Classical island biogeography predicts that island plants should be less defended than their mainland relatives due to relaxed herbivory pressure. However, mounting evidence challenges this assumption, revealing substantial variation across systems and defence types. A more nuanced framework argues that understanding island–mainland differences requires considering the diversity of plant defensive strategies. For example, if island plants experience more variable herbivory regimes—such as episodic insect outbreaks—they may rely more heavily on inducible rather than constitutive defence traits or on tolerance strategies. In this study, we conducted two complementary greenhouse experiments to compare defence- and tolerance-related response in seven island–mainland species pairs of oaks (Quercus) spanning three biogeographical regions: Bornholm Island vs. mainland Sweden, the Balearic Islands vs. mainland Spain, and Lesbos Island vs. mainland Greece. Seedlings were exposed to controlled foliar herbivory by the generalist caterpillar Lymantria dispar, with undamaged plants serving as controls. Defence was quantified through chemical defences—total phenolics and volatile organic compounds—assessed for both constitutive levels and inducibility. Tolerance was measured as growth compensation, quantified through height regrowth following damage. Constitutive chemical defences showed limited differentiation between island and mainland plants, with one notable exception: hydrolysable tannins were consistently higher in island species across all three regions. In contrast, inducible chemical responses did not differ between islands and mainlands. Although overall tolerance at the end of the experiment was similar, we observed greater early regrowth (within the first two weeks post-damage) in island seedlings compared to mainland ones. Together, these findings demonstrate that island-associated herbivory pressures can shape both defence and tolerance strategies, but in trait-specific and temporally dynamic ways—highlighting the complexity of plant defence evolution on islands.

We conducted all analyses using R version 4.2.1 (R Core Team, 2013). We first tested insularity effects separately on constitutive levels of chemical defences and their inducibility across study regions, using the concentrations of phenolic compound groups; flavonoids, hydrolysable tannins, condensed tannins, and hydroxycinnamic acids, as well as total phenolics (the sum of all groups) as proxies. For constitutive levels, we fitted Generalized Linear Mixed Models (GLMMs) using the glmmTMB function from the glmmTMB package (Brooks et al. 2017). Fixed factors included insularity (two levels: island vs. mainland), study region (three levels: Balearic Islands vs. mainland Spain, Bornholm vs. mainland Sweden, and Lesbos vs. mainland Greece), and their two-way interaction. Only data from control plots (without herbivory) were used, and seedling height was included as a covariate. Random factors included population, oak species, block, and the interaction between insularity and block. Because model residuals were not normally distributed, we fitted GLMMs with a Gamma distribution and log link function, appropriate for continuous, non-negative, and right-skewed ecological data (Zuur et al. 2009). For condensed tannins, a Tweedie distribution was used to account for zero inflation (Dunn and Smyth 2005). The significance of fixed effects was assessed using Wald χ² tests via the Anova function from the car package (Fox et al. 2001), and post-hoc mean contrasts for significant factors were conducted using the emmeans function from the emmeans package (Lenth 2023).

To test insularity effects on the inducibility of chemical defences, we applied a bootstrap approach following Moreira et al. (2013). For each seedling in the induced treatment, inducibility was calculated as the difference between its induced value and that of each of the four control plants from the same population (one from the same maternal tree and three from other maternal trees within the population). This procedure yielded four estimates per induced seedling, which were treated as repeated measures in the analysis. When fewer than four control seedlings were available per population, we randomly duplicated values to obtain four estimates. This procedure was applied to each phenolic group as well as total phenolics. We then fitted repeated-measures GLMMs using glmmTMB with a Gaussian distribution and identity link function (Faraway 2006), including insularity, study region, and their two-way interaction as fixed factors. Seedling height and herbivory damage (to account for variation in feeding) were included as covariates. Random factors included population, oak species, block, the interaction between insularity and block, and seedling ID (the latter accounting for repeated measures). Fixed effects and post-hoc contrasts were assessed as described above.

Next, we tested insularity effects on constitutive VOC emissions using data from one block. Constitutive analyses included both quantitative (total emission amounts) and qualitative (compositional variation) assessments. For quantitative analyses, we fitted a GLMM on total VOC emissions using glmmTMB with a Gamma distribution and log link function (Zuur et al. 2009). Fixed factors included insularity, study region, and their two-way interaction, using only control plants. Seedling height was included as a covariate, and population and oak species were included as random factors. Fixed effects and post-hoc contrasts were assessed as described above. For qualitative analysis of constitutive VOCs, we performed a Permutational Multivariate Analysis of Variance (PERMANOVA) using the adonis function in the vegan package (Oksanen et al. 2010) with 10,000 permutations on Bray–Curtis dissimilarity matrices based on compound abundances, controlling for among-population variation. Differences in VOC composition between island and mainland populations for each region were then visualized using non-metric multidimensional scaling (NMDS) based on Bray–Curtis pairwise dissimilarities, implemented with the metaMDS function in the vegan package.

We then estimated VOC inducibility for each seedling as the difference between its induced value and that of the control seedling from the same maternal tree. Unlike phenolics, only one block was used for VOCs, so no repeated measures were generated. When no control was available within a population, a control from another population of the same species was randomly selected. We then fitted GLMMs using glmmTMB with a Gaussian distribution and identity function (Faraway 2006), including insularity, study region, and their two-way interactions as fixed factors. Inducibility values were standardized (mean = 0, SD = 1) to improve the numerical stability (Faraway 2006). Seedling height and herbivory damage were included as covariates, and population and oak species as random factors. Fixed effects and post-hoc contrasts were assessed as above.

Finally, we tested insularity effects on seedling regrowth capacity post damage by fitting a GLMM using glmmTMB with a Gaussian distribution and identity link (Faraway 2006), estimated by REML (Restricted Maximum Likelihood) to perform an ANCOVA on plant height. Specifically, we tested whether regrowth capacity following herbivory differed between island and mainland seedlings and whether any such pattern varied across study regions. The model included time (weeks since larval removal), insularity, study region, and their two- and three-way interactions as fixed factors, with herbivory as a covariate. Population, oak species, block, and the interaction between insularity and block, and seedling ID were included as random factors (the later accounting for repeated measures over time). We assessed fixed and random effects as described above. We focused on two key outputs: (1) a significant time × insularity interaction revealed that regrowth capacity differed between island and mainland populations over time, and (2) a significant time × insularity × study region interaction indicating that the effect of insularity on regrowth capacity varied among regions over time.