Table S1 - Stress-Induced Tradeoffs in a Free-Living Lizard across a Variable Landscape: Consequences for Individuals and Populations

Candidate model analysis summaries. Nine candidate models were tested to estimate survival probability with Program MARK. The best model tested included CORT reactivity as a covariate. In addition to CORT reactivity, the second and third best models also included bactericidal ability and individual sex as covariates, respectively. Although both of these models are within 2 ΔAICc of the best model, the AICc weight of the best model indicates it supported approximately 2–2.5 times as much by the data as these subsequent models. Further, because the confidence intervals of the beta parameters of these candidate models overlap zero, these models are most likely not statistically significant and survival estimates from these models may not be accurate. If the second and third best models are discarded for these reasons, the next best model is one which only includes bacterial killing ability as a covariate. The beta parameters of this model does not overlap zero, suggesting statistical significance. However, given that the ΔAICc between this model and the best model is greater than 2, the difference between these models is statistically significant. However, it should be noted that the survival estimates of the candidate model including bacterial killing ability as a covariate shows similar trends as those of the best model (Table 1). Thus, though our data suggests that CORT reactivity exerts the greatest influence on survival, bactericidal ability may still be a significant factor. The inaccurate estimates provided by the model that includes both CORT reactivity and bacterial killing ability may actually be an artifact of not having enough power to include two covariates into our model. Because the model which only includes individual sex as a covariate performed worse than the null model, it would seem that sex most likely is not a significant factor contributing to survival. At least one beta parameter of the covariates for all other models overlapped zero, suggesting that statistical significance for those models are unlikely. Further, given the ΔAICc of the best model and all other models, and a comparison of their AICc weights, our data strongly supports the model that includes only CORT reactivity as a covariate, suggesting that CORT response is an important driving factor in the variation in survival observed in our populations of U. stansburiana. (DOCX)候选模型分析总结。本研究共测试9个候选模型，借助Program MARK软件估算生存概率。经测试的最优模型以皮质酮（CORT）反应性作为协变量。除皮质酮反应性外，表现排名第二、第三的模型还分别纳入了杀菌能力与个体性别作为协变量。尽管这两个模型与最优模型的ΔAICc差值均在2以内，但最优模型的AICc权重表明，数据对该最优模型的支持度约为后续这两个模型的2~2.5倍。此外，由于这些候选模型的β参数置信区间均包含0，因此这些模型大概率不具备统计学显著性，基于其得到的生存估计值也可能不够准确。若基于上述原因剔除表现第二、第三优的模型，则次优模型为仅纳入杀菌能力作为协变量的模型。该模型的β参数置信区间不包含0，提示其具备统计学显著性。但由于该模型与最优模型的ΔAICc差值大于2，二者间的差异具有统计学意义。不过需要指出，仅以杀菌能力作为协变量的候选模型所得到的生存估计趋势，与最优模型的结果较为相似（见表1）。因此，尽管本研究数据表明皮质酮反应性对生存的影响最大，但杀菌能力仍可能是一项重要的影响因素。同时纳入皮质酮反应性与杀菌能力的模型所得到的不准确估计值，实则可能是由于该模型纳入两个协变量时统计效力不足所致。仅以个体性别作为协变量的模型表现甚至不如零模型，这提示性别大概率并非影响生存的显著因素。其余所有模型的协变量β参数中至少有一个包含0，表明这些模型难以达到统计学显著性。综合最优模型与其余所有模型的ΔAICc差值以及AICc权重对比结果，本研究数据强烈支持仅纳入皮质酮反应性作为协变量的模型，这表明皮质酮反应性是我们研究的U. stansburiana种群生存变异的重要驱动因子。（DOCX）