Ecological immunisation: In situ training of free-ranging predatory lizards reduces their vulnerability to invasive toxic prey Electronic Supplementary material (ESM) Figure 1. Patterns of toad-induced mortality of goannas across the floodplain in the first (2013/14) and second (2014/15) year of the study. (X=Longitude, Y=Latitude). We recorded 15 deaths in the first year (almost exclusively located in the southern half of the floodplain) and 17 in the second year, spread across the southern and the northern parts of the floodplain (total n=32). This reflects the south to north dispersal of the cane toad invasion front at our field site. S1: Field Methods Surveys and radiotelemetry We radio-tracked goannas from November 2013 to May 2015 (cumulative time in the field 12 months). During this period, we searched for goannas from 0600 to 1300 h each day. Goannas were hand-caught via noose and then transported back to the research station where we measured their mass and snout to vent length (SVL). We then attached radio transmitters (Holohil, RI-2B VHF radio transmitters) to their tail following methods in previous studies (1), and returned them to their initial point of capture for release within 24 h. Mortality sensors in the radio transmitters triggered after 12 h of non-activity. We attached a thermal data-logger (Thermochron TC, OnSolution Pty. Ltd.) to each transmitter, to measure ambient temperatures; these data also pin-pointed time and date of death for animals that died between field-trips. Marked goannas were radio-located at least once every two to three days on each trip. Conditioned Taste Aversion training We exposed free-ranging goannas to small live toads (30 – 70 mm snout-urostyle length [SUL], 10 to 25 g), <2% of the lizard’s body mass (criterion for non-lethality, based on laboratory trials; (2)). To reduce toxin content, we squeezed the toads’ parotoid glands prior to trials. The “teacher toad” was presented on fishing line attached to a 4-m fishing rod, with the toad attached to the fishing line via a cotton-thread waist belt (such that the toad was easily detached if seized by the lizard). We approached foraging goannas during the day and dropped the toad to the ground within a metre of the goanna’s head. If the toad was ignored, the procedure was repeated ten minutes later (up to three times each trial). We attempted to watch interactions with the toad, but from far enough away that the lizard was undisturbed. We scored whether each goanna ignored the toad, or responded by approaching it, tongueflicking, biting or eating. Training trials were concurrent with radiotelemetry, conducted opportunistically with every goanna. This study abided by strict ethical guidelines and we adopted several methods to minimise the suffering of the animals involved. Unless a toad was attacked by a goanna, it did not sustain any injuries during trials. Attacks were rapid and almost instantly fatal to the toad. Most attacks ended in consumption; the few toads that were only bitten were humanely euthanized immediately thereafter. Each toad was used in only a single trial; if not consumed, it was euthanized humanely. Prior to trials, toads were housed in high quality conditions with excellent husbandry. S2: Overall survival estimates in program MARK Based on their presence or absence on each field trip, we scored encounter histories for each individual goanna across eleven occasions (field trips) and corrected for time intervals between trips. We initially assessed the goodness-of-fit of a fully time-dependent Cormack– Jolly–Seber (CJS) model, using Test2 and Test3 of program RELEASE implemented in MARK (3). These tests did not indicate any significant lack of fit of the CJS model (P > 0.29), suggesting that the CJS model offers an appropriate framework to estimate survival and recapture rates. Therefore, we fitted eight CJS models in which survival (phi) and recapture rates (p) were either held constant, varied over time, or varied as a function of training, location at the field site (North or South) or an interaction between training and location. As the best initial model contained p(t), we did not pursue any models that included main affects and p(.). This analysis included records of all telemetered individuals regardless of their fate (i.e. records were not censored). We used Akaike Information Criterion to identify the best supported model in the MARK analysis (4) (Table 1.). This model included the training status of the goanna (trained vs. untrained), the location of the goanna (north vs. south i.e. as a proxy for toad exposure) and the interaction between training and location. Untrained goannas in the south (i.e. the area with longest exposure to toads) had the lowest rates of survival. Survival rates of trained goannas in the south were similar to those of both untrained and trained goannas in the north of the floodplain (Table 2 and Figure 2). Hence, we excluded the northern population from subsequent analyses. Table 1. Models run in MARK and their associated AICc and Delta AIC values. (T = training status, L = location on floodplain i.e. North vs South, I = training * location interaction term) Model AICc Delta AIC {Phi( T + L + I ) p(t) PIM} 521.6507 0 {Phi( T + L ) p(t) PIM} 523.60004 1.9497 {Phi( L ) p(t) PIM} 523.3116 4.6609 {Phi(t) p(t) PIM} 532.8191 11.1684 {Phi( T ) p(t) PIM} 535.2775 13.6268 {Phi(t) p(.) PIM} 535.5416 13.8909 {Phi(.) p(t) PIM} 538.5761 16.9254 {Phi(.) p(.) PIM} 543.2208 21.5701 Table 2. Overall survival estimates for trained and untrained goannas in the North and South of the floodplain. Category Annual survival SE Untrained South 0.0029 0.0028 Trained South 0.111 0.08 Untrained North 0.162 0.075 Trained North 0.192 0.18 Figure 2. Estimates of annual rates of survival of all goannas (trained and untrained) in the North and South using MARK. This analysis included all individuals regardless of their training status, tracking history or ultimate fate. References 1. Ujvari B, Madsen T. Increased mortality of naive Varanid lizards after the invasion of non- native cane toads (Bufo marinus). Herpetological Conservation and Biology. 2009;4(2):248-51. 2. Pearson DJ, Webb JK, Greenlees MJ, Phillips BL, Bedford GS, Brown GP, et al. Behavioural responses of reptile predators to invasive cane toads in tropical Australia. Austral Ecology. 2014;39(4):448-54. 3. Amstrup SC, McDonald TL, Manly BFJ. Handbook of Capture-Recapture Analysis: Princeton University Press; 2010. 4. Burnham KP, Anderson DR. Multimodel inference - understanding AIC and BIC in model selection. Sociological Methods & Research. 2004;33(2):261-304.