Many mutualisms are exploited by third-party species, which benefit without providing anything in return. Exploitation can either destabilize or promote mutualisms, via mechanisms that are highly dependent on the ecological context. Here we study a remarkable bird–human mutualism, in which wax-eating greater honeyguides (Indicator indicator) guide humans (Homo sapiens) to wild bees’ nests, in an exchange of knowledge about the location of nests for access to the wax combs inside. We test whether the depletion of wax by mammalian and avian exploiter species either threatens or stabilizes the mutualism. Using camera traps, we monitored feeding visits to wax comb made available following honey harvests. We found that greater honeyguides face competition for wax from conspecifics and nine exploiter species, five of which were not previously known to consume wax. Our results support the hypothesis that heterospecific exploiters stabilize the mutualism, because wax depletion by these competitors likely limits feeding opportunities for conspecific exploiters, favouring the early-arriving individual that guided humans to the bees’ nest. These findings highlight the importance of the ecological context of species interactions and provide further evidence for how mutualisms can persist because of, and not in spite of, exploitation by third-party species.

Study site

We carried out this study in a 28 km² area within the Niassa Special Reserve in northern Mozambique. Our study area is in range of Yao honey-hunters’ foraging trips from Mbamba village (12º12'S, 38º01'E; ca 2,000 inhabitants including > 20 regular honey-hunters). Yao honey-hunters traditionally reward greater honeyguides after a successful honey-harvest by leaving a small pile of beeswax near the harvested bees’ nest (figure 1), and therefore the main source of wax for greater honeyguides in this landscape is that left behind or exposed by humans at the harvest site of a bees’ nest. Within this area, the costs and benefits of the human-honeyguide mutualism appear to approximate those under which it presumably evolved: there is little apiculture and a minimal cash economy for buying sugar instead of honey. The habitat is deciduous Miombo woodland (dominated by the tree genera Brachystegia and Julbernardia) punctuated by granite inselbergs and narrow strips of riverine forest along the Lugenda River and seasonal tributaries (altitude 400–450 m). The climate is sub-humid tropical with mean minimum and maximum air temperatures ranging between 16–33°C in the dry season (May–October) and 22–32°C in the wet season (November–April). Rainfall begins in November and ends in late April or early May; during this period, precipitation averages 250–350 mm per month. Bees’ honey stores, which build up throughout the rains with the flowering of dominant species, peak in May–June, deplete as the dry season progresses and then peak again in November–December following the flowering of trees prior to the following rainy season.  Data were collected from 24 September to 25 October 2015, 29 August to 15 October 2017, 4 November 2018 and 24 September 2021 to 7 October 2021.

Honey-harvests, wax and camera trap placement

Wax-eating data were collected at 26 small piles of wax comb (ca. 0.1–1.5 kg) placed on the ground or a horizontal tree log, in a manner reflecting the honeyguide rewarding culture displayed by Yao honey-hunters . These were located at 26 bees’ nests (six nests in 2015, four in 2017 and 16 in 2021) which were found during honey-hunts conducted in a traditional manner. Eleven sites from 2015 and 2017 where a bees’ nest was harvested and camera traps placed, were excluded from the analysis either due to camera trap malfunction or because the wax was left without harvesting the bees’ nest. To initially locate the bees’ nests, one or two researchers (DJL, CNS, JSC) accompanied two Yao honey-hunters (one of whom was a main assistant and is a co-author: OY) on a honey-hunt as they elicited guiding behaviour from greater honeyguides using stereotypical calls. Twenty-two of the 26 nests were located by guiding from a greater honeyguide, 2 were located opportunistically prior to being guided while walking in the same habitat, and 2 were found in the same habitat after unsuccessful guiding by a greater honeyguide. All bees’ nest locations were previously unknown to us or the honey-hunters prior to the study, and at all 26 sites honey and wax were extracted by two honey-hunters using traditional methods (smoke, axes). The honey-hunters retained the combs containing honey and left behind a pile of wax combs without honey: these included a mixture of wax types, including smaller pieces of newly-produced, empty white wax comb (most favoured by greater honeyguides), older wax comb with bee brood in it, and dark wax comb containing old larval casings and no brood (least favoured by greater honeyguides, pers. obs. and [16]). In all cases, the wax piles were consistent with those that Yao honey-hunters naturally leave as reward following a successful harvest at a bees’ nest they were guided to by a greater honeyguide.

To record animals eating wax combs during the day and night, a camera trap (Acorn 6210, Ltl Acorn, Denmark, Wisconsin) was set up for 1–9 days at each site (total = 88.8 trap days; mean ± SE duration per site = 3.41 ± 0.39 days) at a height of 80–90 cm above ground and 1.5–2 m from the wax pile, facing slightly downwards (26.6 trap days in 2015, 15.7 trap days in 2017, 46.5 trap days in 2021). Each camera trap was set to trigger with a 5 s delay and take photos at 10 s intervals for as long as the camera was motion-triggered. The cameras also recorded a video clip of 10 s or 30 s, alternating with sets of 3 photos for as long as the camera was motion triggered. Coordinates of the bees’ nest were marked with GPS (Garmin eTrex 30, Garmin USA) and all camera traps checked every 1–2 days.

For each animal detected, the duration of wax feeding events was estimated from image and video clip time stamps. Feeding events were defined as contact between the mouth or bill of the animal and any part of the wax comb pile. Whenever possible we recorded which food type was eaten (wax only, brood only, or wax and brood together), but due to the resolution of the images we were not able to reliably score which type of wax was eaten by each animal. Short feeding events, such as when the animal disappeared prior to a second image being taken, and therefore without a reliable ‘end time’, were recorded as having a 2 s duration. At one wax site, feeding greater honeyguides disappeared to feed inside the log from which the honey and wax was harvested, and were observed emerging with small pieces of white wax. For these few observations (5 of 39 visits at this site), the period that the bird was out of view was recorded as feeding time.

 Statistical analysis

All statistical analyses were carried out using R (4.0.3 version). To document the level of competition that greater honeyguides face for beeswax, we summarised the following for each wax placement site (n = 26): number of wax-eating visits by each species, first arriving species, and species which ate the last remaining wax (where known). Feeding times for all wax-eating species were plotted over 24 h and compared to median sunrise and sunset times (generated using the R package ‘suncalc’) for our study duration).

To test prediction (i) of hypothesis one (competitors deplete the wax before greater honeyguides feed) we calculated the proportion of sites at which greater honeyguides fed, the frequency of feeding visits per hour, and the proportion of visits which resulted in the wax pile being depleted for each species. Then, using data for the eight most frequent wax-eating species (defined as those with > 5 visits: greater honeyguide; scaly-throated honeyguide, Indicator variegatus; lesser honeyguide, Indicator minor; striped bush squirrel, Paraxerus flavovittis; African civet, Civettictis civetta; honey badger, Mellivora capensis; yellow baboon, Papio cynocephalus; Meller’s mongoose, Rhynchogale melleri) we fitted a univariate Cox’s proportional hazards survival model for arrival at wax (the event) for greater honeyguides and the other seven competitor species (pooled) using the R-package ‘survival’. The response term was time since wax placement, and a binary variable denoting whether the final visit time was unknown (right-censored events). The proportional hazards assumptions of this Cox regression model were met (checked by visual inspection of proportional hazard plots and tested using the cox.zph function in the ‘survival’ package; all p > 0.05). The results are presented as hazard ratios (HR) with corresponding 95% confidence intervals (CI).

To test prediction (ii) of hypothesis one (most important competitors are diurnal), we tallied the number of visits of species which consumed wax, and the time of day at which depletion (time when final piece of wax is consumed) occurred. To test prediction (iii) of hypothesis one (diurnal competitors consistently displace greater honeyguides from a wax resource), we first calculated the proportion of visits by diurnal competitors at wax (scaly-throated honeyguides, lesser honeyguides, striped bush squirrels, yellow baboon and slender mongoose) which were simultaneous with greater honeyguides, then calculated the proportion of greater honeyguide feeding visits which were cut short by either of these five competitor species, and the number of greater honeyguide visits which were immediately before or after competitor species (within 10 s).

To test prediction (i) of hypothesis two (greater honeyguides are the first-arriving species after the wax has been exposed), we calculated the proportion of sites where greater honeyguides arrived first, along with the previous Cox’s proportional hazards survival model of arrival at wax by greater honeyguides and competitor species. To test prediction (ii) of hypothesis two (the majority of greater honeyguide feeding events occur before major wax competitors), we first observed that the larger-bodied mammals (>1 kg) appearing in the images were honey badger, yellow baboon, Meller’s mongoose and African civet. Preliminary observations suggested that although civets frequently visited wax sites, they were messy and often left substantial wax available, only depleting all available wax at 2 of 26 sites. For this analysis we therefore defined the major wax competitors as yellow baboon, honey badger and Meller’s mongoose, and counted the number of greater honeyguide visits to wax that occurred before or after the visits of these species, and also the number of greater honeyguide visits spent looking for pieces of wax after the wax was depleted. These counts were fitted as the response term in a generalised linear mixed effects model (GzLM) with a Poisson distribution. The number of camera trap days for each interval was included as an offset to account for variation in sampling effort, and wax placement site was included as a random term in addition to an observation level random to account for overdispersion. Similarly, we compared the feeding durations of greater honeyguides before and after the arrival of major wax competitors using a GzLM with a Gamma distribution (selected due to the data having positively-skewed errors) with time spent feeding as the response term and visit interval (before major wax competitors, after major wax competitors, after wax depletion) as the predictor, and wax placement site as the random term. For both models we report chi-squared statistics of an analysis of variance (ANOVA) between the model of interest and the null model. Assumptions of normality for both GzLMs were assessed by visual inspection of the distribution of residuals. Effect sizes (estimated marginal means) were calculated using the R package ‘emmeans’. Feeding rates for greater honeyguides were calculated by dividing the total number of feeding visits by the total number of daylight hours the wax was available for (daylength was calculated using ‘suncal’ package), over all sites. This was repeated for visits before and after major wax competitors. 

R Statistical Program is require to open .R code and .csv files.