Treatment effects ----------------- Fertilization was the strongest driver of insect herbivores among the experimental treatments (see [Appendix S2][1] for complete treatment effect tables). It increased herbivore abundance in fertilized compared to unfertilized islands ([Fig. 1][2]a, F_1:23 = 40.4, p<0.001), reduced 35 evenness ([Fig. 1][3]b, F_1:23 = 17.2, p<0.001) and reduced among-islands spatial turnover in composition (***R²*** = 0.22, [Fig. 1][4]c). Most herbivore taxa present in fertilized islands were also present in the unfertilized ones (see nested pattern in [Fig.1][5]c). Therefore, the homogenization effect of fertilization on amongislands insect herbivore composition unfolded because of a filtering effect where fewer and less spatially variable, but more abundant herbivorous taxa could persist and dominate fertilized islands. Defoliation by mowing did not have detectable individual or interactive effects on herbivore richness, evenness and abundance, however it significantly affected among-islands herbivore taxa composition by supporting taxa that were otherwise mostly absent, and excluding taxa that were otherwise present in unmowed islands (***R²*** = 0.07, [Fig.1][6]d). Interestingly, neither island size nor spatial constraints had any significant direct impacts on insect herbivores. Mediated effects ---------------- The path model comparison analysis showed that the observed effects of fertilization and defoliation on insect herbivores were mainly explained by plant composition changes (Bottom-up model AICc confidence weight = 1, see [Table 1][7]) and that predator-mediated effects were negligible (ΔAICc _{Bottom-up vs. Top-down} = 106.13, [Table 1][8]). The selected bottom-up path model showed that the observed homogenization effect of fertilization on insect herbivores unfolded because nitrogen addition reduced plant richness (***γ_Rr:N*** = -0.506, [Fig.2][9], for detailed SEM model outputs see Appendix S3) and altered among-islands plant composition (***γ_Rr:N*** = 0.637, [Fig.2][10]) by supporting fewer specific forb species (mainly *Chenopodium* spp.). This, in turn, favored higher abundance and dominance of specific herbivore groups (mainly Miridae and Cicadellidae) within fertilized islands. The changes in insect herbivore composition associated with defoliation unfolded because it affected among-islands plant composition (***γ_Rc:D*** = 0.125, [Fig.2][11]), which in turn favored higher abundance of Aphidoidea and Scarabidae herbivores in mowed islands. Mainland-Islands ---------------- By contrast to the islands, insect herbivores in the untreated and continuous mainland were associated with a mixture of top-down effects from predators and bottom-up effects from the plants ([Table 1][12]). Similar to what was found in the islands, plant composition was the most important driver of herbivore composition (***β_Hc:Rc*** = 0.648), richness (***β_Hr:Rc*** = 0.455), and abundance (***β_Ha:Rc*** = 0.849). However I also found evidence for top-down effects from predators on herbivores with predator richness negatively affecting herbivore richness (***β_Hr:Pr*** = -0.349) and abundance (***β_Ha:Pr*** = -0.846). This transition from mixed top-down/bottom-up in the mainland to a solely bottom-up regulation in the fragmented and perturbed islands was associated with a parallel homogenization of the plant community ([Fig. 3][13]d), a decline in herbivorous taxa richness, and top-predator abundance ([Fig. 3][14]a-b), and a change in herbivore composition ([Fig. 3][15]c). Overall, these results illustrate that despite the potential for interactive effects of multiple stressors on each trophic level, plant-mediated processes mostly predicted the responses of herbivore abundance (***R²*** = 0.42, [Fig.1][16]) and composition (***R²*** = 0.44, [Fig.1][17]) to fertilization and defoliation. The transition from the continuous unperturbed mainland to the fragmented islands was characterized by a parallel homogenization in plant composition and a shift toward species-poor herbivore communities, dominated by specific and more abundant herbivorous taxa. [1]: https://osf.io/j9sxq/ "Appendix S2" [2]: https://osf.io/jevm6/ "Figure 1" [3]: https://osf.io/jevm6/%20%27Figure%201%22 [4]: https://osf.io/jevm6/ "Figure 1" [5]: https://osf.io/jevm6/ "Figure 1" [6]: https://osf.io/jevm6/ "Figure 1" [7]: https://osf.io/6h3uz/ "Table 1" [8]: https://osf.io/6h3uz/ "Table 1" [9]: https://osf.io/y28en/ "Figure 2" [10]: https://osf.io/y28en/ "Figure 2" [11]: https://osf.io/y28en/ "Figure 2" [12]: https://osf.io/6h3uz/ "Table 1" [13]: https://osf.io/qby29/ "Figure 3" [14]: https://osf.io/qby29/ "Figure 3" [15]: https://osf.io/qby29/ "Figure 3" [16]: https://osf.io/jevm6/ "Figure 1" [17]: https://osf.io/jevm6/ "Figure 1"