| dc.description.method | MATERIALS AND METHODS
Subjects
Experiments were completed between April – August 2023 at ‘Adventures with Elephants’, a research and tourism facility near Bela Bela, Limpopo Province, South Africa. We tested six semi- domesticated (7 to 29 years old; four females, two males). For all trials, professional elephant handlers were used to ensure the comfort and safety of the elephants. Handlers were not assigned to individual elephants, but rather rotated among the elephants within and between days. Elephants foraged free-range daily in natural highveld savanna, from which we harvested the native plant species used in our study. Handlers, full-time staff veterinarians, and senior management provide daily care and ensure proper elephant welfare practice is in place. Experiments were performed in accordance with the Animal Ethics Committees of the University of Sydney (#2022/2196) and the Australian Code for the Care and Use of Animals for Scientific Purposes.
Apparatus and Materials
We built a giant Y-maze to test whether the neighbours around a high-quality focal plant affects its susceptibility to herbivores (Fig. 1, detailed schematic Supplementary Fig. S1). The structure of the Y-maze meant that decisions made by elephants occurred at a distance (9.5 m), well beyond their own body length and the distance they could reach with their trunk (5 m). Focal plants, plant neighbours, and artificial odour vials were placed out of sight inside an enclosed annex off the side end of each arm of the Y-maze (Fig. 2a, b, detailed schematic Supplementary Fig. S2), to ensure that odour was the only available cue to elephants.
Focal plants and plant neighbours within an annex were placed on a raised shelf, 0.8 m off the ground and, using a mesh door, were caged off so elephants could smell but not access any real plant material or artificial odour vials (Fig. 2b). The bottom half of each annex was left open and accessible to elephants once it was reached. Here, we placed a clear plastic tub which held a reward sample comprising about one third of the biomass of any real plants presented on the shelf above (Fig 2b, c). All reward plants were mixed thoroughly to simulate foraging in a mixed species plant patch. Elephants were able to walk past, pick up, and consume (if they wished) these reward samples. Rewards were provided to ensure elephants were continually motivated to participate in experiments if higher-quality patches were selected, and to create a foraging cost if lower-quality patches were selected as elephants would need to sort through low-quality plant material to access high-quality plants.
The Y-maze was big enough for a large male elephant to walk through (Fig. 1, detailed schematic Supplementary Fig. S1); height 2.4 m, walkways 3.8 m wide, >1 m wider than the largest elephant used in our study. The single entrance path into the Y-maze was 3.2 m long, and each arm was 7.3 m long. The end of each arm was left open to allow elephants to exit the maze. Beyond the end of each arm, we placed 4.2 m long shade cloth screens to reduce distractions to the elephants from behind the Y-maze (Fig. 1a). To ensure the elephants could smell plants and artificial odour vials from the start of the Y-maze, we placed a desk fan in each annex, which blew odours down each arm of the Y-maze (Fig. 2c).
General experimental procedure
To test our aims, we ran several choice trials using a protocol modified from (Schmitt et al. 2018). Details on how elephants were habituated and trained to use the Y-maze are in Supplementary Note 1. Using the Y-maze we presented elephants with one option down each arm of the maze: a high-quality focal plant (i) alone or (ii) within a plant patch treatment. A patch treatment comprised either real plant material or an artificial odour vial (Fig. 2c). Elephants were asked to ‘choose’ a side of the Y-maze to walk down from the start of the maze, 9.5 m away, using olfactory information alone.
At the start of each choice trial, each elephant was instructed by a handler to stand on a mat at the start of the Y-maze and smell down each arm for ca. 2 s (Supplementary Video 1), then to ‘choose’. At that point, it walked down one arm of the Y-maze and could consume the reward if it wished. To avoid bias and odour contamination, all observers and handlers stood directly behind the elephant. When plants were replaced in a reward tub after selection, we entered the maze from the end of the Y from both sides to prevent leaving an odour trail within the maze. In addition, plants on the side of the Y-maze not selected by the elephant were also lifted and dropped back into the reward tub and mixed at the same time to ensure elephants were not using sound as a cue to choose a side.
Within a trial, each treatment was repeated 12 times per elephant. The 12 repeats were separated into four periods, each with three consecutive ‘walk-throughs’ per elephant. A period per elephant was conducted either once or twice a day, up to seven days a week. If two periods were run in a single day, periods were 6-7 h apart and interspersed by natural foraging. Within a trial, not all elephants were presented the same treatment each day. We used a random number generator to randomize the treatment each individual elephant was presented with. No elephant was presented with the same treatment across two consecutive days. The position of the two options within a patch pair in the Y-maze was randomized (by coin toss) for each elephant with two constraints. First, across three consecutive ‘walk-throughs’ a given option was on one side no more than two times to prevent elephants using memory of position as a cue. Second, across all 12 ‘walk-through’ repeats, a given option was presented half the time of on the left and half on the right positions among elephants, to balance other potential confounding effects. All experiments were run blind (plants not visible) to both elephants and handlers. Trials were only run on windless mornings and afternoons.
Plant quality ranking
Using the foraging acceptability index developed by (Schmitt 2017; Schmitt et al. 2018) on elephants at our study site, we selected three plant species to use in our study, two of which elephants selected for (Pappea capensis and Dombeya rotundifolia), and one which they avoided (Euclea undulata). We then used preference trials (Supplementary Note 2) to confirm the quality ranking of these three species: P. capensis (high quality) was more preferred than D. rotundifolia (moderate quality), and E. undulata was avoided (low quality). We used P. capensis as our focal plant.
Trial 1: Does plant odour drive patch choice hence associational effects?
To test whether plant odours underpin patch choice facilitating associational effects of neighbours on a high-quality focal plant we compared the elephant choices between (i) high-quality focal plant alone (175 g of P. capensis) versus (ii) the same (175 g of P. capensis) in one of four patch treatments: (a) no neighbour (focal plant alone as a control); mixed with 875 g of (b) the same species (focal plant + high-quality neighbours), (c) moderate-quality D. rotundifolia (focal plant + moderate-quality neighbours), or (d) low-quality (avoided) E. undulata (focal plant + low-quality neighbours).
Based on neighbour quality, we predicted that elephants would choose treatment (a) equally to the high-quality focal plant alone (since they were the same); but choose (b) and (c) more, and (d) less than the focal plant alone. From an associational effect perspective, these predictions are consistent with neighbours in (b) and (c) resulting in associational plant susceptibility to the high-quality focal plant, but neighbours in (d) providing associational plant refuge.
Trial 2: Defining odour information elephants use to recognise and select between plant patches
Here, we tested whether a subset of odours, selected to be informative for low-quality E. undulata (Orlando et al. 2022), could elicit the same patch choice decisions by elephants, as for real E. undulata neighbours, in providing associational refuge. To select the putative informative odour subset, we first quantified the complete E. undulata odour profile from ‘headspace’ VOC sampling and GC-MS analysis following (Schmitt et al. 2018), detecting 36 VOCs. Based on the method of (Orlando et al. 2022) we then used two ‘rules of reliability’ to hone in on VOCs defined in pairs. VOC pairs needed to be emitted (a) consistently (by more than 85% of plants sampled), and in (b) precise proportions (between 0.08 (high precision) to 0 (absolute precision)). From these rules, we selected five VOCs from a band of putatively informative VOC pairs (Fig. 3): γ-terpinene, limonene, β-pinene, decanol, and nonanol.
We then created three artificial odour treatments: (a) informative, (b) uninformative, and (c) flipped proportion. The informative treatment combined the five VOCs (four pairs) in correct informative proportions (based on the average) for E. undulata. The uninformative treatment combined five new VOCs (four pairs) that were detected in E. undulata but fell below our chosen threshold of reliability (sabinene, octanoic acid, benzaldehyde, undecane, and dodecane). The flipped proportion treatment inverted the relative proportions of informative VOCs within pairs. This treatment allowed us to test whether the relative amounts of informative VOCs mattered (as predicted by Orlando et al 2022), not simply their presence. VOCs in corrected identified paired proportions were then mixed into small glass amber vials (‘homeopathic vial’, West Pack, South Africa, Fig 2c, d).
Through several choice trials (Supplementary Note 3) we concluded that the VOC emission rate from 1.5 mL of informative odour mixture was comparable to the VOC emissions from 875 g of low-quality E. undulata as used as low-quality neighbours in Trial 1 (treatment (d)), in eliciting the same response by elephants. A single odour vial of 1.5 mL was therefore used in treatments in Trial 2.
To test if artificially-synthesized odours were informative to elephants, and could influence patch choice decisions, we compared elephant choices between (i) a high-quality focal plant alone (175 g of the high-quality plant P. capensis) versus (ii) the same (175 g of P. capensis) in one of four patch treatments: (a) no neighbour (focal plant alone as a control), (b) mixed with 875 g of low-quality E. undulata (focal plant + low quality neighbours), or next to (c) an informative odour vial, (d) a flipped proportion odour vial, or (e) an uninformative odour vial (as per Figure 2c).
Based on neighbour quality, we predicted that elephants would choose treatment (a) equally to the high-quality P. capensis focal plant alone (since they were the same). If informative odour vials were informative of low-quality E. undulata to elephants, we predicted that elephants would choose treatments (b) and (c) equally (i.e., the patches would be avoided). If the ‘rules of reliability’ (Orlando et al. 2022) are correct, we predicted that treatments (d) and (e) should not be informative to elephants of E. undulata, and therefore should be chosen randomly.
Statistical analysis
We focused on the probability of patch treatment options being chosen (1,0) over a high-quality focal plant alone. We used a generalized linear mixed model with a binomial distribution and logit link function, with Treatment as a fixed effect and individual elephant as a random factor (‘lme4’ package in R (Bates et al. 2014)). Based on AIC model comparison (Burnham & Anderson 2003), Position of treatment in the Y-maze (left, right) and Test Number (1-12) were excluded from final models.
When Treatment was significant, we compared among treatments by performing multiple pairwise comparisons with the Tukey–Kramer adjustment (indicated by alphabetical superscript in figures) (‘dplyr’ package in R (Wickham et al. 2022)). Within each treatment, we also tested whether the choice of the Treatment option was significant by testing whether the probability was significantly different from 0.5 (specifically, whether the log of the odds ratio was significantly different from zero) (‘emmeans’ package in R (Lenth 2022)). All analyses were performed using R Statistical Software (version 4.2.0; R Core Team, 2022). Data were plotted using R (‘ggplot2’ package; (Wickham, Chang & Wickham 2016).
Trial S.1
S.1.1 MATERIALS AND METHODS
To compare how many artificial odour vials were required to influence plant patch selection by elephants, we compared elephant choice between (i) a high-quality focal plant alone (175 g of the high-quality plant P. capensis) versus (ii) the same (175 g of P. capensis) in one of four patch treatments: (a) no neighbour (focal plant alone as a control), (b) mixed with 875 g of low-quality (avoided) E. undulata (focal plant + low-quality neighbours), or next to (c) five informative odour vials, and (e) one informative odour vial. All informative odour vials were mixed to a total 3 mL volume.
Trial S.2
S.2.1 MATERIALS AND METHODS
To compare informative odour vial to real E. undulata VOC emission rates, we compared elephant choice between (i) a high-quality focal plant (175 g P. capensis) mixed with 875 g of real low-quality E. undulata, versus (ii) five patch treatments: (a) 175 g P. capensis mixed with 875 g of real E. undulata (control), and 175 g P. capensis next to informative odour vials mixed to volumes of (b) 3 mL, (c) 1.5 mL, (d) 0.5 mL, and (e) 0.1 mL.
Where elephant selection of an informative odour vial treatment reached 0.5 (representing random choice), VOC emission rate between 875 g low-quality E. undulata and the informative odour vial volume was deemed comparable. | en |
