Taste aversion training can educate free-ranging crocodiles against toxic invaders

Apex predators play critical ecological roles, making their conservation a high priority. In tropical Australia, some populations of freshwater crocodiles (Crocodylus johnstoni) have plummeted by >70% due to lethal ingestion of toxic invasive cane toads (Rhinella marina). Laboratory-based research has identified conditioned taste aversion (CTA) as a way to discourage consumption of toads. To translate those ideas into landscape-scale management, we deployed 2,395 baits (toad carcasses with toxin removed and containing a nausea-inducing chemical) across four gorge systems in north-western Australia and monitored bait uptake with remote cameras. Crocodile abundance was quantified with surveys. Free-ranging crocodiles rapidly learned to avoid toad baits but continued to consume control (chicken) baits. Toad invasion at our sites was followed by high rates of crocodile mortality (especially for small individuals) at a control site but not at nearby treatment sites. In areas with high connectivity to other waterbodies, repeated baiting over successive years had continuing positive impacts on crocodile survival. In summary, we succeeded in buffering the often-catastrophic impact of invasive cane toads on apex predators.

Methods
This data was recorded during baiting trials run in the wild to test the effectiveness of taste aversion training with freshwater crocodiles. The data has had minimal processing but was collected and analysed as below.

Eliciting taste aversion

Baiting trials were conducted all sites in September and October of 2021. We ran a repeat baiting at Bandiln͟gan sites 1 and 2 in September 2022 (but not other sites), to explore the longevity of our intervention in a large population experiencing toads for the first time. We also wanted to clarify population dynamics between years at dry-season waterbodies in this important gorge ecosystem, which would allow us to design future conservation protocols. Trials were timed to coincide with late dry-season peak overlap between crocodiles and toads, when pools were relatively isolated. Based on our spotlight counts just prior to trials, we calculated the number of bait stations required in each area to keep the ratio of crocodiles to bait stations below 2:1, while maintaining 50–100 m distance between stations along both sides of the waterbodies. Our rationale for these calculations was twofold and based on the spatial ecology detected during spotlight surveys. Pragmatically, we aimed to maximise the number of bait stations available to crocodiles while limiting competition between animals, to ensure wide application of the baiting strategy. Ethically, we aimed to minimise the chance of single animals consuming too many baits and becoming overly ill.

At each bait station, two 90-cm metal stakes were placed 2 m apart, angled outwards at 30o, with their bases in water to prevent ant attack (following methods in Aiyer et al. [35]). Baits swung freely 100 mm above the surface of the water to be visible to crocodiles while discouraging consumption by freshwater prawns or fish. Baits were suspended by bulldog clips, allowing easy release when pulled (figure 1b).

To induce generalised aversion to cane toads, each bait consisted of the fresh carcass of an adult toad (~100 g, 70 mm) with internal organs, head and parotid glands removed to render it non-lethal [36]. To ensure the baits would elicit taste aversion, we added 5 mL of lithium chloride (LiCl; concentration 0.5 Mol [37]) via intramuscular injection into both back legs of the cane toad carcass. We honed the dosage rate through pilot studies with captive crocodiles. Generally, chemicals used to elicit taste aversion should be odourless and tasteless, such that the target aversive ‘cue’ (in this case, the scent or taste of cane toads) remains untainted [10].  Because Lithium Chloride (LiCL) is salty, it is often (in lab studies at least) administered via injection or drenching into the target animal. This was not possible in the context of our in-situ conservation study, with large crocodilians. However, crocodiles were unlikely to detect the LiCl in baits: a) because they are ambush predators, initiating fast attacks from the water then consuming toad-sized prey whole and b) they spend time in seasonally brackish environments and are accustomed to consuming prey within saline water [23]. A paired control bait (chicken neck ~40 g, 50 mm) was hung on one arm at each station on Day 1 and Day 5 of trials, so that visiting animals could choose between both bait types. This design allowed us to distinguish between satiation and taste aversion at the end of the trial.

To identify consumers of bait, we set up a remotely triggered wildlife camera (Little Acorn; model Ltl6310WmC) on the riverbank behind every second or third bait station (depending on the number of stations at each site). Cameras were set to high sensitivity, and recorded a burst of three photos every time they were triggered by motion or a thermal differential. We later reviewed photographs using IrfanView v4.62 and scored the taxa to species level if possible.

Baits were deployed from canoes for five successive nights, between 1500 and 1700h, and checked between 0600 and 0800 h the next morning. We recorded whether or not baits were eaten, and then removed all remaining baits and replaced them that afternoon (see Aiyer et al. [35] for details).

Analysis of data

For all data from baiting trials, we ran binomial Generalised Linear Mixed Models with a logit link function, using the GLIMMIX procedure in SAS v9.4 M8 (SAS Institute, Cary, NC). When comparing multiple sites, we incorporated ‘bait station ID’ nested within ‘site ID’ as a random factor in the analyses, to account for pseudo-replication and repeated measures.

To assess whether rates of toad bait consumption declined over the five-day baiting period (demonstrating aversion), our dependent variable was bait ‘Eaten/Remain’ and we ran an analysis including the independent variables of trial day (1–5), crocodile density and population connectivity (open/closed). We included interaction terms which might change patterns of bait offtake from those expected with the development of learned aversion: trial day * crocodile density (increased competition throughout the trial), trial day * population connectivity (immigration of naïve individuals during the trial), and crocodile density * population connectivity (to compare general levels of activity and bait offtake).

To compare the offtake of toad vs. control chicken baits on Day 1 and Day 5 of trials (further demonstrating learned aversion), our dependent variable was bait ‘Eaten/Remain’ and we ran an analysis including the independent variables of trial day (1–5), treatment (toad/chicken), crocodile density and population connectivity (open/closed). We included interaction terms which might influence whether baits were differentially consumed during the trial: trial day * treatment (choice over time: satiation [neither bait consumed on Day 5] vs. learned avoidance [decreased consumption of toad baits only on Day 5]), trial day*treatment*crocodile density (increased competition overriding learning on Day 5), and trial day*treatment*population connectivity (immigration of naïve individuals during the trial eating equal numbers of baits on Day 5). We clarified whether offtake of both bait types declined over time by conducting GLMMs as above, with the independent variable of trial day, analysing chicken and toad datasets separately.

To compare the response of crocodiles in Bandiln͟gan (Windjana Gorge) National Park to baiting trials conducted in 2021 vs. 2022, we pooled data together for each year (i.e., sites 1 and 2, because sites were not as discrete in 2022). Our dependent variable was bait ‘Eaten/Remain’ and we ran a full factorial analysis including the independent variables of trial day (1–5), year (2021/2022) and the interaction between trial day and year. ‘Bait station ID’ was included as a random factor in this analysis.

To investigate the effect of baiting on crocodile deaths in Dan͟ggu Geikie Gorge National Park (where we began to witness crocodile mortalities in 2018), we ran a logistic regression analysis with a binomial distribution and logit link function, to model the relative numbers of dead crocodiles (counts of carcasses) compared to live crocodiles (from population spotlighting counts; i.e. the trials/events ratio) in each site (treatment vs control) and time period. The independent variables were treatment (baited vs. unbaited control), intervention status (pre-intervention – 2018-2020/post-intervention – 2021-2022) and the interaction between the two.  Analyses were conducted in SAS 9.4 (SAS Institute, Cary NC).

To determine whether the cane toad invasion imposed size-selective mortality, we compared mean SVLs of dead crocodiles to live crocodiles. Data from live animals (n=209) were collected during monitoring programs at Bandiln͟gan (Windjana Gorge) National Park over four years pre-cane toad invasion (2014–2018). In that catch-and-release program, the crocodile population was sampled from the lower sites in the gorge extensively each year via seine netting of dry-season pools (selecting a random sample of individuals), with morphological measures taken prior to marking and release. Throughout that period the population was stable and displayed a normal size distribution (Department of Biodiversity Conservation and Attractions (DBCA), unpublished data, 2023). We compared the means using a one-way ANOVA in JMP v16 (SAS Institute, Cary, NC).