Teacher toads: Buffering apex predators from toxic invaders in a remote tropical landscape

Even after research identifies new approaches for wildlife management, translating those methods for delivery can be logistically challenging. In tropical Australia, invasive cane toads (Rhinella marina) fatally poison many anuran‐eating native predators. Small‐scale trials show that vulnerable predators exposed to small (nonlethal) toads can learn to delete toads from their diets, increasing survival of those predators after toads invade. We deployed this method in the remote Kimberley region of tropical Australia, with >200,000 eggs, tadpoles, or metamorph toads released in advance of the expanding toad invasion front. Remote camera surveys before and after “teacher toad” deployment showed that large monitor lizards (Varanus panoptes) were almost extirpated from control plots but remained abundant in treatment plots, indicating broad‐scale success of this novel intervention.

Conditioned taste aversion (CTA) is a form of learning that has been used to discourage predation on domestic stock by canid predators (Ellins et al., 1977) and induce avian predators to stop consuming the eggs of endangered birds (Tobajas et al., 2020).The approach exploits a predator's ability to learn from negative feeding experiences: a novel prey type whose consumption induces nausea is less likely to be eaten again.To render a prey type undesirable, we can add nausea-inducing chemicals as in the examples above.Another way to induce aversion to a toxic invasive species is to expose naive predators to small (less toxic) individuals of the novel prey type such that the aversioninducing stimulus not only resembles the lethal stimulus, but also behaves in similar ways (Ward-Fear et al., 2017).We have used this approach to buffer native predators against the invasion of toxic cane toads (Rhinella marina) in tropical Australia.By radio-tracking individual predators, we have shown that exposure to juvenile cane toads can teach marsupials (Llewelyn et al., 2010;O'Donnell et al., 2010;Webb et al., 2008) and reptiles (crocodiles- Somaweera et al., 2011;lizards-Price-Rees et al., 2011, 2013;Ward-Fear et al., 2016) to avoid toads large enough to be fatal if ingested.
Demonstrating the potential utility of CTA to buffer toad impacts is only the first step.To be useful for management, we need to roll out the method in the field and at a larger spatial scale: a daunting logistical challenge in a complex tropical area such as the Kimberley region of Western Australia.To take on this challenge, we assembled a coalition of stakeholders including researchers, wildlife management agencies, nongovernmental organizations, private landowners, and indigenous groups (see www.canetoadcoalition.com) in the Kimberley region of northwestern Australia.We deployed live juvenile "teacher toads" ahead of the invasion frontline to elicit taste aversion in yellow-spotted monitors (Varanus panoptes).By preventing local extinctions of an apex predator, we demonstrate the promise of this technique for conservation challenges globally.

Study site and species
The Fitzroy Valley (18 The yellow-spotted monitor (V.panoptes, Varanidae) is among the largest lizards worldwide (to 2 m long, 7 kg; see Figure 1b), and is most active in the wet season (November-April; Christian et al., 1995).The broad diet of these lizards includes amphibians (Shine, 1986;Ward-Fear et al., 2020).Yellow-spotted monitors are an important cultural species and traditional food item for indigenous people (Ward-Fear et al., 2016) and play major ecological roles (Doody et al., 2017;Pettit et al., 2021b).
Cane toads (Rhinella marina) are large bufonid anurans native to South America (Figure 1c).After their release in northeastern Australia in 1935, these amphibians spread across the tropics (Urban et al., 2008) into the Kimberley (Shine et al., 2018).The spread of invasive cane toads has devastated populations of yellow-spotted monitors, with decline of >90% due to ingestion of lethally toxic toads (Pettit et al., 2021a;Figure 1d).

Experimental design
To determine whether broad-scale deployment of aversion-inducing stimuli to native species would buffer the impact of toads, we released small cane toads in soon-to-be-invaded areas and used remotely triggered camera traps to measure abundances of monitors therein.
We used a multiple before-after control-impact (MBACI) design based on a random allocation of four control sites (no toads deployed) and three treatment sites (Figure 1a).All were near riparian zones (separated by 3-15 km), the preferred habitat of yellow-spotted monitors and adjacent to water sources where toads could survive when released.Our study ran from March 2018 to March 2021.
To encompass seasonal variation in monitor densities and toad impacts, we conducted two annual surveys, in the early wet season (November-December) and the late wet season (March-April), guided by the onset and duration of monsoonal rains.

Predator surveys
We surveyed monitor populations using remote infrared and motion-detecting cameras (model Ltl6310WmC; Little Acorn).Each site had seven cameras beside a waterway, 150-200 m apart, and secured 1 m above the ground while facing downward at 45 • toward a punctured can of sardines in oil.We ran each camera survey for 10 days.We analyzed camera trap images in IrfanView v4.62 to identify species, and individual yellow-spotted monitors during each trapping session (but not between years), based on size, color, and pattern (see Supporting Information).

"Teacher toad" deployment
Shortly before toads arrived in an area and after wet season rains had created pools of water, we deployed captive-bred "teacher toads" as either metamorphs (10 mm long, 0.1 g) or eggs and early-stage tadpoles.We deployed 1000 metamorph "teacher toads" per treatment site in the late wet season of 2018, and approximately 75,000 late-stage eggs or early-stage tadpoles per treatment site across the wet season of 2019.Where possible, we revisited sites multiple times in the following 30 days to confirm whether or not tadpoles or metamorphs had survived the translocation (see Supporting Information).

Data analysis
To conform to the design of MBACI, our initial surveys were done before toad arrival (March and November 2018) and continued after toad arrival (March and December 2019; March and November 2020; March 2021).We modeled changes in the proportion of sites occupied by yellow-spotted monitors using a single-species multiseason occupancy model including site occupancy, colonization, and extinction probabilities over the four wet seasons (and seven trapping sessions) of surveys using PRESENCE v2.13.46 (MacKenzie et al., 2003).We incorporated sitespecific covariates of treatment ("teacher toads" added vs. control) and survey-specific covariates of "season" (early [November] vs. late [March-April] wet season) and the cumulative wet season rainfall prior to each trapping session.
We recorded the abundance of monitors at each site as the total number of individual lizards (not sight visitations) captured on remote cameras per trapping session.
To determine which variables influenced lizard abundance, we ran a generalized linear mixed model (GLMM) with a negative binomial distribution and log link function.Our dependent variable was the number of lizards "captured" per trapping session, at each site.We included the independent factors of "season" (early vs. late), wet season "rainfall" prior to the trapping session (mm), and the interaction between the two, as well as "toad invasion status" (before/after toads), "treatment" ("teacher toads" vs. control), and the interaction between the treatment and toad invasion ("impact: before/after"; i.e., the BACI interaction term).We incorporated the number of trap nights (log transformed) as an offset variable to account for any variation between sites and sessions.We also included "trip number" as a repeated measure and site as a random factor.See Supporting Information for detailed methodology and rationale for data analysis.

RESULTS
Over

Arrival of the cane toad invasion
Cane toads were first sighted in the Fitzroy Valley area in March 2019, becoming increasingly common over the next two wet seasons.Mean densities of toads detected on cameras per season were similar in treatment versus control sites (F 1,6 = 2.13, p = 0.2), suggesting that cane toads arrived at about the same time and in about equal numbers in control and treatment sites.

Site occupancy and extinction modeling
Overall, detection probabilities for yellow-spotted monitors averaged 95% in the late wet season and 50% in the early wet season.Detection increased after higher rainfall and because "rainfall" was a better predictor of detection, we removed "season."In our highest ranked model (with an Akaike Information Criterion weight of 79%), the probability of site colonization was equal from year to year, the probability of monitor extinction was determined by whether or not a site had been treated with "teacher toads," and detection probability increased with the amount of wet season rainfall preceding the trapping session (see Table 1 for model summary).The second-best model (∆AIC of 3.31) was the same except that colonization probabilities were allowed to fluctuate among years.Together these models (which both included treatment) had an AIC weight of 94.75%.The null models (where all processes were either held constant or fluctuated among years) had ∆AICs of 5.46 and 14.57, respectively.In the best-fitting model, the extinction coefficient predicted by treatment was −44.69, while the closest null model that did not include treatment had an extinction coefficient of −1.77.Occupancy for yellow-spotted monitors was 100% at all sites prior to toad arrival, and extinction probabilities after toad arrival were estimated as 0% for treatment sites and 50% for control sites.

Abundance of monitor lizards before and after toad invasion
In our GLMM neither season, the amount of wet season rain prior to the trapping session, or the interaction between the two predicted the number of lizards recorded at each site (p > 0.8).The interaction between treatment and toad invasion status (before/after toad arrival) was the best predictor of lizard abundance (F 1,37 = 89.588,p < 0.001) at each site, throughout the study.Sites at which "teacher toads" were deployed had more monitors remaining after the toad invasion arrived.The abundance of monitors declined at all sites when toads invaded, but that decline was ultimately transitory in all three treatment sites yet persisted at the four control sites (Figure 2).Although lizard abundances per site fluctuated across the early and late wet season trapping sessions, the late wet

DISCUSSION
Invasive cane toads have devastated populations of yellowspotted monitors across tropical Australia, with little recovery over ensuing decades (Pettit et al., 2021a).The virtual disappearance of these hitherto-abundant giant lizards has had flow-on effects to species consumed by these voracious generalist predators (Doody et al., 2017) and to ecosystem services (e.g., rates of carrion removal: Pettit et al., 2021b).The rapid and sustained declines in monitor abundance (as judged by encounter rates) at our four control sites after the arrival of cane toads accord well with reports from other sites (Brown et al., 2013;Burnett, 1997;Ujvari & Madsen, 2009;Ward-Fear et al., 2016).In contrast, populations of monitor lizards persisted at the sites where we released "teacher toads."Our intervention thus achieved its aim.The impact of "teacher toad" release was evident both from occupancy modeling (presence/absence) and from monitor abundance (counts of individuals).The multiseason occupancy model structure was well suited to our BACI design.In addition to dealing with the potential problem of imperfect detection (a common criticism of remote camera studies: McIntyre et al., 2020), we were able to include covariates that might influence rates and trajectories of change in lizard populations after toad arrival.Treatment (or lack thereof) was the best predictor of monitor population extinction.Although it is difficult to interpret individual covariates within dynamic occupancy models, extinction coefficients were substantially higher in models that included treatment as a factor (−44.7) than in null models (−1.7).Nonetheless, we note that relying solely on occupancy modeling to detect a treatment effect is weakened by seasonal differences in detection coupled with our relatively small sample size (seven sites and seven trapping sessions).
The impact of "teacher toads" was also clear in the significant interaction between treatment and toad presence in the GLMM analysis.That is, monitor lizards were abundant in all sites prior to toad arrival, but remained relatively common postinvasion only in treatment areas.Toads arrived at the same time in control and treatment sites, so the difference in resilience of monitor populations cannot be attributed to invasion chronology.Could abiotic factors have influenced our measures of lizard abundance and site occupancy?Shifts in detectability are driven by seasonal shifts in activity of monitors, driven in turn by the timing of onset, intensity, and duration of wet season rains (Christian et al., 1995;Shine, 1986;Ward-Fear et al., 2020).To control for variation in rainfall across the four wet seasons of our study, we surveyed at the same points (early and late wet season) each year, guided by environmental characteristics rather than calendar months.We also incorporated season and cumulative rainfall as covariates into our analyses.Although our occupancy analysis showed that wet season rainfall influenced detection (presence/absence), recolonization probabilities were consistent among years despite fluctuating rainfall.Our GLMM showed no effect of rainfall on abundance (despite fluctuations in the numbers of lizards recorded in early vs. late season trapping sessions).Instead, the strongest driver of lizard abundance was the "treatment × toad invasion" interaction term.After toads arrived, declines in monitor populations at our control sites were sustained across all future trapping sessions regardless of season.As a result, the effects of season and rainfall on lizard abundance became irrelevant.
Camera trapping presents a cost-effective way to document declines in wildlife populations (including with yellow-spotted monitors: e.g., Doody et al., 2023), but such measures can be misleading if potential confounding factors are not considered.Our use of counts of individuals as abundance indices is appropriate because (a) we used the same mechanical methodology, unlikely to change in effectiveness over time and (b) we surveyed the same seasonal periods, habitat, and sites.Importantly, we compared control and treatment sites that were surveyed concurrently using the same methods.Any changes due to environmental conditions or animal habituation to camera traps should have impacted capture rates equally across sites.
Ideally, we would have monitored the survival of individual lizards across the study (as in Ward-Fear et al., 2016), but were unable to reliably recognize individual animals over successive years.Attempting to assign individual IDs to lizards across years would have introduced a high level of subjectivity and a high error rate.Radio tracking would provide a way to obtain long-term data on individuals but was not feasible in our study both for logistical reasons (seasonal inaccessibility) and ethical reasons (rapid lizard growth requires frequent adjustment of the transmitter harness).High levels of predation by pythons (Ward-Fear et al., 2018) mean that most monitors present in the fourth season would have been the descendants of survivors, not the original animals.Hence, we focused on population abundance over time rather than on individual survival.This kind of compromise (in this case, lack of precision about individual IDs) is almost inevitable when one tries to scale up an experimental study to a broader scale deployment.
Although we cannot document the fates of individual lizards, our results provide compelling evidence that deployment of "teacher toads" facilitated the persistence of populations of yellow-spotted monitors.This conclusion accords well with a previous study demonstrating the ability of taste-aversion training to enhance survival rates of individual monitor lizards following the invasion of toxic cane toads (Ward-Fear et al., 2016).
We initially selected four treatment sites but our attempts to deploy toads failed at one site (Site 4), suggest-ing that "teacher toad" methods need to be fine-tuned to local habitat features.As our main aim was to test whether successful deployment of "teacher toads" would influence postinvasion persistence of monitor populations we did not include Site 4 in our main analysis.Unsurprisingly, inclusion of Site 4 data changed the results of our analyses: the order of rankings in the occupancy model shifted, bringing the null model closer to the treatment model (∆AIC 1.75), and our GLMM results were no longer significant (see Supporting Information).Suitability of sites may be difficult to assess a priori and pragmatically, "teacher toad" deployment may be best suited to larger ephemeral or permanent waterbodies.In sites where all waterbodies are unstable (i.e., rapidly ephemeral); and thus, unsuitable for "teacher toad" deployment, long-term effects of toad invasion may be buffered by immigration of predators from nearby areas where "teacher toad" deployment is feasible.
We employed two different methods to elicit taste aversion.We released metamorph toads in the first year, because this was done too late in the season for larvae to complete development before waterbodies dried out.However, releases commenced early in the following wet season, allowing us to use eggs and tadpoles.We were thus able to trial the production of multiple forms of taste-aversion stimulus.Releasing tadpoles has practical advantages (it removes the need to raise metamorphs prior to release) as well as disadvantages (no information about the numbers of metamorphs emerging).Future research could clarify the rates of growth and dispersal of toads versus tadpoles and their relative effectiveness in inducing taste aversion in vulnerable predators.
In summary, our study offers an unusual example of a management strategy evolving from detailed research but then upscaled to delivery across multiple predator populations.Previous attempts to protect vulnerable wildlife have focused on culling cane toads directly (Shine et al., 2018).Such methods rarely have long-term impacts; even a few surviving toads may be enough to fatally poison all the apex predators in an area (Shine, 2018).In contrast, deploying "teacher toads" can buffer predator populations over a large area, and because the main risk to predators comes when the invasion front (dominated by large adult toads) first arrives, a single such deployment may allow subsequent persistence of the predator population (Ward-Fear et al., 2016).That is, after cane toads begin breeding at a newly colonized site, the abundance of small (nonlethal) metamorphs adds "teacher toads" to the system every year, without requiring additional releases.Future work could explore how long aversion persists, whether predator populations survive long term after a single deployment, and how rapidly predators recolonize control sites adjacent to treated areas.
Releasing large numbers of an ecologically damaging invasive species in order to buffer the impact of that species on wildlife sounds highly counterintuitive.Our initial proposals to deploy "teacher toads" met with widespread opposition from community groups, an opposition overcome over a decade-long period by an intensive publiceducation campaign, field studies demonstrating proof of concept, and the large-scale involvement of diverse conservation and management agencies and indigenous rangers in those field studies (Shine, 2018;Ward-Fear et al., 2016, 2020).
More broadly, the results of this study and our earlier research on taste-aversion techniques (e.g., Ward-Fear et al., 2016, 2017) support the plausibility of manipulating the behavior of vulnerable native species to buffer them against the impacts of invasive taxa (Greggor et al., 2019).Such behavioral interventions can have ethical and logistical advantages over traditional methods of wildlife management based on the often-impossible aim of eradicating invasive species (van Dooren et al., 2023).

A C K N O W L E D G M E N T S
This project began with the knowledge and of Bunuba Elder, M. Aiken.We thank the Bunuba people, A. Rethus, C. Forward, and other DBCA staff for their logistical support, and M. Elphick for manuscript preparation.The work was funded by the Australian Research Council (LP170100013) and wider Cane Toad Coalition: Department of Biodiversity Conservation and Attractions (DBCA), Australian Wildlife Conservancy, World Wide Fund for Nature Australia, Kimberley Land Council, Natural Resource Management Rangelands, Dunkeld Pastoral, and Matsos Brewery.This study was approved by the Animal Ethics Committee of Macquarie University (2019/025-15) and conducted under research permit number FO25000052-2 from DBCA.

C O N F L I C T O F I N T E R E S T S TAT E M E N T
The authors declare no conflicts of interest.

F
I G U R E 1 (a) Map of our study area with treatment and control sites, and inset location of the Kimberley region in northwestern Australia, (b) the yellow-spotted monitor (Varanus panoptes) is northern Australia's largest lizard, (c) the toxic cane toad (Rhinella marina), an invasive species from South America, introduced as a biocontrol in the 1930s, (d) yellow-spotted monitors are fatally poisoned by cane toads.Photo credits: (b) J. Farquhar, (c) and (d) G. Clarke.
Supporting datasets for the study are available at the Dryad data repository via DOI: 10.5061/dryad.wm37pvmvxO R C I D Georgia Ward-Fear https://orcid.org/0000-0002-4808-1933R E F E R E N C E S Summary of model selection in our multiseason dynamic occupancy model.Models incorporated multiple processes.Initial occupancy (i.e., before toads arrived; psi) was fixed, estimates of colonization (gamma) were held constant or varied from year to year, extinction probabilities (eps) varied as a function of the treatment status of a site (i.e., treatment with "teacher toads" or control sites without), and detection probability was either held constant or varied with the cumulative wet season rain, pretrapping session (in mm).−LogLike is the likelihood of the model, K is the number of parameters in the model, ∆AIC values of 0-2 provide substantial support, 4-7 considerably less, and >10 none.The AIC weight of any model depends on the entire set of candidate models and varies from 0 (no support) to 1 (full support).
(Christian et al., 1995) is the most reliable activity period for yellow-spotted monitors(Christian et al., 1995and supported by our occupancy analysis).At the final late wet season trapping session (March 2021, 3 years post-toad), capture rates at treatment sites were 35%, 57%, and 140% of initial late wet season values, prior to toad arrival (March 2018).Conversely, capture rates at control sites were 6%, 20% and at two sites 0% (i.e., no monitors recorded) of initial pretoad values (March 2018).