Drivers of geophagy of large-bodied amazonian herbivorous and frugivorous mammals

Abstract

Mineral licks, critical for the survival of many large-bodied mammals in the Amazon, serve as keystone resources that influence the behavior and ecological dynamics of these species. This study presents the most comprehensive analysis to date on the drivers of geophagy—the consumption of soil by animals—at mineral licks in the Peruvian Amazon. Using a combination of camera traps and soil analyses from 52 mineral licks, we examined the visitation patterns of six large-bodied mammals: the black agouti (Dasyprocta fuliginosa), paca (Cuniculus paca), collared peccary (Pecari tajacu), Brazilian porcupine (Coendou prehensilis), lowland tapir (Tapirus terrestris), and red howler monkey (Alouatta seniculus). Our results reveal that mineral licks provide essential nutrients, particularly sodium (Na), which may be deficient in the diets of frugivorous species such as agouti, paca, and red howler monkey, supporting the mineral supplementation hypothesis. Conversely, the toxin adsorption hypothesis, which posits that animals consume soil to mitigate dietary toxins, was most strongly supported for the herbivorous Brazilian porcupine. The omnivorous collared peccary and the mixed-diet tapir exhibited complex interactions between soil characteristics, suggesting that both mineral supplementation and toxin adsorption play roles in their geophagy. This study highlights the importance of mineral licks for the conservation of Amazonian mammals, emphasizing their role in supporting biodiversity by providing critical nutritional resources that enhance species fitness and ecological resilience. Our findings underscore the need for the protection of these sites, which are integral not only to the survival of individual species but also to the health of the broader Amazonian ecosystem.

Supplementary Information

The online version contains supplementary material available at 10.1038/s41598-024-80237-0.

Introduction

The Amazon rainforest holds some of the richest biodiversity on the planet¹, and the conservation of that biodiversity is a priority in the face of global environmental disturbance and change^2,3. However, little is known about the diet, behavior, physiology, and movement of many of the large-bodied mammals in the Amazon, several of which are keystone species⁴ that uphold the very structure of the forest itself^5–7. Many of these species are also targeted as prey for human hunters, forming the foundation of food security and dietary protein for local and Indigenous hunters^8,9. Understanding the resource needs of large-bodied mammals is critical to promoting successful growth and reproduction, and therefore conservation and management.

Mineral licks, also called salt licks or clay licks, are critical resources for birds and mammals in the tropics and can themselves be considered a keystone resource¹⁰. Animals visit mineral licks to exhibit geophagy, consuming soil that may alleviate physiological stressors that, in turn, can inhibit growth and reproduction. Mineral licks are also critical for predators, making them ‘trophic nodes of attraction¹¹, and studies show that mineral licks are common sites of predation of arboreal species in particular^12–14. Human hunters also visit mineral licks seeking prey¹⁰ and mineral licks often hold key places in local culture⁸. The most common visitors to Amazonian mineral licks include the red brocket deer (Mazama americana), collared peccary (Pecari tajacu), black agouti (Dasyprocta fuliginosa), Brazilian porcupine (Coendou prehensilis), paca (Cuniculus paca), tapir (Tapirus terrestris), red howler monkey (Alouatta seniculus), and more^15–17.

The services that mineral licks provide individual animals likely vary based on the animal’s diet. The mineral supplementation hypothesis has received the most attention in previous studies on geophagy, and states that soil may provide dietary minerals that are critical to growth and reproduction that are missing from the diets of many herbivores and frugivores^18–20. Fruits and leaves in the Amazon are notoriously mineral-poor and are particularly low in sodium (Na)²¹. Accordingly, Na limitation in particular has been proposed as a driver of geophagy for many species of mammals and birds²². The toxin adsorption hypothesis may also describe geophagy of some species. Folivores in particular may consume alkaloid-laced leaves or other plant material that then causes gastrointestinal discomfort^23–25. Clays consumed in mineral lick soils may adsorb these dietary toxins, alleviating this discomfort. Animals may also select mineral licks based partially on the habitats that the lick occurs in Tobler et al.²⁶, showing the importance of testing habitat-based characteristics when assessing drivers of geophay.

The drivers of geophagy of specific species are difficult to assess overall because mineral licks are often located in remote areas and are difficult to find, resulting in a low sample size of licks in most studies. Many studies only test some soil characteristics and are therefore limited in scope; for example, examining concentrations of a few cations and looking at the correlations between their concentrations and animal visitation. Others do not test control soil samples or do not include a robust regression-based analysis of results. Inferences that can be drawn from these studies are often limited. In this study, we combine camera trap and soil data from 52 mineral licks in the same watershed in the Peruvian Amazon in a robust generalized linear modeling framework to assess the drivers of geophagy of six large-bodied Amazonian mammals. We predicted that species may derive many benefits from mineral licks, but that the mineral supplementation hypothesis would better describe geophagy of frugivores while the toxin adsorption hypothesis would better describe geophagy of folivores. Species with a varied diet would have weak evidence supporting either of these hypotheses.

Results

Visit frequency was highly variable across mineral licks and species. Agouti had a mean visit frequency of 24.17 visits per 100 camera nights (SD = 25.54, n = 30 licks), paca had a mean visit frequency of 19.55 visits per 100 camera nights (SD = 32.08 n = 32 licks), the collared peccary had a mean visit frequency of 16.00 visits per 100 camera nights (SD = 21.99, n = 27 licks), the Brazilian porcupine had a mean visit frequency of 22.91 visits per 100 camera nights (SD = 25.46, n = 30 licks), the tapir had a mean visit frequency of 15.36 visits per 100 camera nights (SD = 25.57, n = 29 licks), and the red howler monkey had a mean visit frequency of 8.21 visits per 100 camera nights (SD = 8.78, n = 14 licks). Red howler monkeys were the most selective species, visiting the fewest licks with the lowest visit frequency.

General results from soil analyses are reported in Griffiths et al.²⁷, but summarized here. Mineral licks soils differed greatly from control samples, and from each other. In general, mineral lick soils had elevated concentrations of all biologically active macro- and micro-nutrients, had a more basic pH, and higher clay content.

Regression results were variable amongst species (Table S2). Agouti were more likely to visit mineral licks as Na, Al, Cu, and Mn concentrations and soil pH increased. The effects of Na, Al and pH were highest, with similar effect sizes that were highly significant (Table 1). Agouti visit frequency was predicted to increase as B, Cu, Fe, Mn, Zn, Na, and Al concentrations, pH, and surface roughness increased, and decrease with increasing K, P, and N concentrations, clay content, and distance from the closest stream (Fig. 1). pH and Zn had the strongest positive effect, while N had the strongest negative effect on predicted visit frequency. Cu and Fe, Cu and Zn (Fig. 2), and Fe and Mn (Fig. S1), were interacting in the optimal model.

Table 1.

Coefficients and SEs for optimal zero-inflated models testing the drivers of geophagy of six amazonian mammals at Interior forest mineral licks.

Covariate	Agouti	Paca	Collared Peccary	Brazilian porcupine	Tapir	Red howler monkey
Intercept	P: 1.738 (0.114)***; B: 1.722 (0.434)***	P: 3.607 (0.072)***; B: 2.097 (0.465)***	P: 2.573 (0.218)***; B: 2.213 (0.525)***	P: 2.213 (0.163)***; B: 3.324 (0.915)***	P: 1.348 (0.201)***; B: 3.563 (1.313)**	P: 1.595 (0.145)***; B: 2.678 (0.417)***
Exchangeable Al (meq/100 g)	B: − 1.776 (0.527)***	P: − 0.522 (0.094)***; B: − 2.118 (0.628)***	P: − 1.457 (0.218)***; B: − 2.335 (0.858)**	P: − 0.899 (0.106)***; B: − 3.477 (0.973)***	P: 0.247 (0.135); B: − 1.205 (0.726)
Soluble B (ppm)	P: 0.772 (0.108)***; B: 0.605 (0.468)	P: 0.189 (0.048)***	B: 1.328 (0.59)*	P: 0.757 (0.12)***; B: − 2.109 (0.655)***	P: 0.127 (0.071)
Exchangeable Ca (meq/100 g)						B: − 0.681 (0.267)*
Clay content (%)	P: − 1.21 (0.121)***; B: 1.158 (0.453)*	P: 0.122 (0.075); B: 1.467 (0.644)*	P: − 0.824 (0.249)***	P: 1.057 (0.091)***; B: − 1.642 (0.854)	P: 0.86 (0.121)***; B: − 2.381 (1.109)*
Cu (ppm)	P: 0.021 (0.1); B: − 1.35 (0.464)**	P: − 0.31 (0.09)***; B: − 2.072 (0.643)***	P: − 0.369 (0.148)*; B: − 2.287 (0.703)***	P: 0.395 (0.139)**; B: − 0.556 (0.579)	P: 0.555 (0.195)**; B: − 1.686 (1.09)
Cu (ppm):Fe (ppm)	P: − 0.172 (0.093)	P: 0.319 (0.073)***		P: − 0.284 (0.154); B: − 2.202 (0.838)**	P: 0.367 (0.114)***; B: 1.131 (0.624)
Cu (ppm):Zn (ppm)	P: 0.531 (0.136)***	P: − 0.735 (0.1)***		P: − 0.7 (0.175)***; B: 1.273 (0.74)
Distance from closest stream (km)	B: 0.625 (0.37)	P: 0.366 (0.082)***; B: 0.746 (0.392)	P: 1.195 (0.168)***; B: 0.935 (0.548)		P: − 1.746 (0.164)***; B: − 1.111 (0.771)
Elevation (m)		P: − 0.304 (0.066)***	P: 1.853 (0.222)***; B: 1.089 (0.601)	P: − 1.445 (0.107)***; B: − 1.892 (0.761)*	P: 0.702 (0.167)***; B: − 1.962 (0.898)*
Fe (ppm)	P: 0.739 (0.094)***	P: 0.072 (0.062); B: 0.402 (0.41)	P: − 0.624 (0.093)***	P: 1.804 (0.195)***; B: 0.671 (0.702)	P: − 1.794 (0.218)***; B: − 1.868 (0.975)
Fe (ppm):Mn (ppm)	P: − 0.368 (0.056)***	P: − 0.217 (0.086)*; B: 0.849 (0.483)	P: 0.166 (0.105)		P: 0.645 (0.1)***
Exchangeable K (meq/100 g)	P: − 0.467 (0.099)***	P: 0.835 (0.12)***; B: 0.921 (0.559)	P: 0.854 (0.197)***; B: 1.635 (0.62)**	P: − 0.335 (0.118)**; B: 1.563 (0.827)	P: − 1.042 (0.149)***; B: 1.944 (1.439)	P: 0.167 (0.105)
Mn (ppm)	P: 0.127 (0.048)**; B: − 0.894 (0.337)**	P: − 0.944 (0.111)***; B: − 1.452 (0.548)**	P: − 0.591 (0.124)***; B: − 1.441 (0.573)*	P: − 0.695 (0.084)***; B: − 2.171 (0.644)***	P: 0.047 (0.1); B: − 2.006 (0.789)*
Total N (%)	P: − 2.632 (0.206)***		P: − 1.427 (0.171)***; B: − 1.375 (0.52)**	P: − 1.557 (0.197)***	P: 0.991 (0.135)***
Exchangeable Na (meq/100 g)	P: − 0.095 (0.036)**; B: − 2.035 (0.942)*	P: 0.197 (0.033)***; B: − 1.17 (0.555)*	P: 0.441 (0.056)***; B: − 0.775 (0.527)		P: 0.614 (0.165)***	P: 0.239 (0.053)***; B: − 0.448 (0.23)
Free P (ppm)	P: − 0.443 (0.07)***	P: − 0.013 (0.072); B: − 0.049 (0.314)	P: 0.394 (0.131)**	P: − 0.552 (0.099)***	P: 0.154 (0.092); B: − 1.512 (0.703)*	B: − 0.721 (0.27)**
pH	P: 1.26 (0.089)***; B: − 1.724 (0.568)**	P: − 1.063 (0.129)***; B: − 3.072 (0.834)***	P: − 1.463 (0.238)***; B: − 2.992 (0.87)***	P: 0.312 (0.109)**; B: − 3.266 (0.969)***	P: − 0.894 (0.156)***; B: − 4.721 (1.719)**
pH: free P (ppm)		P: − 0.373 (0.071)***; B: 0.946 (0.4)*	P: 0.464 (0.096)***	P: 0.441 (0.074)***	P: − 0.467 (0.111)***
Surface roughness (m)	P: 0.322 (0.079)***	P: − 0.203 (0.066)**	P: − 1.176 (0.235)***	P: − 0.618 (0.082)***; B: 0.855 (0.515)	P: 0.367 (0.149)*; B: 3.755 (1.377)**
Zn (ppm)	P: 0.926 (0.099)***	P: 0.248 (0.121)*	P: 0.353 (0.127)**	P: 0.704 (0.149)***; B: − 0.631 (0.469)	P: − 0.647 (0.189)***; B: − 1.268 (0.942)

Poisson half of the model denoted by P: and binomial half of the model denoted by B: * denotes p < 0.05; ** denotes p < 0.01; *** denotes p < 0.001.

Fig. 1.

Model results from optimal zero-inflated models showing the marginal effects of soil and landscape characteristics of Amazonian interior forest mineral licks on visit frequency of five herbivores. Each line represents the given covariate varying across the inner quartile range of values found in mineral licks that the species visited. For visualization purposes, all other covariates are held at the mean value of mineral lick soils where the species visited.

Fig. 2.

Model results from optimal zero-inflated models showing the marginal effects of the interactions between soil Cu and Fe concentration, and Cu and Zn concentration, on visit frequency of herbivores which had one of these interactions included in their optimal model. Each line represents the given covariate varying across the inner quartile range of values found in mineral licks that the species visited. For visualization purposes, all other covariates are held at the mean value of mineral lick soils where the species visited.

Paca were more likely to visit mineral licks as Al, Cu, Mn, Na, and P concentrations and pH increased, and less likely to visit as K and Fe concentrations, clay content, and distance from the closest stream increased (Table 1). pH, Al and Cu had the strongest effect on visit probability, and all were statistically significant. Visit frequency was predicted to increase as B, Fe, K, Na, and Zn concentrations, clay content, and distance from the closest stream, increased (Table 1; Fig. 1), with Cu interacting with Zn and Fe (Fig. 2). Predicted visit frequency decreased as Al, Mn, P, and Cu concentrations, elevation, and surface roughness increased. The effect of pH on visit frequency depended largely on P concentration (Fig. S2).

Collared peccary were more likely to visit mineral licks as Al, Cu, Mn, N, and Na concentrations, and pH all increased, and were less likely to visit as K and B concentrations, elevation and the distance to the closest stream increased (Table 1). Overall, pH had the strongest positive effect on the likelihood that peccary visit mineral licks, and K concentration had the largest negative effect; both effects were statistically significant. Visit frequency of the collared peccary was predicted to increase as Na, K, P, Cu, and Zn concentrations, elevation, and distance from the closest stream increased (Table 1; Fig. 1). Predicted visit frequency decreased with increasing Al, B, Fe, Mn, and N concentrations, clay content, pH, and surface roughness. The strongest positive effects on visit frequency were elevation and distance from the closest stream, while the strongest negative effects were from Al and N concentrations and pH (Table 1). The optimal model also included interactions between Mn and Fe (Fig. S1) and pH and P (Fig. S2).

The Brazilian porcupine was more likely to visit mineral licks as Al, B, Cu, Mn, and Cn concentrations, clay content, elevation, and the distance to the closest stream all increased, and as Fe, K, and surface roughness decreased (Table 1). Visit frequency was predicted to increase as B, Cu, Fe, Zn, and Mn concentrations, pH, and clay content increased (Fig. 1). Predicted visit frequency decreased with increasing K, P, and N concentrations, surface roughness, and elevation (Table 1). Fe concentration had the highest positive effect on visit frequency, while N concentration and elevation had the highest negative effect (Table 1). The optimal model also included interactions between Cu and Fe, Cu and Zn (Fig. 2), and pH and P (Fig. S2).

The probability of a tapir visiting a mineral lick increased as Al, Cu, Fe, P, and Zn concentrations, clay content, pH, distance from the closest stream, and elevation increased, and decreased as surface roughness increased (Table 1). pH had the largest influence on visit probability. Predicted visit frequency of the tapir increased as Na, Al, B, Cu, Mn, and N concentrations, clay content, surface roughness, and elevation increased and K, Fe, and Zn concentrations, pH, and distance from the closest stream decreased (Table 1; Fig. 1). N concentration had the largest positive effect on visit frequency, while distance from the closest stream and Fe concentration had the largest negative effects. The tapir was the only species which had a positive coefficient estimate for N concentration. The optimal model also included interactions between Cu and Fe (Fig. 2), Fe and Mn (Fig. S1), and pH and P (Fig. S2).

The optimal model for the red howler monkey was a submodel of the major cations global model which included Ca, Na, and P concentrations in the binomial portion of the model and Na and K in the Poisson portion (Table S2). All of the covariates had positive influences on predicted visit frequency (Table 1). The model with habitat covariates only was the next highest ranked with one submodel scoring a ∆AIC = 7.26 compared to the optimal model. However, this model and all others performed relatively poorly, with a cumulative Akaike weight of 0.02 for all models that were not from the major cations candidate set.

Overall, several results from models showed similar tendencies across species. For example, the effect of Na on visit frequency was similar for the paca and agouti, and the collared peccary, tapir, and red howler monkey (Fig. 1). Responses to K were similar for the tapir and red howler monkey, and the agouti and Brazilian porcupine. The Brazilian porcupine, paca, and agouti were predicted to respond similarly to P concentration, as were the tapir and red howler monkey. Predicted responses to B were similar between the paca and agouti, and collared peccary and tapir. The collared peccary and tapir also responded almost identically to Cu, Fe, Mn, and pH. The Brazilian porcupine and agouti were also closely aligned in their responses to Cu, Zn, N, and pH (Fig. 1).

Discussion

Overall, both hypotheses were generally supported as potential drivers of geophagy at Amazonian interior forest mineral licks. The majority of covariates that were tested related to the mineral supplementation hypothesis (Table S1). Every mineral that was tested appeared in the optimal model of at least one species, and every species (except the red howler monkey) had at least eight minerals appear in the visit frequency half of their optimal model. The agouti and paca both had strong support for the mineral supplementation hypothesis, and their results were quite similar for many of the minerals examined. The red howler monkey also had the strongest support for the mineral supplementation hypothesis. These three species are all frugivorous^28–33 and may not receive sufficient dietary minerals from the fruit that they eat. In particular, they may be Na-deficient²² and mineral lick soils could provide needed supplementation of Na. Na also had a strong impact on tapir visitation, and it also likely has a largely frugivorous diet in this region. The tapir has been known to consume fruit when it is in season, and exhibit searching behavior for fruit³⁴, but consumes browse when fruit is unavailable. This region is rich in aguaje³⁵, a preferred fruit of the tapir³⁴.

All species tested except for the red howler monkey had microminerals included in their optimal models, like Cu and Zn. For example, tapir were much more likely to consume soils high in Cu but low in Zn, and the strength of this interaction suggests that a high Cu: Zn ratio might be critical for the tapir to maximize Cu absorption. Cu plays a critical role in enzyme function, immune and nervous system health, and antioxidant defense in large bodied mammals³⁶. B also appeared in the optimal model of the agouti, paca, collared peccary, and porcupine; however, the role of B as a biologically active mineral remains largely unknown. Al also appears in several models having a negative impact on visitation. This makes sense since Al is toxic to many animals at high concentrations³⁷. While other studies have measured some of these minerals in soils at mineral licks^17,38–41, we are proposing here that they also may partially drive geophagy, and certainly can supplement the diets of mineral lick visitors. The tapir was the only species for which N had a positive effect. Tapirs have a specialized digestive system that includes a large caecum where microbial fermentation takes place⁴². These microbes may be fertilized by N in the diet of the tapir.

Several previous studies on mineral licks around the world have concluded that the mineral supplementation hypothesis may explain the behavior of visiting species, though sample sizes are often limited. The importance of Na in particular has been highlighted as a potential driver of geophagy^43–46, specially for Amazonian frugivores which may be Na deficient due to low Na concentrations in many Amazonian fruits^21,22. Notably, a review of the drivers of geophagy of New World monkeys by¹⁹ found that the functional aspects of geophagy were largely unknown, a conclusion supported more recently by a review done by⁴⁷.

The toxin adsorption hypothesis had support in models from the Brazilian porcupine and the tapir. The porcupine in particular was partial to high-clay soils, where any lick with > 35% clay content was predicted to be heavily visited. Similarly, pH was among the most influential factors in selection of soils by porcupines. The species is known to be exclusively herbivorous with a diet of mainly buds, bark, and seeds^48,49, which may be laden with toxic alkaloids. The tapir also selected for high-clay soils, but pH had a negative effect on visitation rate. Therefore, evidence is limited for the toxin adsorption hypothesis; however, since the tapir may also consume browse as well as fruit, the species likely receives some dual benefits from licks.

In many cases the coefficients of both pH and clay content, the core covariates of this hypothesis, were negative in the Poisson halves of optimal models. The agouti, paca, collared peccary, and the red howler monkey all fit this pattern, with limited to no support for the toxin adsorption hypothesis based on the model results for pH and clay content.

Several other studies on mineral licks around the world have found associations between pH, clay content, and geophagy, though often with limited sample sizes^40,50. Even more studies have found that visiting animals may derive multiple benefits from soils, including both mineral supplementation and toxin adsorption^{17,19,23,39,51–53}, and that is likely the case in our study site as well. Our large sample size of mineral licks and camera nights expands upon the conclusion that dual benefits may be received by offering specific insights into the relationships between mineral lick visitation and soil characteristics for six understudied species.

All species studied also showed some variation in visit frequency across habitats, with distance from the closest stream and elevation appearing in the optimal models of four species, and surface roughness appearing in five species models. While the inclusion of habitat-based covariates may explain variation in visit frequency that is associated with differential use of habitats, it may not fully account for any differences in species distributions at the study site. Some animals, like the tapir⁵⁴ may exhibit searching behavior for mineral licks, leaving their preferred territories to specifically visit mineral licks they select. For these species, visit frequency is likely a strong metric for mineral lick preference. Previous studies⁵⁵ have indicated that the black agouti, collared peccary, and paca are widespread in this region, and therefore may choose mineral licks based on the services they provide. However, others, like the red howler monkey, may simply be rarer in some parts of the study site compared to others, and their lack of visitation at many mineral lick sites may simply be because they do not live nearby and do not roam to search for mineral licks. Therefore, we recognize that results from the red howler monkey may not be fully able to explain its mineral lick choice, and that spotty distribution may influence the results.

Conclusion

Our predictions were confirmed for several species: the frugivorous agouti, paca, and red howler monkey had greatest support from the mineral supplementation hypothesis; the folivorous Brazilian porcupine had greatest support from the toxin adsorption hypothesis, and the omnivorous collared peccary had greatest support from the habitat selection hypothesis. The tapir, which eats both fruit and leaves and may shift its diet seasonally according to fruit abundance, had support from both the mineral supplementation and toxin adsorption hypotheses.

This study provides the most comprehensive analysis of the drivers of geophagy to date and gives key insights into the dietary and behavioral ecology of six understudied mammals. The importance of so many soil characteristics in species models emphasizes the services that mineral licks may provide to visiting mammals. The high density of mineral licks and visitation rates by mammals in our study site show that mineral licks may be more common in the Amazon than previously thought, and are likely extremely important for species conservation in the region, providing boosts to the productivity of species that in turn support their ecosystems and, in many cases, local human communities.

Methods

Study site

This study took place in the Sucusari River basin in the Peruvian Amazon, in the Department of Loreto, Peru. The Sucusari River is a tributary of the Napo River, and lies about 120 km from the city of Iquitos, the capital of Loreto (Fig. 3). The area is encompassed by the titled lands of the Maijuna community of Sucusari, and the Maijuna-Kichwa Regional Conservation Area (MKRCA), a 391,000-ha protected area that adjoins the Sucusari titled lands to the north. The dominant habitat types in the Sucusari River basin are terra firme primary upland forests and seasonally flooded, or floodplain, forests⁵⁶. The climate is characterized by a mean annual temperature of 26ºC and a mean annual rainfall of 3,100 mm⁵⁷. All mineral licks in the study were identified with the support of Maijuna hunters, who pass down the locations of mineral licks from generation to generation⁸.

Fig. 3.

Map of the study site in the Sucusari River basin in the Peruvian Amazon, including the titled lands of the Maijuna community of Sucusari and the Maijuna-Kichwa Regional Conservation area, which contains 83 interior-forest mineral licks. Map created using QGIS (version 3.32.0)⁵⁸.

Camera trapping

We set camera traps (Bushnell Aggressor, Boly Scout Guard) at 80 mineral licks in the Sucusari River basin (Fig. 3) between August, 2018, and June, 2019, in a rotation. First, we set cameras in 25 mineral licks randomly chosen throughout the river basin. After 80 days, cameras were retrieved, batteries and SD cards were changed, and cameras were rotated to new mineral licks. This process repeated for four total rotations. Some mineral licks which experienced camera malfunctions received camera traps during two different rotations to maximize the number of mineral licks which achieved at least 55 camera nights of survey effort. If a mineral lick was flooded (had several cm of standing water) when it was visited, sampling at the site was deferred to the next rotation. Mineral licks with fewer than 55 camera nights were excluded from the study because they did not meet the minimum sampling requirements of camera nights for species richness estimations established by⁵⁹, leaving a total sample size of 52 licks which had enough camera nights to be included in analyses.

Camera traps were placed 50 cm from the ground facing the active “face” of the lick, which was determined by looking for recent teeth and claw marks, following²⁶. In licks where the active face could not easily be determined, Maijuna hunters were asked where they have observed animals feeding at the site, and camera traps were placed facing that location. In licks where there were multiple active faces, we placed multiple camera traps to achieve complete coverage of the lick. Camera traps were set to burst photo mode, taking three still images for each trigger, with a delay of two minutes between triggers.

We identified all large-bodied (> 1 kg) mammals in camera trap images⁶⁰ using CameraBase⁶¹. We then classified images into independent visit events, which we considered any visit to the mineral lick by the same species that was separated from the last previous visit by that species by at least one hour⁶². We calculated a visit frequency for each species at each mineral lick as the total number of independent visits recorded per 100 camera nights of sampling effort.

Soil collection and analysis

We revisited all mineral licks in the Sucusari River basin in January, 2022, to extract soil samples. Some licks were not revisited because they were inaccessible at the time. At each mineral lick, we collected at least three soil samples: one “consumed” sample, and two controls. For the consumed sample, we identified recent teeth marks that indicated the presence of a feeding site, and scraped the subsoil from the site horizontally with a clean trowel until about 1 kg of soil was removed⁶³. This soil was placed in a sealed and labeled plastic bag. In licks where there were multiple feeding sites that had visually different soil types (based on texture and color), multiple “consumed” samples were collected. To extract control samples, we walked 5 m from the feeding site in a random direction, outside of the lick. We scraped away organic debris from the spot, then removed 1 kg of subsoil with a clean trowel, which was placed into a separate labeled plastic bag. We repeated this process for the second control sample. All soil samples were dried in the shade in clean plastic bins for at least 24 h before being rebagged and relabeled for analysis. For mineral licks that lie in a cluster, with < 50 m separating licks, only one set of soil samples was taken from one mineral lick in the cluster unless soil types were visibly different. Samples were analyzed for their physical and chemical properties by SGS Peru in Lima, Peru²⁷. This process yielded a total sample size of 165 soil samples.

Data analysis

We assessed the drivers of geophagy for six focal species using a generalized linear modeling approach. The six focal species were the most common mineral lick visitors, apart from the red brocket deer, which formed the pilot for this study given that it visited most mineral licks⁶⁴. These six species were the black agouti, paca, collared peccary, Brazilian porcupine, lowland tapir, and red howler monkey. First, we collapsed the soil data at each mineral lick into one consumed sample and one control sample by taking the mean of soil characteristics when multiple consumed or control samples existed for a single lick, to account for pseudoreplication caused by repeated sampling. We calculated a mean visit frequency for each species at each lick the species visited, with units of number of independent visits by the species at the lick per 100 camera nights of sampling effort. Visit frequencies for licks which were never visited by the species were assigned the value of 0, assuming that each species is distributed throughout the Sucusari River basin and could visit any mineral lick they chose. These visit frequencies were merged with the corresponding consumed soil samples from the relevant licks. Samples of control soils from mineral licks were also used in the regression analysis, each of which was assigned a visit frequency of 0, assuming that they were never consumed since they were outside of a mineral lick. As a result, each species had a sample size of n = 129 soil samples. To summarize, control samples and unvisited consumed samples had a visit frequency of 0; visited consumed samples had the calculated visit frequency by the species.

We used a zero-inflated model with a combined binomial/Poisson distribution to assess drivers of geophagy of each species, with visit frequency as the response variable. This model type was uniquely suited to the data because the binomial portion of the model tests which soil or landscape characteristics drive whether a mineral lick is visited or not, while the Poisson portion of the model estimates which characteristics may drive changes in visit frequency. For all species except the red howler monkey, we constructed a global model for each species, and visually examined residual plots to check model fit. Covariates that were included in the model were chosen based on the leading hypotheses for drivers of geophagy of herbivores (Table S1), and were all scaled so that effect sizes could be compared. We also included habitat-based covariates under the hypothesis that species detection frequency may be influenced by habitat, which may help explain variation in visit frequency data. We checked covariates for collinearity before including them in the global model, with a correlation threshold of 0.7 for inclusion⁶⁵. We tested correlated covariates one at a time in the global model and used the one that resulted in the lowest AIC for model selection and inference. We followed a backwards-stepwise model selection process to choose the optimal model for each species, dropping one covariate at a time which resulted in the largest drop in AIC⁶⁶; the dropped covariate could come from the binomial or Poisson portions of the model.

Since red howler monkeys visited only 14 licks in the study, we did not include all covariates in a single global model to aid in model convergence. Instead, we formed four competing global models which directly related to the hypotheses for geophagy as our candidate set: one with major cation minerals, one with micronutrient minerals, one with toxin adsorption covariates, and one with habitat-based covariates only (Table S1). Then, we used the stepwise selection process on each of these four models to obtain the models with the lowest AIC of each set. We ranked these four models against each other using AIC to determine which hypothesis had the most support; the model with the lowest AIC of the four was considered the optimal model.

All zero-inflated models were constructed using the glmmTMB package⁶⁷ in R (version 4.1.1)⁶⁸.

Electronic supplementary material

Below is the link to the electronic supplementary material.

Author contributions

Brian M. Griffiths, Lesa Griffiths, and Michael P. Gilmore conceived the ideas for this study, and Brian M. Griffiths and Michael P. Gilmore designed the methodology. Brian M. Griffiths and George Hansbrough collected and analyzed the data; Brian M. Griffiths led the writing of the manuscript; Diego Valderrama provided key statistical insights and editing of the final manuscript, along with critical mentorship. Brian M. Griffiths, Lesa Griffiths, and Michael P. Gilmore pursued funding for the study.

Data availability

Data are publicly available at: Griffiths, Brian; Hansbrough, George; Griffiths, Lesa; Valderrama, Diego; Gilmore, Michael (2024), “Drivers of geophagy of large-bodied Amazonian herbivorous and frugivorous mammals”, Mendeley Data, V2, doi: 10.17632/9vpzd6rmrn.2.

Declarations

Competing interests

The authors declare no competing interests.