Matching physical properties of food and tools in three populations of wild capuchin monkeys

Robust capuchin monkeys (Sapajus) are known for accessing mechanically challenging food. Although presenting morphological adaptations to do so, several populations go beyond the body limitations, using tools, mainly stone tools, to expand their food range. Those populations are diverse, some using stones more widely than others. We know stone tool size correlates with the target's resistance within some populations, but we have no detailed comparisons between populations so far. This study described and compared general environmental data, food’s physical properties, and stone tools features on three populations of bearded capuchin monkeys (Sapajus libidinosus), including a new site. The differences we observed regarding stone tool use between the new site (CVNP) and the previously studied ones could be partially explained by ecological factors, such as the raw material and resource availability. However, other differences appear to be more related to behavioral traditions, such as the processing of Hymenaea at CVNP, where the monkeys use bigger stones than other populations to process the same kind of food, which present similar physical properties between sites. Possible cultural differences need to be compared within a larger number of areas to better understand capuchin monkey behavioral variability.


Introduction
Although only recently methods had been developed to measure the food's physical properties in the eld 1 , robust capuchin monkeys (genus Sapajus) are widely and long known for accessing mechanically challenging food [2][3][4][5] . The genus Sapajus exhibits adaptation in the skull and postcranial skeleton that should produce increased masticatory or postcranial forces when feeding 6,7 . However, the several degrees of morphological variance between species is not straightforward correlated with species diet difference. For example, Sapajus apella has a more robust skull and skeleton than S. libidinosus, but the latter has a diet with a high maximum and average toughness value 4,5 . This counterintuitive fact may represent behavioral adaptation in opposition to morphological adaptation since S. libidinosus' use of stone tools to exploit the most challenging foods in their diet [8][9][10][11][12][13][14] . The same may happen with yellowbreasted capuchins (S. xanthosternos), which also have a more gracile skull morphology than S. apella 7 and can access hard food resources using tools 15 .
Sapajus is one of the few extant primate genera (together with Pan, Homo, and Macaca) to customarily use stone tools to access encapsulated food 16,17 . However, the use of stone tools is not the same in all populations, with variations in the material and dimensions of tools used and the food resources explored. It is believed that the stone tools are chosen chie y depending on the physical property of the target resource, and that was shown for populations of S. libidinosus, where heavier tools are used for harder food, as the stone tools used to process large palm nuts [18][19][20] . Additionally, chemical properties can also be a factor to in uence the choice of a tool, as in the case of fresh cashew nut cracking at Serra da Capivara, where monkeys choose larger tools to, apparently, protect themselves from the caustic liquid from the softer fresh cashew nuts 21 , different from other population where the same resource is explored without stone tools, using a rubbing technique 22 . Therefore information on ecology, physical environment, and behavior, including dietary ecology and food material properties, are needed to understand the evolutionary driving forces behind anatomical and cultural changes related to food processing.
The S. libidinosus population from Serra da Capivara National Park (SCaNP) is known to present, to date, the most diverse use of stone tools. Besides using stone tools to crack open encapsulated plant resources, as in other savanna populations, they also use stone tools to aid digging for food, and for other percussive behavior, such as stone on stone to obtain powder to ingest and anoint 20,23 . More recently, the population of S. libidinosus of Serra das Confusões National Park (SCoNP), 100km from SCaNP, was described to use stone tools to process encased food, exploring different resources, or on different frequencies, from other populations 8 . Another well-studied population of S. libidinosus is from Fazenda Boa Vista (FBV), where the monkeys appear to be specialized in cracking open hard resistant palm nuts 12,24 , although they also occasionally process less resistant food 25 . Other populations of S. libidinosus have been described, directly or indirectly, to use stone tools to process encapsulated, hard-tobreak food resources 9,10 . In no population, except for SCaNP, have capuchins been observed, so far, to habitually use stone tools for anything other than cracking encased food.
The relation between the food resource characteristics and stone tools has been studied to some extent.
Although the studies do not measure the resistance of the food, they argue that stone tools are most frequently used to process high resistant fruits, such as palm nuts 9,19,[26][27][28] , and there is a correlation between hammerstone weight and the apparent higher resistance of the resources 8,19,20 . However, there is almost no information about actual values for resistance of the resources and no comparison between capuchin populations concerning the physical properties of the accessed food regarding the features of the stone tools used in each area.
Accessing the physical properties of the food can shed light on the tool choice by capuchins to process those foods and better understand the relation between resistance of food resources and tool features.
The effectiveness of the forces produced when the tool contacts the target depends on the characteristics of the food surface. These characteristics are external (e.g., size and shape) and internal mechanical properties (e.g., elastic modulus and toughness). Once the food is hit, the resistance to fragmentation is de ned by the internal mechanical properties. The size, shape, and physical properties of the tool interact with the same properties of the food, and we can infer that the food breakdown is optimized to the animals' needs, saving energy. To measure the internal mechanical properties, tests record the force applied to the object and the deformation that it produces. The deformation can be registered as a forcedisplacement graphic. Hardness is the resistance to the initial loading force; in other words, the bigger the hardness value, the higher the force necessary to initiate a fracture. Elastic modulus corresponds to the material's rigidity and can also be used to estimate food resistance to the beginning of a fracture; thus, less elastic material (lower values of the elastic modulus) would be more rigid 29 .
In this study, we will describe and compare lithic raw material availability, food resources physical properties, and features of the stone tools used (hammer and anvil) of three bearded capuchin monkeys (S. libidinosus) populations ( Fig. 1) that use stone tools to process food resources, to test the following hypotheses: 1. Capuchin monkeys in each population select stones of adequate size to process the resources present in their living area Prediction 1a -Hammerstone dimension will be correlated to resource hardness and elasticity, heavier stones to harder and less elastic resources

Results
Stone tools dimensions of the three study sites are presented in Table 1. Stone tools at CVNP were the heaviest and largest across sites. The stone tools were composed mainly of quartzite (46.8%), followed by sandstone (25.2%) and limonite (13.8%). However, each site had different patterns of stone tool material ( Table 2).
The lithic raw material characteristics available for each area are presented in Table 3. The primary food resource found at the anvils during the mapping at CVNP was Attalea palm nut (91%); Hymenaea remains were found in only 3% of the mapped anvils. At SCoNP, the main resources found in the anvils were Buchenavia fruits (38%), Manihot (31%), Hymenaea (19%), and Attalea maripa nuts (11%); the remaining 1% was not identi ed 8 .
The physical properties test results (Indentation Hardness and Reduced Elastic Modulus) on the resources sampled in each area (Fig. 2), plus the average of stone tools recorded to process each resource in each area, are presented in Table 4.  30 . We compared them at the genus level in this work.

Comparison within populations
The resource physical properties and stone tools used to process those items in each population are summarized in Fig. 3
Regarding anvils, we found no differences in the length related to the food resources processed with stone tools (Kruskal-Wallis, 2 =1.516, df=3, p=0.679).

Comparison between populations
To analyze the differences between populations, we compared the resources present in more than one site and the stone tools used to process those items.
Attalea palm nuts, Buchenavia seeds, and the associated stone tools were not quantitatively evaluated because they are present in only one of the study sites.
Fresh cashew nuts sampled in SCaNP and SCoNP were similar in elasticity (Mann-Whitney, Z=55, p=0.845) but had signi cantly different values for hardness (Mann-Whitney, Z=25, p=0.025). As for dry cashew, we found signi cant differences in both physical properties (Mann-Whitney, Z=23, p=0.017; Mann-Whitney, Z=23, p=0.018), with SCaNP nuts being harder and less elastic. We could not compare stone tools for cashews because SCoNP monkeys had not been observed to explore this resource, with or without tools 8,31 .
The stones used to process those resources were heavier at SCoNP, the difference was not signi cant, but there was a trend (Mann-Whitney, Z=387, p=0.085).
The Hymeanae pods were the only resource analyzed in all three sites. We found no differences in physical properties between the jatobá fruits (Hardness, Kruskal-Wallis, 2

Discussion
This work compared wild capuchin monkeys from three areas, describing the stone tool use in a new S. libidinosus population and comparing the resources processed within and between sites. We also describe the physical properties of the resources and the stone tools used to process those foods and compare the data between the three sites and the literature.
The new site described, CVNP, has a typical tropical savannah (Cerrado) capuchin stone tool use pattern. The capuchins mostly use stone tools to crack open a high resistant palm nut (Attalea), similar to those processed in other savannah environments 9,15,18 . However, the weight of the tools and the physical properties of the food accessed in this site were bigger than the other two populations analyzed (and also from the literature). Stone tools in CVNP had an average weight of 1672g. The heaviest stone tool registered in our mapping was 5700g (Table 1). In latter camera-trap surveillance (data not presented here, see supplementary video les for examples), we recorded monkeys effectively using a 4650g stone tool (Fig. 5), an astonishing weight to be lifted by monkeys with a body mass average of 3500g for adult males and 2100g for adult females 32 . At the other two sites, the resources explored had lower hardness values, and the stone tools used to pound food resources were lighter, as expected based on other capuchin populations choosing stones according to the resistance of the resource processed 8,12,19 .
CVNP has a high density (215 stones/ha) of quartzite pebbles (the best raw material for stone tools) ( Table 3), making it much more likely for the monkeys to encounter and use it than other stones available. Furthermore, the quartzite pebbles in this site are on average larger than the stones available at the other sites, maybe allowing CVNP´s monkeys to more often use heavy stones, close to the limit of their lift capacities. Thus, this environmental difference could partially explain the use of larger tools for the CVNP population.
The high availability of quartzite in SCaNP (the higher among the compared sites, 322 stones/ha) could explain the more diversity of stone tool use observed in that area than any other site 8,20,23,33 . However, we probably do not have a complete knowledge of the repertoire of stone tool use at CVNP to compare those sites thoroughly and better understand the mechanism underlying tradition on the use of tools.
One of the resources explored with stone tools in CVNP is jatobá pods (Hymenaea sp.), although much less frequently than the palm nuts, at least in the study areas we mapped. The stone tools were, as expected, lighter than the ones used to process the harder palm nuts, although not statistically signi cant. This result may be related to the small sample of Hymenaea tools.
Jatobá pods were the only resource present in all three populations compared here. Our results showed that the physical properties of the pods were not signi cantly different across sites, meaning the challenge to open the pods would be similar in all three areas. However, the weight of the stone tools used in each population was different. CVNP monkeys used much larger stones to process the pods than the monkeys from the other two populations. One explanation could be the availability of the potential material to be used as tool. As mentioned before, CVNP has the larger/heavier raw stone material availability (Table 3). An alternative explanation would be related to the other resources explored. At the CVNP area, the monkeys explore, at the same nut-cracking sites, and with higher frequency, a much harder food (Attalea nuts), which could bias the tools selection to larger stones because of those more resistant palm nuts.
To better understand the tool use repertoire at CVNP, we still need to directly observe the monkeys or perform experimental tests to check the existence (or not) of tool use observed in other populations (e.g., probe tools, digging stones). The use of camera-traps recording nut-cracking sites (analysis ongoing) will also allow the observation of the stone tool use, its e ciency, and variation.
The rst prediction of hypothesis 1 was not supported in any of the analyzed populations. Heavier stone tools were not correlated to process harder food. In the case of CVNP, we had a small sample for stone tools used for Hymenaea, and both resources analyzed were found close to each other, so the stone selection at that area could have been biased to the process of the harder Attalea nuts. In the case of SCoNP, stone tools' weights did not differ between Hymenaea and Manihot seeds. Two things could explain this difference. First, in SCoNP, the lithic raw material is scarcer than in CVNP and SCaNP, restricting the selection of tools. Second, although small (Avg length, 16.42mm), Manihot seed showed surprisingly higher hardness values, perhaps requiring a similar stone tool to the ones used to open the larger Hymenaea. Considering only the hardness value, we would expect that the monkeys would use heavier stones to access Manihot. However, in SCaNP, the stones used to crack Manihot were lighter than Hymenaea. One possible explanation is that although Hymenaea is much larger, scaling hardness with size does not consider the combination of structural and material compensation (the result of the heterogeneity and anisotropy of the food resource). Taking into account the structure of the food resources, could be helpful in future studies to clarify the stone dimension and food physical properties.
The second prediction for H1 was not supported also. The anvil length did not vary by resource size. Our explanation is that an adequate anvil can be successfully used to process most resources. So, not being a limiting factor that needs to be different for most resources, e.g., the same anvil used to process an Attalea nut could be potentially used for Hymenaea pods, particularly when both species occur near each other. Even when a small target, such as manihot, is being processed, the choice of an anvil could be driven by the larger resource in the area (rule of thumb to choose larger anvils), as this larger anvil can be used for any resource and there are no energetic costs, as it is not transported. A factor that should be analyzed in the future is the inclination of the anvil or the presence of pits, as some resources are rounder and more prone to roll from a tilted anvil, meaning this factor could be more critical for anvil selection depending on the type of resource to be placed on the anvil.
Our results show a more complex association between food physical properties and tool selection than previous studies 12,19,20 . The data presented here also suggest that variables can interact, such as food size and hardness, and re ect on tool selection. In this way, our results have reshaped our understanding of tool selection considering physical properties. Our results suggest a hierarchical approach to the tool choice considering food hardness, size, and stones available to choose in the environment. Future work testing the relationships between physical properties and tool selection in other taxa will improve the generalizability of these results.
Hypothesis 2 was supported for Hymenaea pods and fresh cashews but not for dry cashews and manihot seeds.
Hymenaea pods and fresh cashews did not differ in the physical properties between sites, even though the Hymenaea in each place are from different species. That allows an interesting comparison between the use of stone tools to open this resource, as it is processed with stones of different weight even being similar regarding physical properties. The same is true, in part, for the green cashew nuts, although in this case, the chemical characteristics could also be a factor of in uence 21 .
We found differences in hardness and elasticity between the study sites for manihot and dry cashews. The difference between physical properties for the same species in different sites can be related to the time lying on the ground. Cashew nuts can lay on the oor for a long time (days or even months) before being picked by the monkeys (T.F. personal observation), and our sampling re ects that, as we collected a mix of dry nuts from the ground and also directly from the tree. Therefore, we would expect the loss of water content to turn the tissues less elastic and harder (Gerolamo et al., submitted). The in uence of the effects of the environment on the physical properties could also explain the difference found for manihot seed; even if similar to the naked eye, we may have sampled seeds in different stages of desiccation.
Robust capuchin monkeys are known to explore mechanically tough food, however, populations that do not use stone tools must rely only on their bodies to do so. That presents a series of limitations.
Nevertheless, capuchins have features that facilitate the production of large muscle and bite forces without compromising gape and so can exploit mechanically challenging foods at relatively large jaw gapes. But Attalea and Hymenaea are larger/bigger than the jaw gape of the monkeys and could only be opened by hand or using tools, and that appears to be the case. The monkeys at the analyzed populations (and other previously studied) have shown that the use of stone tools provides access to hard-encased food that would not be accessible otherwise. Even for smaller food targets, e.g., Manihot seeds and cashew nuts, the e ciency of using tools to access those targets could be higher, especially if the population already has the tool use behavior in their repertoire because of the more challenging targets that obligatory need stone tools to be accessed, as appears to be the case of the populations in this study.
The differences we observed regarding stone tool use between the new site (CVNP) and the previously studied ones could be partially explained by ecological factors, such as the availability of lithic material and food resources. When processing similar food, such as Hymenaea, each population appears to be limited by the stones available, re ecting on the stone choice. Even the presence of other food resources can bias the tool choice, as the Attalea nuts presence appears to do at CVNP. However, some differences are more subtle. Even though cashew nuts were available and softer than cashew nuts present in the SCaNP population, SCoNP monkeys do not eat cashew nuts, with or without stone tools 31 . Since the absence of consumption of this food resource in SCoNP is neither because of availability or impossibility of access, one could argue that this difference is a cultural behavior. Cashew nuts possess mechanical and chemical defense mechanisms 21 that need to be dealt with, and that behavior could be socially learned.
To better understand the variation on resource exploitation, including tool use, by capuchin monkeys, we still need more information from new sites to have other examples to correlate with the possible ecological and cultural factors.

Study sites
Serra da Capivara National Park (SCaNP, -8.833239, -42.552377) and Serra das Confusões National Park (SCoNP, -9.213317, -43.498371) are located in the south of Piauí state (Fig. 1), Brazil. Both parks have Caatinga (thornbush savannah) as the predominant biome and semiarid as the prevailing climate. Still, the west area of SCoNP is an ecotone with the biome Cerrado (tropical savannah). The vegetation is xerophytic, especially at higher elevations, but there are patches of deciduous forest in the humid valleys surrounded by high cliffs.
SCaNP has an area of 130,000 ha, with an average temperature of 28°C (range 10-47°C) and a mean annual rainfall of 689 mm. This region is composed of Serra Grande formation, which has sandstones and conglomerates, and Canindé Group, which has laminated shale, sandstones, tempestites, mudstones, cobbles, and siltstones (for more details, see 34,35 ).
SCoNP has 823,435 ha of area, in which the average temperature is 28°C (range 12-45°C), and the mean annual rainfall is 650 mm. There are also more humid areas (Cerrado) in the east and north of SCoNP and some intermittent rivers in the valleys. This park presents three geological formations: Cabeça, which has hard sandstones; Pimenteiras, having red shale, sandstones, and siltstones; and Serra Grande, which has conglomeratic sandstone and quartz pebbles (for more details, see 8 ).
Chapada dos Veadeiros National Park (CVNP, -14.093696, -47.373306) was created in 1961 in the northeast of Goiás State (Fig. 1), in a central region of Brazil. This site has an area of 240,611 ha characterized by Cerrado biome and a tropical savanna climate. It has annual rainfall ranging from 1,500 to 1,750mm, most of it occurring during the wet season (November to March), and average temperatures ranging from 20° to 27°C 30 . The region encompasses waterfalls and three rivers descending from high plateaus of altitudes up to 1,676m, erosional scraps, and intra-plateau depressions 36 . This high altitude and water availability enable a mosaic of vegetation, including more open areas as grassland and open scrubland, wet areas as gallery forests, until more dense areas with taller trees, as forest and wooded savanna 30 . In terms of geomorphology, CVNP is in the Brazilian Central Plateau and is formed by Araí Group, that includes the basement granite gneiss and contains quartzites, conglomerates, calcareouspelitic rocks, sandstones, basalts, and siltstones; and also by Paranoá Group, which is above the Araí group and contains metasediments, including quartzites, metasiltstones, limestones and dolostones 36 . Tool sites and environmental sampling The stone tool use site mapping at CVNP was done following the same method of previous work 8  A "processing site" was characterized by the following items: (1) "anvil", a at surface used as a substrate for the processed encased food, (2) "hammer" on top or beside the anvil (within 1m), and (3) remains of the processed encased food atop or adjacent (within 30 cm) to the anvil 37 .
When a processing site was identi ed, we recorded the following variables: GPS coordinates, anvil material and size (maximum length and width), stone tool material, weight and size (length, width, and thicknesses), and processed encased food. We used digital scales (to nearest 0.1 g), calipers (to nearest 0.1 mm), and measuring tapes (to nearest mm). We visually identi ed processed encased foods in each site and collected samples from nearby trees for botanical identi cation. We collected the remaining of processed resources at anvils for the physical properties test (see below).
Samplings of processing sites and description of S. libidinosus tool use at ScaNP and SCoNP were done in previous studies 8,20 . Still, we revisited both locations in October/2019 to collect food samples for physical properties testing and plot samples (see below). We collected Manihot, Annacardium, and Hymenaea samples in the same areas of the previous works. At SCaNP, we collected additional behavioral data on tool use throughout two study periods with the Pedra Furada (PF) group (2018 and 2019). TF followed this group for a total of 242 hours, recording all occurrences of tool use and measuring the tools used with the same protocol as above after the monkeys left. The PF group was also studied and described in previous studies in this area 20,21,23,38 . The data on the stone tools is available in Supplementary Table 1.
To characterize the availability of raw lithic material, we did plot samples at the three sites, in the living area of groups Pedra Furada (SCaNP), Gruta do Boi (SCoNP), and Mariri (CVNP). Forty-ve plots of 10x10m were done along 4.5km transect lines, in 100m intervals, covering the living area of at least one group of capuchin monkeys as part of a broader survey for ecological description. We identi ed stone tool use sites with the same methodology as above in each plot. To evaluate stone material and the amount available in the areas, we did a 50x50cm subplot at the SW corner of each plot, where all stones with more than 3cm were counted and classi ed by rock type, measured, and weighted.
Food resources physical properties test Mechanics were measured with a Lucas Scienti c FLS-2 portable mechanical tester 1,39 . Although hardness is not a property per se, it is a derived concept traditionally used to measure the resistance to deforming under indentation. The technology of the equipment used can measure indentation of millimeter dimensions bringing more accurate results for such tests. The test used is called sharp indent (for more details, see 40 ) and, in addition to the hardness values, gives us the elastic modulus, de ned as the ratio of stress to strain in the elastic region 41 . Following an initial force loading, the displacement is held constant while recording the force decay for further 90s or until the load stabilizes, forming a curve.
Fitting a curve to the relaxation behavior allows the calculation of both an instantaneous (Ei) and in nite (E∞) elastic modulus. "Ei" represents the elastic modulus of a material if it could be loaded instantly, while "E∞" estimates the elastic behavior under an in nitely slow loading regime. In this study, we use "Ei" measurements, which are likely more relevant when investigating how food behaves under chewing loads 42 .    Hammerstone tool weight to process each resource at each site.