Subjects and study sites
We collected data on wild vervet monkeys (Chlorocebus pygerythrus) during February and March 2020 at the Inkawu Vervet Project (IVP) field site, located in Mawana game reserve (28° 00.327 S, 031° 12.348 E) in KwaZulu-Natal, South Africa. The study site is home to multiple wild groups of vervet monkeys, six of which are habituated to humans, regularly observed by researchers, and partake in experimental studies. Our data set comprised four of these groups, three of which are habituated since 2010 (Baie Dankie: N = 57, Noha: N = 39, Lemon Tree: N = 24) and the fourth since 2013 (Kubu: N = 19). In addition, the study area sustains at least three unhabituated groups, with many more living throughout the rest of the reserve. To enable data collection on unhabituated monkeys and to record any potential interactions with the novel stimuli, we placed motion-triggered video camera traps below two known sleeping trees of an unhabituated group (Congo: N = 11).
We collected data on the captive population in March 2020 at the Wild Animal Trauma Centre & Haven (WATCH) vervet sanctuary, in Vryheid, KwaZulu Natal, South Africa. At the time of data collection, the WATCH sanctuary housed three groups of vervet monkeys. For logistical reasons, we only included two groups in our study (Poena: N = 17 and Boeta: N = 3). Most of the monkeys arrived at the sanctuary and were cared for by humans, since they were a few weeks old, and only a few individuals arrived at a later life stage. At first, infant monkeys arriving at a very young age are housed indoors and bottle nursed by human caretakers. Once they reach 3 months of age, they are slowly integrated into a group of conspecifics of mixed ages. Since the goal is to release these individuals back into their natural habitat (if circumstances allow), caretakers, and occasionally researchers, limit their contact with the monkey groups as much as possible.
We presented all four habituated groups with eight novel stimuli representing distinct materials, structures, and odours. We categorized four of these items as human-made or processed: boiled pasta (green, red, natural coloured), popcorn, toy mice (with Baldrian herb scent), and plastic toy cars (yellow, blue, green, and red). One item, white seashells, represented a completely natural occurring object. We chose the remaining three items to have “naturalistic features”: fish (dead organic material in form of canned sardines), beef meatballs (raw organic material), and rubber butterflies of different colours (man-made material which mimics naturally occurring organisms) [Supplementary information (SI) Fig.S1]. We randomized the order of presentation of the novel stimuli across groups to avoid order effects and presented one type of novel stimulus at a time, on the ground, always with several items of each type to avoid potential monopolization by higher ranking group members. To attract the wild monkeys’ attention to the experimental area, prior to the start of the experiment, we placed a handful of familiar food (corn) in the middle of the area where the novel items were spaced out. The habituated monkeys are used to eating corn as this food item has been introduced during both the habituation process as well as during previous experimental studies (van de Waal et al. 2013; van de Waal et al. 2017). Our main goal was to record any potential behavioural reactions towards the novel stimuli after the monkeys had been attracted to the area (within 20 m) and thus seen the novel stimuli. We did all experiments during the early mornings 1–2 h after dawn and we presented only one category of novel stimuli per group per day. We video recorded all experiments with Sony handycams HDR-CX200, two mounted on tripods from different angles, and a third that was handheld by an observer zooming in on any observed explorative behaviours. We presented all novel stimuli to the monkeys for 20 min, to allow enough time for lower ranking individuals to also approach in case the most dominant individuals were present at the start of the experiment preventing the lower rankers from approaching. Because the microhabitats vary slightly across groups as well as within each groups’ home range, depending on their location on the day of our experiments, we categorized each experimental setup into three distinct habitat structures: open savannah (no canopy protection and no high grass), high grass (high grass but no canopy protection), and below tree (the experimental area was protected by canopy). In the open savannah, vervet monkeys are exposed to aerial predators like eagles and monkeys are observed to restrict their movement in high grass as the study area is home to a high abundance of pythons, capable of capturing vervet monkeys. Consequently, below trees represents the safest habitat structure for the monkeys as the tree canopy serves as protection from aerial predators and these areas do not have high grass.
The unhabituated group would not tolerate any human presence, as individuals from this group run away when human observers approach. They were however already habituated to eat corn when placed out in their habitat. We used an identical set up as with the habituated group, where we placed a small amount of corn in the middle of the area with the novel items. To record data from the unhabituated group, we placed the video camera traps in a way that they captured two different angles of the novel items, which we presented to the monkeys below two of their known sleeping trees. We used all the same novel stimuli as those used for the habituated groups. Because of the uncertainty regarding when the group would pass by the experimental location or when the monkeys would exactly use those sleeping trees, we presented the novel stimuli for 2 days in a row (unless a recording of any approaches by the group took place before the end of 2 consecutive days). Recordings from the camera traps thus allowed us to distinguish whether the group approached the novel stimuli on a single or multiple visits. For comparisons with the other group types, we only used the responses observed during the first visit.
At the WATCH sanctuary, we placed the novel stimuli in the main enclosure of the monkeys, who we moved into a side enclosure during the preparation of the experiment, and then let back into their normal enclosure. We used the same experimental protocol as for the habituated groups, including categories and numbers of novel stimuli, experimental duration, video camera placements, and recordings. As the captive monkeys were not used to corn, we used a few peanuts instead as the familiar food that would attract their attention to the experimental area.
Video coding and measurements
We coded all behavioural responses from video recordings. We recorded the number of close proximity approaches—those made to within 1 m of any of the multiple novel stimuli (food or item)—by any monkey that was present within a 20-m radius of the experimental location. As we defined a close proximity approach as each time a monkey approached within 1 m the novel stimuli, in any case where a monkey left the experimental area and then approached within 1 m again, this represented two approaches. For each approach that was made to the experimental area, we also distinguished whether or not the approach was made alone (when no other monkey was present within 1 m of the novel stimuli) or socially (when there was at least one other monkey present within 1 m of the novel stimuli). Once a monkey made physical contact with a novel item (0 m), we coded following exploration events: the number of smelling and tasting events, the number of times a monkey touched the novel item by hand, the number of times when a monkey chewed/bit the novel stimuli and the number of times a monkey lifted and moved an item. We then summed these behaviours into one exploration score labelled number of exploration events for each group and item. For each novel-food item, we additionally scored whether or not a monkey tasted it, defined as an event where a monkey licked a novel-food item, or every time a monkey put its lips onto a food item without ingesting it. All definitions of the coded behaviours as well as the frequencies of approaches and exploration events per group can be found in the ethogram in Table S1 and Fig. 4S in the Supplementary material.
To estimate the variation in human habituation among groups, we additionally performed a habituation test with the wild-habituated and captive groups. During this test, we exposed the monkeys to a human male that they had never seen before. The wild-habituated groups are familiar with researchers and project volunteers who attempt to distinguish themselves from other humans such as poachers by always wearing a turquoise blue cap while in the presence of the monkeys. In the habituation test, the man was dressed all in black clothes and wore a black cap. The man walked calmly towards the group of the monkeys shaking a Tupperware with corn as this is a familiar signal to the monkeys when they participate in research experiments. In the wild setting the man then placed the closed box with corn at his feet and as a group level habituation index, we measured the proportion of monkeys that approached the man to a distance of 1 m out of all the monkeys present within 20 m. In the captive setting, the man placed himself right at the enclosure mesh and placed peanuts right at his feet, which were in touchable distance to the monkeys. This test lasted 20 min in total.
We conducted the statistical analyses in R (version 3.6.1; R Core Team, 2020) and RStudio (version 1.2.5031; RStudio Team, 2020). We z-transformed covariates (habituation index and group size) to have a mean of zero and standard deviation of one before including them in the models to facilitate the interpretation of the coefficient estimates (Schielzeth 2010).
We first conducted a series of Spearman correlations to investigate whether any of the response measures (number of close proximity approaches, number of exploration events, and number of individuals within each group that tasted the food items) were correlated (Table 2). Since the number of individuals that tasted the different stimuli strongly correlated with the other response measures and this variable contained multiple missing values (N = 24), we excluded this variable from further analysis.
To address the study aims, we fitted four different Generalized Linear Mixed Models (glmm) to the data (Table 1). We checked all models (Model 1a,1b, 2, 3) for overdispersion and overall stability (see Supplementary material) and z-transformed continuous variables (Habituation index and group size) before including them as fixed effects (Table 1). We draw inference by comparing the full model with a reduced (null) model lacking the predictors of interest but containing all other model elements (Forstmeier and Schielzeth 2011) using a likelihood ratio test (test “Chisq”' in the R function anova, (Dobson 2002). We implemented this approach to avoid “cryptic multiple testing” and to maintain type 1 error rates at the desired nominal level of 0.05 (Forstmeier and Schielzeth 2011). We calculated individual p values for each predictor using the function drop1 and R squared using the function r.squaredGLMM.
In the first model (Model 1a), we investigated the effects of group type (three levels: wild habituated, wild unhabituated, and captive) and stimuli type (8 levels, see above) on the number of approaches (response variable, count data) observed in a given group. For Model 1a, which had a Poisson error structure and log-link function, we fitted the function glmer from the package lme4 (Bates et al. 2014). To account for group identity, we included the random intercept of group ID (7 levels, see above) into the model. We also included the logarithm of group size as an offset term to account for the different number of individuals in each of the groups.
To evaluate variation in close proximity approaches in relation to habituation level, we fitted a second model (Model 1b) using the same response variable, random structure, and the same offset as in Model 1a, but we changed the fixed effect structure. In addition to group type (2 levels: wild habituated and captive) and object type, we included the habituation index into Model 1b. As unhabituated groups did not have a habituation index, we excluded this group from those models where this variable was included (Model 1b and Model 2, see below).
In Model 2, we evaluated potential differences in explorative behaviour among the different group types (2 levels: wild habituated and captive), habituation level and stimuli type (8 levels). Similar to Models 1a and 1b, in Model 2, we included the random intercept of group ID as well as group size as an offset. To avoid overdispersion problems, we fitted this model using a quasi-Poisson model with a negative binomial distribution and the optimizer “bobyqa”.
In Model 3, we evaluated whether the proportion of social approaches varied according to habitat structure, habituation level, and stimuli type. In Model 3, we only included data from wild-habituated groups as these groups were the only ones that had been tested at locations with different habitat structures (below tree: N = 14, high grass: N = 10, and open savannah; N = 13). Model 3 was a binomial model with a response variable in the form of a matrix containing two columns corresponding to the number of social approaches and the number of individual approaches per trial (Baayen et al. 2008). Using such response variable, we account for the different number of approaches observed in different trials. Given that binomial models do not allow including offsets, we included group size as a control predictor. As before, we also included the random intercept of group ID was included in the model (although note that in this case group ID only had 4 levels, which is the threshold generally used to substitute a fixed by a random effect, meaning that it could have also been included as a control predictor).