Fueling a predator death-trap: Trophic subsidies and the risk of management-induced collapse in a predator-prey system

4 1. Anthropogenic subsidies create complex ecological dynamics, yet their interaction with invasive species 5 management is poorly understood. Managing subsidies or predators in isolation risks perverse outcomes, 6 including population collapses, demanding a more holistic understanding. 7 2. We employed a Pattern-Oriented Inference framework to synthesize multiple lines of evidence for the 8 endemic Noronha skink (Trachylepis atlantica) across an archipelago. We analyzed three key patterns 9 using complementary methods: (i) population density, estimated via capture-mark-recapture and 10 negative binomial GLMs of count data; (ii) individual body size, using gamma GLMs; and (iii) injury 11 rates, via tail-autotomy analysis. 12 3. We documented a dramatic density gradient driven by predation pressure. Skink populations were over 13 three times denser on predator-light secondary islands (0.411 ind/m², 95% CI: 0.297–0.568) than in 14 predator-rich PARNAMAR (0.125 ind/m², 95% CI: 0.090–0.174). Although anthropogenic subsidies 15 boosted density by 82% in the inhabited APA (0.227 ind/m², 95% CI: 0.174–0.297), this was 16 insufficient to overcome the severe impact of invasive predators. 17 4. This created a paradox where density , a traditional population success metric, was inverse to size, an 18 indicator of individual fitness. On the main island, although the subsidized APA was denser than 19 PARNAMAR, individual condition was no better. Adult males in the APA were as small as those in the 20 lowest-density sites and significantly smaller than males on the secondary islands. Predation pressure 21 explains this gradient, with tail-loss injury rates peaking in the APA (29.3%), remaining high in PARNA 22 (25.0%), and dropping about 50% on the secondary islands (15.0%). 23 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint 5. Synthesis and applications: The convergence of these patterns supports a "predator death -trap" 24 hypothesis, where trophic subsidies in the inhabited APA fuel high skink recruitment, masking extreme 25 mortality from a subsidized predator community and other anthropogenic threats. This dynamic 26 produces high population turnover and truncates size structure by selectively removing larger, older 27 adults and may explain patterns seen elsewhere . Our findings have critical management implications: 28 removing food subsidies without concurrent, effective control of key invasive predators could trigger a 29 population collapse. We advocate integrated, multi -species management and provide robust evidence 30 for the threatened status of this endemic reptile. 31

ecological trap, invasive species, island conservation, management trap, Pattern-Oriented Inference, 32 population sink, reptile conservation, Trachylepis atlantica, trophic subsidy. 33 1. INTRODUCTION 34 Due to their geographic isolation, oceanic islands foster unique evolutionary dynamics, resulting in high rates of 35 endemism and unique community structures (Kier et al., 2009). However, this isolation also makes island biota 36 exceptionally vulnerable to anthropogenic disturbances and biological invasions, primary drivers of the ongoing 37 biodiversity crisis and have caused numerous extinctions (Spatz et al., 2017) . Understanding the complex and 38 often interacting impacts of these cumulative pressures is a central challenge for applied ecology (Maeda et al., 39 2019; Oppel et al., 2014). 40 The archipelago of Fernando de Noronha, Brazil, exemplifies this challenge. Noronha is home to the endemic 41 Noronha skink (Trachylepis atlantica), a lizard that evolved in a historically predator-free environment (Rocha 42 et al., 2009) , and an i mportant pollinator species for the native mulungu tree (Erythrina velutina). The main 43 island of the archipelago is now inhabited and hosts a rich suite of invasive species, including black rats (Rattus 44 rattus), stray and feral cats (Felis catus), tegu lizards (Salvator merianae), and the cattle egret (Bubulcus ibis), 45 all of which are documented predators of the Noronha skink (Gaiotto et al., 2020; Gasparini et al., 2007; 46 Micheletti et al., 2020) . In contrast, t he smaller, uninhabited secondary islands, while still harboring invasive 47 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint rats—except for one island, where eradication has been successful—are largely free from cats and tegu lizards 48 (Abrahão et al., 2019; Russell et al., 2018). 49 Complicating this dynamic is the direct and indirect influence of human settlement on the main island, which is 50 administered as both a protected Environmental Protection Area (APA) and a more strictly protected National 51 Park (PARNAMAR). While both regions host the full suite of invasive predators, the APA, which contains the 52 urban areas, provides anthropogenic food subsidies (e.g., refuse) that can alter resource availability for both 53 skinks and predators (Gasparini et al., 2007) . These potential benefits, however, are coupled with a suite of 54 concentrated negative pressures uniqu e to the APA, including direct persecution linked to tourism -related 55 concerns (e.g., business owners fearing that lizards may disturb visitors), alongside predation from domestic and 56 stray animals, and profound habitat modification . This creates a classic management dilemma: decisions must 57 be made in a complex system where such countervailing effects are difficult to separate , but obtaining enough 58 data to inform all parameters is logistically infeasible. For example, although a general decline in skink density 59 on the main island has been noted (AEMA, 2017), the lack of a clear understanding of the relative importance 60 of these interacting drivers has impeded the development of robust conservation strategies and has hindered 61 efforts to formally classify the species as ‘Vulnerable’ or ‘Threatened’ under IUCN criteria (IUCN, 2014). 62 This archipelago -wide gradient of invasive predator presence, coupled with differences in human influence 63 between management zones, creates a natural experiment for assessing their cumulative impacts, which we use 64 to address this challenge directly. We adopt a Pattern-Oriented Inference (POI) framework —a novel approach 65 for synthesizing the types of disparate datasets common in applied ecology — inspired by the Pattern-Oriented 66 Modelling approach (Grimm et al., 2005) , which uses the principle that an explanatory hypothesis gains 67 substantial credibility if it can simultaneously explain multiple, independent patterns observed at different scales 68 or levels of organization. By integrating multiple lines of evidence and focusing on a suite of robust patterns—69 (1) population density, (2) individual body size, and (3) rates of injury— we test the overarching hypothesis that 70 the interplay between invasive predators and human activities transforms inhabited areas into a ‘predator death-71 trap,’ which in turn generates a high-turnover population dynamic that functions as an ‘invisible sink’. Our goal 72 is to demonstrate that even with inherent data limitations, a structured synthesis can provide a clear understanding 73 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint of system dynamics and generate robust recommendations to guide urgent management decisions while avoiding 74 perverse outcomes, such as intensifying predation pressure and causing a collapse of the Noronha skink 75 population. Ultimately, such a framework can provide a strong e vidence foundation for the effective wildlife 76 management. 77 2. METHODS 78 To untangle the complex interactions between invasive species, human activities, and skink populations, we 79 adopted a Pattern-Oriented Inference (POI) framework. This approach, inspired by Pa ttern-Oriented Modeling 80 (Grimm et al., 1996, 2005; Grimm & Railsback, 2012), uses multiple, independent patterns observed in a system 81 to make robust inferences about underlying ecological processes, even in the face of uncertainty. We focused on 82 three key patterns: population density, individual body size, and non -lethal injury rates, wh ich were analyzed 83 across three distinct management zones within the archipelago, representing a gradient of invasive species 84 pressure and human influence. 85 2.1 Study Area and Design 86 The study was conducted in the Fernando de Noronha archipelago (3°51'13.71"S, 32°25′25.63"W), a Brazilian 87 federal territory located in the Atlantic Ocean (Figure 1). The archipelago encompasses a total terrestrial area of 88 18.22 km², dominated by the main island (16.89 km²) and a series of smaller secondary islands (1.33 km²). The 89 climate of Fernando de Noronha is consistently warm, with sea surface temperatures avera ging 27 °C and 90 atmospheric temperatures generally between 25 and 31 °C. Annual precipitation is approximately 1,400 mm, 91 falling predominantly from January through July, while August to December constitutes the drier period 92 (WeatherSpark, 2024) . Original v egetation on the archipelago is classified as Seasonal Deciduous Forest, 93 showing marked contrasts between the wet and dry seasons (Teixeira & Linsker, 2003). 94 The main island—the only inhabited island—is divided into two contiguous federal protected areas that form 95 the basis of our study design. The first is PARNAMAR, the strictly protected Marine National Park (IUCN 96 Category II), which covers approximately 70% of the island’s terrestrial area (11,82 km 2). This site is 97 uninhabited, lacks direct anthropogenic subsidies, but contains the full suite of key invasive predators: cats (Felis 98 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint catus), tegus, and rats, as well as egrets . The second is the APA, the inhabited Environmental Protection Area 99 (IUCN Category V), which covers the remaining 30% of the island (5,07 km2). In 201 7, this site supported a 100 resident population of approximately 2,900 people and attracted significant tourism. Like PARNAMAR, it 101 contains the full predator suite with a higher free ranging cat density (Dias et al., 2017) , but it is additionally 102 characterized by substantial human presence, including urban development, habitat modification, and th e 103 availability of anthropogenic subsidies. A third site was defined to represent a "baseline" environment. These 104 are the ‘secondary islands’—a group of small, uninhabited islands that are largely free of cattle egrets, invasive 105 cats and tegu lizards. While all except one (Ilha do Meio) also host invasive black rats, all secondary islands lack 106 the direct human pressure and subsidies found in the APA. 107 2.2 Data Collection and Analysis 108 To investigate our three core patterns, we employed distinct analytical approaches tailored to the available data 109 for each pattern. All statistical analyses were performed in R version 4.4.2 (R Core Team, 2024) . The full 110 workflow including all data and code—models, parameterization, outputs, and diagnostics—are freely available 111 on GitHub (see Data Availability Statement) and provided as an annex to this manuscript (Appendix A). 112 Pattern 1: Population Density 113 We estimated skink density on the three sites using two complementary methods. First, we used point count data 114 collected at standardized survey locations with a 2 -meter radius (area ≈ 12.57 m²) to model variation in skink 115 density on PARNAMAR, APA and secondary islands. We fitted a negative binomial generalized linear model 116 (GLM) using the MASS package in R (Venables & Ripley, 2002), with skink counts as the response variable. 117 The model included two fixed-effect predictors: ‘invasive species presence’ (InvasiveSpeciesPresence), a binary 118 factor indicating the presence of the full invasive predator suite versus a reduced suite on the secondary islands, 119 and ‘food supplementation’ (FoodSupplementation), a binary factor representing the presence or absence of 120 anthropogenic subsidies. To account for the size of the survey area and express the results in terms of density, 121 we included the natural logarithm of the survey area as an offset term. We explored two alternative model 122 formulations to test the robustness o f our results. These included (i) a model with an interaction term between 123 ‘invasive species presence’ and ‘food supplementation’ , and (ii) a generalized linear mixed -effects model 124 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint (GLMM) with a random intercept for study site to account for potential spat ial grouping. Model comparison 125 using Akaike Information Criterion (AIC) confirmed that the simpler GLM provided the best fit to the data, as 126 the interaction term was not estimable due to data sparseness and the random intercept in the GLMM showed 127 negligible variance. We therefore selected the main effects GLM for inference and assessed its adequacy using 128 simulation-based residual diagnostics via the DHARMa package in R (Hartig, 2024), which included checks for 129 overdispersion, zero-inflation, and uniformity in residual distributions. 130 Second, for a more detailed assessment of skink density on the main island, we conducted a capture-mark-131 recapture analysis within the APA and PARNAMAR sites. We applied a Poisson-log normal mark-resight model 132 using the RMark package in R (Laake, 2013), which allowed joint estimation of the resighting probability (alpha) 133 and the super -population size ( U) at each site. We constructed a candidate set of five biologically motivated 134 models to test the influence of site and sampling effort on detection and population size parameters. Models were 135 ranked based on AIC corrected for small sample sizes (AICc), and the model with the lowest AICc was selected 136 for inference. 137 Pattern 2: Individual Body Size 138 To investigate impacts on individuals body size, we analyzed head length as a stable morphometric trait. Unlike 139 variables such as tail length or body mass, which may vary due to injury or recent feeding, head length reflects 140 skeletal structure and is not affected by tail loss or transient weight changes. It also provides a more reliable and 141 consistent measure than snout–vent length, which can be more error-prone due to curvature or movement during 142 handling. Given that head length is a continuous and strictly positive variable, we fitted a gamma generali zed 143 linear model (GLM) with a log link function. An initial model tested for main effects of sex and site on head 144 length. To assess whether environmental conditions affected males and females differently, a second model 145 included an interaction term ( sex × site). Model fit was evaluated using diagnostic plots, including residuals 146 versus fitted values and Q-Q plots, to verify assumptions of the gamma distribution. Estimated marginal means 147 and post-hoc pairwise comparisons among group levels were performed wit h Tukey-adjusted p-values were 148 computed using the emmeans package (Lenth, 2025). 149 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint Pattern 3: Injury Rates (signs of tail autotomy) 150 To assess variation in predation pressure across regions, we quantified the frequency of tail autotomy (i.e., tail 151 loss or signs of regrowth) . While intraspecific aggression is a known cause of tail autotomy in this species 152 (Gasparotto, 2021), a difference in prevalence of injury across sites can serve as a robust proxy for sublethal 153 predation pressure. For each individual, the presence or absence of autotomy was recorded, and a contingency 154 table was constructed to summarize the number of inju red versus uninjured individuals across the three sites. 155 We then used Pearson’s Chi -squared test, implemented in base R (R Core Team, 2024) , to test whether the 156 proportion of individuals with autotomy differed significantly among sites. Given the potential for low statistical 157 power due to limited sample sizes, we conducted a detailed post -hoc power analysis to formally evaluate the 158 sensitivity of our Chi -squared test. This analysis, using the pwr package in R (Champely, 2020), included: (i) 159 calculating the achieved power of the 3x2 omnibus test based on the observed effect size (Cohen’s w); (ii) 160 estimating the total sample size required to achieve 80% power; (iii) calculating pairwise power for two -161 proportion tests with unequal sample sizes; and (iv) determining the minimum detectable differ ence (MDD) at 162 80% power for key contrasts involving the smallest sample sizes. To validate these analytic results, we also 163 performed a simulation-based power analysis with 5,000 replicates. 164 3. RESULTS 165 Our analyses revealed three distinct patterns in the Noronha skink populations converging to one theory: the 166 complex interplay between invasive predators and anthropogenic subsidies. 167 Pattern 1: A Density Gradient Driven by Invasive Predators and Food Subsidies 168 The skink population density varied significantly across the three sites. For the negative binomial generalized 169 linear model of point counts, residual diagnostics confirmed a good model fit, with simulated residuals closely 170 following expected distributions and no evidence of systematic deviation, overdispersion, or outliers. The model 171 explained a modest but reasonable proportion of the variance (Nagelkerke’s R² = 0.17), as is typical for 172 ecological count data. The model estimated the highest density on the secondary islands (0.411 individuals/m², 173 95% CI: 0.295 –0.573) and significantly lower densities on the predator -rich main island, with the subsidized 174 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint APA (0.204 ind/m², 95% CI: 0.153–0.273) having a higher density than the unsubsidized PARNAMAR (0.156 175 ind/m², 95% CI: 0.115–0.213), although not statistically significant (Figure 2). This pattern was, however, further 176 clarified by the more detailed capture-mark-recapture analysis. 177 The top-ranked capture-mark-recapture model, which included an effect of site on super -population size ( U), 178 had the lowest AICc (855.39; ΔAICc = 1.80 compared to the next best model; Appendix A, p. 40). Estimated 179 population sizes were significantly lower in PARNAMAR colonies compared to those in the APA (β = -2.00, p 180 < 0.001), despite a significantly higher observation probability for PARNAMAR (β = 0.41, p < 0.05). Together, 181 these results establish a clear pattern, where secondary islands present a significantly larger population density 182 as APA, which in turn, presents a larger population than PARNAMAR. 183 Pattern 2: Size Structure Reveals Demographic Truncation on the Main Island 184 The population's size structure did not directly correspond to the density pattern, revealing a critical discrepancy 185 on the main island. While the APA supported a higher density than PARNAMAR, its adult males were no larger, 186 pointing to differential mortality pressures (Figure 3). A gamma generalized linear model that included an 187 interaction between sex and site provided a better fit to the head length data (AIC = 24.116) than a model with 188 only main effects (AIC = 25.051). The sex × site interaction term was significant (sexM:sitePARNA, p = 0.0285), 189 indicating that differences in size between sites were primarily driven by changes in the male population. Model 190 diagnostics confirmed data quality with no outliers detected, and residuals displayed no major deviations from 191 model assumptions (see Appendix A). 192 Post-hoc pairwise comparisons (Tukey-adjusted) revealed the specific nature of this interaction (Figure 3). While 193 there were no significant size differences among females across the three sites (e.g., females on secondary islands 194 vs. females in the APA, p = 1.00; females in the APA vs. females in the PARNAMAR, p = 0.99), the effect on 195 males was pronounced. Males on the secondary islands were significantly larger than males from both the APA 196 (p = 0.0214) and PARNAMAR (p = 0.0195). There was no statistical size difference between males inhabiting 197 the APA and PARNAMAR (p = 0.9998). This establishes a clear pattern of demographic truncation, where the 198 largest size classes of males are notably absent from the main island populations, regardless of subsidy presence. 199 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint Pattern 3: Injury Rates Align with a High-Pressure Environment on the Main Island 200 Providing a mechanistic link to the observed size truncation, rates of non-lethal injury (tail autotomy) followed 201 a trend of elevated pressure on the main island. The proportion of injured individuals was highest in the APA 202 (29.3%), followed by PARNAMAR (25.0%), and was lowest on the Secondary Islands (15.0%). Despite this 203 clear directional pattern, the overall difference was not statistically significant (Pearson’s Chi-squared test: χ² = 204 1.61, df = 2, p = 0.45). Our post-hoc power analysis revealed that this was a consequence of low statistical power. 205 With our realized sample sizes, the achieved power to detect an effect of the observed magnitude was only ~19%, 206 and an estimated ~729 individuals would be required to reach the standard 80% power threshold. 207 The minimum detectable difference for comparisons involving the small secondary island sample was 32.6–34.1 208 percentage points, indicating only very large effects could be reliably detected. Together, these analyses confirm 209 our interpretation that the non-significant test result is due to statistical limitations rather than the absence of a 210 biological effect. The observed trend of higher injury rates on the main island the refore provides a final, 211 corroborating line of evidence that aligns with the body size data and supports the hypothesis that skinks, 212 particularly in the APA, experience a higher rate of sublethal predator encounters. 213 4. DISCUSSION 214 Making robust management dec isions for species of conservation concern often requires synthesizing 215 information from multiple, imperfect sources of data. In this study, we employed a 'Pattern-Oriented Inference' 216 framework to integrate results from population density models, individual morphometrics, and rates of injury to 217 build a cohesive understanding of the complex pressures facing Trachylepis atlantica , the island endemic 218 Noronha skink lizard. The convergence of these distinct patterns provides a powerful line of evidence that 219 transcends the limitations of any single analysis. It resolves an apparent ecological paradox by revealing that the 220 inhabited area of the main island functions as a pred ator death -trap, a dynamic with profound and urgent 221 implications for the conservation of this endemic species. 222 Our results establish a clear density gradient across the sites (secondary islands > APA > PARNAMAR), which 223 can be explained by the interplay of top-down and bottom-up forces. The high density on the secondary islands 224 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint serves as a crucial baseline, demonstrating the species' potential in an environment with reduced predator 225 pressure, even without anthropogenic subsidies. Conversely, the significantly suppressed density in the predator-226 rich PARNAMAR reflects the strong top-down control exerted by the full suite of invasive predators, particularly 227 cats, tegus, cattle egrets, and rats (Gaiotto et al., 2020; Micheletti et al., 202 0). The most insightful finding, 228 however, is the intermediate density in the APA. Here, anthropogenic food subsidies appear to provide a bottom-229 up stimulus, likely boosting skink recruitment and survival enough to elevate the population density above that 230 of the unsubsidized PARNAMAR. However, while subsidy -driven increases have been observed elsewhere 231 (Plaza & Lambertucci, 2017) , our work demonstrates a critical caveat: this apparent recovery can conceal the 232 transition of a habitat into a high-mortality sink. 233 The key to deciphering the system lies in how size-selective predation resolves the apparent paradox on the main 234 island. Predation is rarely random, and traits that increase conspicuousness, such as large body size and territorial 235 behaviour, often elevate mortality risk. For instance, Bateman and Fleming (2011) demonstrated that large, 236 territorial adult m ale anoles suffered significantly higher rates of sublethal predation attempts from cats, 237 providing a direct, evidence -based mechanism for the size -selective pressure we propose is acting on the 238 Noronha skink. In the APA, this pressure is amplified because anthropogenic subsidies support high densities of 239 both skinks and their invasive predators (Dias et al., 2017), potentially increasing lethal encounters. Moreover, 240 human activity in the APA reduces the availability of alternative native prey (particularly birds), likely 241 concentrating predation pressure on skinks as one of the few abundant resources remaining. The outcome is a 242 stark pattern of demographic tr uncation, where the selective removal of the largest and oldest males leaves a 243 high-turnover population dominated by smaller, younger individuals. This directly explains why the subsidized 244 APA, despite supporting more skinks, has males no larger than those in the resource -poor, predator -rich 245 PARNAMAR. The trend of higher injury rates in the APA provides a final, corroborating line of evidence, 246 confirming that this truncated size structure is a consequence of relentless mortality, not stunted growth. 247 The con vergence of these patterns —subsidized density, a truncated male size structure, and elevated injury 248 rates—strongly supports our central hypothesis: the inhabited APA functions as a predator death-trap. Ecological 249 traps occur when an environmental cue, whic h normally indicates high-quality habitat, leads an organism to a 250 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint low-quality patch where its fitness is compromised (Patten & Kelly, 2010). In Noronha, the abundance of food 251 in the APA likely serves as an attractive cue for skinks, concentrating their population. However, these same 252 subsidies also support a high -density, subsidized predator community (Dias et al., 2017) . The result is a fatal 253 combination of high prey density and high predator density in the same location, creating an intensified predator-254 prey dynamic that transforms the APA from a subsidized refuge into a population sink. 255 Our findings reveal a critical and non -obvious management trap that could have catastrophic consequences : 256 removing food subsidies in the APA without first reducing invasive predators. Because subsidies currently fuel 257 high recruitment, eliminating them in isolation would suppress prey production while leaving a dense predator 258 community intact —thereby increasing per -capita predation risk and plausibly precipit ating a population 259 collapse, as suggested for the species in theoretical studies (Gasparini et al., 2007) . Effective conservation 260 therefore requires an integrated sequence: (i) implement coordinated, sustained control of key invasive predators 261 (e.g., cats, tegus, cattle egrets, rats); (ii) only then phase down anthropogenic subsidies via improved waste 262 management and feeding restrictions; and (iii) track explicit triggers (density, size‐structure, and injury rates) 263 under a real-time adaptive monitoring fram ework (Micheletti et al., 2025) . This strategy addresses both top -264 down and bottom-up drivers simultaneously, avoiding management-induced feedbacks and aligning actions with 265 the multi-line evidence that the APA currently functions as a predator death-trap. 266 Our integrated results provide strong, cohesive evidence that the Noronha skink is facing cryptic but severe 267 threats that warrant s re-assessing the species as at least “Vulnerable” under IUCN criteria (A2: observed 268 population decline; C1: small and declining population size; IUCN, 2014), a conclusion previously obscured by 269 its apparent abundance in the APA. This concern is further heightened by its insular life -history strategy: the 270 species exhibits a reduced, seasonal reproductive pattern compared with continental congeners (Migliore et al., 271 2017), reflecting an evolutionary history in a predator-free environment. Such traits likely lower its resilience to 272 the novel and intense predation pressures it currently faces. This case study therefore ser ves as a critical 273 cautionary tale for conservation, demonstrating the danger of relying on single metrics like population density 274 in complex, subsidized ecosystems where a high density can mask an underlying population sink and create a 275 dangerous illusion of security. These findings also inform Brazil’s invasive species strategy, highlighting the 276 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint need for integrated subsidy and predator control rather than piecemeal actions. Integrated eradication campaigns 277 on islands elsewhere , such as New Zealand (Griffiths et al., 2012) , demonstrate the feasibility of coordinated 278 invasive species and subsidies’ control (even if for public-safety or operational control), which could provide a 279 model for Noronha. 280 Our findings likely extend beyond Noronha. Similar subsidy–predator interactions have been reported for large 281 lizards (Jessop et al., 2012) and free-ranging cats (Fleming et al., 2022; Kazato et al., 2020; Maeda et al., 2019) 282 on continental areas. F ood waste has even been directly linked to elevating carrying capacity and sustaining 283 predator guilds (Almaraz et al., 2022). We suggest that the predator death -trap dynamic may be a general risk 284 wherever subsidies elevate recruitment in prey populations but simultaneously fuel predator communities. We 285 demonstrate that a Pattern-Oriented Inference framework provides the necessary tool for managers to synthesize 286 multiple data types—even those with limitations—to uncover these hidden dynamics and form a robust basis for 287 policy. Ultimately, our findings deliver an urgent warning for managers of human-altered ecosystems globally: 288 an integrated management plan that correctly identifies a nd prioritizes the primary threatening process, in this 289 case invasive predators, is essential to avoid perverse outcomes, such as triggering a preventable population 290 collapse. 291

292 Abrahão, C., Russell, J., Silva, J., Ferreira, F., & Dias, R. (2019). Population assessment of a novel island 293 invasive: Tegu (Salvator merianae) of Fernando de Noronha. Island Invasives: Scaling up to Meet the 294 Challenge, Occasional Paper SSC no. 62, 317–325. 295 AEMA, (Agência Estadual de Meio Ambiente). (2017). Lista Estadual de Répteis Ameaçádos —Pernambuco. 296 http://www.cprh.pe.gov.br/home/42872;57450;10;3351;17946.asp 297 Almaraz, P., Martínez, F., Morales‐Reyes, Z., Sánchez‐Zapata, J. A., & Blanco, G. ( 2022). Long‐term 298 demographic dynamics of a keystone scavenger disrupted by human‐induced shifts in food availability. 299 Ecological Applications, 32(6), e2579. https://doi.org/10.1002/eap.2579 300 Bateman, P. W., & Fleming, P. A. (2011). Frequency of tail loss re flects variation in predation levels, predator 301 efficiency, and the behaviour of three populations of brown anoles: TAIL AUTOTOMY AND 302 PREDATOR EFFICIENCY. Biological Journal of the Linnean Society , 103(3), 648 –656. 303 https://doi.org/10.1111/j.1095-8312.2011.01646.x 304 Champely, S. (2020). pwr: Basic Functions for Power Analysis. https://CRAN.R-project.org/package=pwr 305 Dias, R. A., Abrahão, C. R., Micheletti, T., Mangini, P. R., De Oliveira Gasparotto, V. P., De Jesus Pena, H. F., 306 Ferreira, F., Russell, J. C., & Silva, J. C. R. (2017). Prospects for domestic and feral cat management on 307 an inhabited tropical island. Biological Invasions, 19(8), 2339–2353. https://doi.org/10.1007/s10530-308 017-1446-9 309 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint Fleming, P. A., Stobo-Wilson, A. M., Crawford, H. M., Dawson, S. J., Dickman, C. R., Doherty, T. S., Fleming, 310 P. J. S., Newsome, T. M., Palmer, R., Thompson, J. A., & Woinarski, J. C. Z. (2022). Distinctive diets 311 of eutherian predators in Australia. Royal Society Open Science , 9(10), 220792. 312 https://doi.org/10.1098/rsos.220792 313 Gaiotto, J. V., Abrahão, C. R., Dias, R. A., & Bugoni, L. (2020). Diet of invasive cats, rats and tegu lizards 314 reveals impact over threatened species in a tropical island. Perspectives in Ecology and Conservation, 315 18(4), 294–303. https://doi.org/10.1016/j.pecon.2020.09.005 316 Gasparini, J. L., Peloso, P. L., & Sazima, I. (2007). New opportunities and hazards brought by humans to the 317 island habitat of the skink Euprepis atlanticus. Herpetological Bulletin, 100, 30–33. 318 Gasparotto, V. (2021). Agonistic and mati ng behaviour of the endemic lizard Trachylepis atlantica from the 319 Fernando de Noronha archipelago, Brazil. Herpetological Bulletin , 158, Winter 2021 , 16 –23. 320 https://doi.org/10.33256/hb158.1623 321 Griffiths, R., Buchanan, F., Broome, K., & Butland, B. (2012). Rangitoto and Motutapu – A Starting Point for 322 Future Vertebrate Pest Eradications on Inhabited Islands. Proceedings of the Vertebrate Pest 323 Conference, 25. https://doi.org/10.5070/V425110573 324 Grimm, V., Frank, K., Jeltsch, F., Brandl, R., Uchmański, J., & Wissel, C. (1996). Pattern-oriented modelling in 325 population ecology. Science of The Total Environment , 183(1–2), 151 –166. 326 https://doi.org/10.1016/0048-9697(95)04966-5 327 Grimm, V., & Railsback, S. F. (2012). Pattern -oriented modelling: A ‘multi -scope’ for predictive systems 328 ecology. Philosophical Transactions of the Royal Society B: Biological Sciences, 367(1586), 298–310. 329 https://doi.org/10.1098/rstb.2011.0180 330 Grimm, V., Revilla, E., Berger, U., Jeltsch, F., Mooij, W. M., Railsback, S. F., Thulke, H. -H., Weiner, J., 331 Wiegand, T., & DeAngelis, D. L. (2005). Pattern-Oriented Modeling of Agent-Based Complex Systems: 332 Lessons from Ecology. Science, 310(5750), 987–991. https://doi.org/10.1126/science.1116681 333 Hartig, F. (2024). DHARMa: Residual Diagnostics for Hierarchical (Multi-Level / Mixed) Regression Models. 334 https://CRAN.R-project.org/package=DHARMa 335 IUCN. (2014). Trachylepis atlantica: Colli, G.R., Fenker, J., Tedeschi, L., Bataus, Y.S.L., Uhlig, V.M., Silveira, 336 A.L., da Rocha, C., Nogueira, C. de C., Werneck, F., de Moura, G.J.B., Winck, G., Kiefer, M., de Freitas, 337 M.A., Ribeiro Junior, M.A., Hoogmoed, M.S., Tinôco, M.S.T., Valadão, R., Cardoso Vieira, R., Perez 338 Maciel, R., Gomes Faria, R., Recoder, R., Ávila, R., Torquato da Silva, S., de Barcelos Ribeiro, S. & 339 Avila-Pires, T.C.S.: The IUCN Red List of Threatened Species 2019: e.T120689136A134890404 340 [Dataset]. https://doi.org/10.2305/IUCN.UK.2019-1.RLTS.T120689136A134890404.pt 341 Jessop, T. S., Smissen, P., Scheelings, F., & Dempster, T. (2012). Demographic and Phenotypic Effects of 342 Human Mediated Trophic Subsidy on a Large Australian Lizard (Varanus varius): Meal Ticket or Last 343 Supper? PLoS ONE, 7(4), e34069. https://doi.org/10.1371/journal.pone.0034069 344 Kazato, K., Watari, Y., & Miyashita, T. (2020). Identification of the population source of free -ranging cats 345 threatening endemic species on Tokunoshima Island, Japan. Mammal Research , 65(4), 719 –727. 346 https://doi.org/10.1007/s13364-020-00528-5 347 Kier, G., Kreft, H., Lee, T. M., Jetz, W., Ibisch, P. L., Nowicki, C., Mutke, J., & Barthlott, W. (2009). A global 348 assessment of endemism and species richness across island and mainland regions. Proceedings of the 349 National Academy of Sciences, 106(23), 9322–9327. https://doi.org/10.1073/pnas.0810306106 350 Laake, J. L. (2013). RMark: An R Interface for Analysis of Capture -Recapture Data with MARK (AFSC 351 Processed Rep. Nos. 2013 –01; p. 25). Alaska Fish. Sci. Cent., NOAA, Natl. Mar. Fish. Ser v. 352 https://apps-afsc.fisheries.noaa.gov/Publications/ProcRpt/PR2013-01.pdf 353 Lenth, R. V. (2025). emmeans: Estimated Marginal Means, aka Least -Squares Means . https://CRAN.R -354 project.org/package=emmeans 355 Maeda, T., Nakashita, R., Shionosaki, K., Yamada, F., & Watari, Y. (2019). Predation on endangered species by 356 human-subsidized domestic cats on Tokunoshima Island. Scientific Reports , 9(1), 16200. 357 https://doi.org/10.1038/s41598-019-52472-3 358 Micheletti, T., Fonseca, F. S., Mangini, P. R., Serafini, P. P., Krul, R. , Mello, T. J., Freitas, M. G., Dias, R. A., 359 Silva, J. C. R., Marvulo, M. F. V., Araujo, R., Gasparotto, V. P. O., Abrahão, C. R., Rebouças, R., 360 Toledo, L. F., Siqueira, P. G. S. C., Duarte, H. O., Moura, M. J. C., Fernandes-Santos, R. C., & Russell, 361 J. C. (2020). Terrestrial Invasive Species on Fernando de Noronha Archipelago: What We Know and 362 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint the Way Forward. In V. Londe (Ed.), Invasive Species: Ecology, Impacts, and Potential Uses (1st ed.). 363 Nova Science Publishers. https://sites.ufpe.br/ceerma/wp-364 content/uploads/sites/73/2021/03/Micheletti_etal_2020_FNoronha.pdf 365 Micheletti, T., Mello, T. J., Verona, C., Gasparotto, V. P. O., Krul, R., Araujo, R., Sampaio, T., & Mangini, P. 366 R. (2025). Swiftly squeaky clean: Lessons learned from eradicating an overpopulati on of rats on an 367 island of constraints [(Submitted to Conservation Biology)]. Ecology. 368 https://doi.org/10.1101/2025.02.19.639051 369 Migliore, S., Braz, H., Barreto -Lima, A., & Almeida -Santos, S. (2017). Reproductive timing and fecundity in 370 the Neotropical lizard Enyalius perditus (Squamata: Leiosauridae). Acta Herpetologica, 187-191 Pages. 371 https://doi.org/10.13128/ACTA_HERPETOL-19981 372 Oppel, S., Burns, F., Vickery, J., George, K., Ellick, G., Leo, D., & Hillman, J. C. (2014). Habitat‐specific 373 effectiveness of feral cat control for the conservation of an endemic ground‐nesting bird species. Journal 374 of Applied Ecology, 51(5), 1246–1254. https://doi.org/10.1111/1365-2664.12292 375 Patten, M. A., & Kelly, J. F. (2010). Habitat selection and the perceptual trap. Ecological Applications, 20(8), 376 2148–2156. https://doi.org/10.1890/09-2370.1 377 Plaza, P. I., & Lambertucci, S. A. (2017). How are garbage dumps impacting vertebrate demography, health, 378 and conservation? Global Ecology and Conservation , 12, 9 –20. 379 https://doi.org/10.1016/j.gecco.2017.08.002 380 R Core Team. (2024). R: A Language and Environment for Statistical Computing. R Foundation for Statistical 381 Computing. https://www.R-project.org/ 382 Rocha, C. F. D., Vrcibradic, D., Menezes, V. A., & Ariani, C. V. (2009). Ecology and Natural History of the 383 Easternmost Native Lizard Species in South America, Trachylepis atlantica (Scincidae), from the 384 Fernando de Noronha Archipelago, Brazil. Journal of Herpetology , 43(3), 450 –459. 385 https://doi.org/10.1670/07-267R2.1 386 Russell, J. C., Abrahão, C. R., Silva, J. C. R., & Dias, R. A. (2018). Management of cats and rodents on inhabited 387 islands: An overview and case study of Fernando de Noronha, Brazil. Perspectives in Ecology and 388 Conservation, 16(4), 193–200. https://doi.org/10.1016/j.pecon.2018.10.005 389 Spatz, D. R., Zilliacus, K. M., Holmes, N. D., Butchart, S. H. M., Genovesi, P., Ceballos, G., Tershy, B. R., & 390 Croll, D. A. (2017). Globally threatened vertebrates on islands with invasive species. Science Advances, 391 3(10), e1603080. https://doi.org/10.1126/sciadv.1603080 392 Teixeira, W., & Linsker, R. (2003). Arquipélago Fernando de Noronha: O paraíso do vulcão . Terra Virgem 393 Editora. 394 Venables, W. N., & Ripley, B. D. (2002). Modern Applied Statistics with S (Fourth). Springer. 395 https://www.stats.ox.ac.uk/pub/MASS4/ 396 WeatherSpark. (2024). Average Weather in Fernando de Noronha (Distrito Estadual), Pernambuco, Brazil Year 397 Round. http s://weatherspark.com/y/31447/Average-Weather-in-Fernando-de-Noronha-(Distrito-398 Estadual)-Pernambuco-Brazil-Year-Round 399 400 .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint 401 402 Figure 1. Map of the Fernando de Noronha archipelago, Brazil, depicting the spatial distribution of study sites and sampling efforts in relation to key environmental drivers. The main island is divided into the protected PARNAMAR and the inhabited APA, representing zones with varying anthropogenic subsidies. Secondary islands (dark grey) represent areas with reduced invasive predator presence. Yellow circles mark point count survey locations, and green circles denote capture-mark-recapture survey locations. .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint 403 Figure 2. Skink density is strongly suppressed by the full invasive predator suite, a conclusion refined by a capture-mark-recapture analysis. The main bars show estimated marginal mean densities of the Noronha skink (individuals / m²) from a Negative Binomial GLM of broad -scale point counts. Error bars represent 95% confidence intervals. Different letters (a, b) indicate stati stically significant differences based on this GLM, which shows that sites with a reduced predator suite support significantly higher densities than main island sites. The bracket and annotation highlight a key finding from a separate, more detailed capture-mark-recapture analysis conducted only on the main island: skink population size was confirmed to be significantly higher in the subsidized APA compared to the unsubsidized PARNAMAR (p < 0.001). .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint 404 Figure 3. Demographic truncation in male skinks is correlated with higher rates of non -lethal injury. While female skinks (left panels) show no significant differences in head length across sites (mean ± 95% CI ), males (right panels) are significantly larger on the secondary islands compared to the main island sites (APA and PARNA) where we observe lower invasive species and anthropogenic pressure . This smaller male body size on the main island coincides with a h igher proportion of individuals exhibiting tail autotomy (a proxy for predation pressure), particularly in the APA. Letters indicate significant differences (p < 0.05) in head length only. .CC-BY 4.0 International licenseavailable under a (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in perpetuity. It is made The copyright holder for this preprintthis version posted August 30, 2025. ; https://doi.org/10.1101/2025.08.26.672345doi: bioRxiv preprint