per day. When rocks were found inside hollow trees we counted and weighed them using a spring scale (35 kg Travel Blue Spring scale). We extracted the following information from each video that contained observations of chimpanzee accumulative stone throwing: time, date, number of individuals present, age and sex class for each individual, and type of behaviour exhibited. The accumulative stone throwing behaviour was classified according to the type of rock handling observed: 'hurl' , 'bang' or 'toss' , and we also noted the occurrence of other behaviours such as pant hooting and elements of the ritualized chimpanzee display (Fig. 3).
The density of hollow trees was calculated either by observations of these trees while conducting strip transects, or from 20 m by 20 m habitat plots that were surveyed at each TRS (208-246 habitat plots per TRS). Habitat plots were centered along transects at 100 m intervals. Stone availability was estimated using strip transect methodology, whereby observers count the number of loose rocks found within 1 m of either side of the line transect or recce being walked. A detailed protocol for all PanAf methods is freely available online on our website http:// panafrican.eva.mpg.de.
Ethics statement. The data collection methods for this study were strictly non-invasive and were approved by the Ethical Board of the Max Planck Society. As such, the study was conducted in accordance with Germany's laws and the rules and regulations governing animal research in the European Union. In every African country where field data were collected, all relevant permissions were first obtained from the country's governmental institutions before starting data collection.