1. Species in one ecosystem can indirectly affect multiple biodiversity components and ecosystem functions of adjacent ecosystems. The magnitude of these cross-ecosystem effects depends on the attributes of the organisms involved in the interactions, including traits of the predator, prey and basal resource. However, it is unclear how predators with cross-ecosystem habitat interact with predators with single-ecosystem habitat to affect their shared ecosystem. Also, unknown is how such complex top-down effects may be mediated by the anti-predatory traits of prey and quality of the basal resource.
2. We used the aquatic invertebrate food webs in tank bromeliads as a model system to investigate these questions. We manipulated the presence of a strictly aquatic predator (damselfly larvae) and a predator with both terrestrial and aquatic habitats (spider) and examined effects on survival of prey (detritivores grouped by anti-predator defense), detrital decomposition (of two plant species differing in litter quality), nitrogen flux and host plant growth. To evaluate the direct and indirect effects of each predator type on multiple detritivore groups and ultimately on multiple ecosystem processes, we used piecewise structural equation models. For each response variable, we isolated the contribution of different detritivore groups to overall effects by comparing alternate model formulations.
3. Alone, damselfly larvae and spiders each directly decreased survival of detritivores and caused multiple indirect negative effects on detritus decomposition, nutrient cycling, and host plant growth. However, when predators co-occurred, the spider caused a negative non-consumptive effect on the damselfly larva, diminishing the net direct and indirect top-down effects on the aquatic detritivore community and ecosystem functioning. Both detritivore traits and detritus quality modulated the strength and mechanism of these trophic cascades. Predator interference was mediated by undefended or partially defended detritivores as detritivores with anti-predatory defenses evaded consumption by damselfly larvae but not spiders. Predators and detritivores affected ecosystem decomposition and nutrient cycling only in the presence of high-quality detritus, as the low-quality detritus was consumed more by microbes than invertebrates.
4. The complex responses of this system to predators from both recipient and adjacent ecosystems highlight the critical role of maintaining biodiversity components across multiple ecosystems. 

Between January and March 2010 (35 days), we manipulated predators and detritivores within bromeliads settled within experimental cages. Each cage consisted of a rectangular wire frame (0.50m × 0.60m) covered with a white fabric mesh that allowed moderate light to pass through, fixed to the top edge of a pot into which the experimental bromeliad was planted. An emergence trap, made of an inverted plastic bottle, was placed on top of each cage to capture adults of detritivores that emerged during the experiment.

We randomly arranged these 48 bromeliads into 12 blocks of four bromeliads, with a minimum distance of 20 meters between blocks. We randomly assigned each block to the following treatments: (i) only detritivores, without predators (control treatment); (ii) detritivores + damselfly larva; (iii) detritivores + spider; (iv) detritivores + damselfly larva + spider. We used an additive design, in which the density of each predator species is held constant since this design, unlike a substitutive design, avoids confounding the absolute effect of interspecific interactions (in the predator co-occurrence treatment) with intraspecific interactions (Sih et al.1998; Romero & Srivastava, 2010). This allows us to assess predator impact regarding individual and multiple effects (i.e., considering potential interspecific interactions between spider and damselfly larvae). Prior to the inclusion of the organisms, we added portions of approximately 15 previously dried and weighed leaves of Eugenia uniflora (Myrtaceae) (mean = 0.24g; SD ± 0.002g) and of Ocotea pulchella (Lauraceae) (mean = 0.57g; SD ± 0.009g) distributed equally among the bromeliad tanks. The litter of both plant species is commonly found inside the bromeliad tanks in the study area. We consider E. uniflora to be the higher quality litter for two reasons: (1) it has higher specific leaf area than O. pulchella, (97.77 cm².g^-1 ± 27.3, Migliorini et al., 2018; and 67.9 cm².g^-1 ± 5.0, Boerger & Wisniewski, 2003, respectively), often an indicator of higher palatability (Kurokawa et al., 2010; Schädler et al., 2003), which can also increase detrital decomposition rates (Migliorini et al., 2018); (2) it was fertilized with ¹⁵N in order to trace nitrogen flux from detritus to bromeliads. 

Once the litter was added, we added ten chironomids, five culicids, three limoniids, three scirtids, and two trichopterans in each experimental plant. The abundance of detritivorous groups in this experiment was based on natural abundance in bromeliads of similar size to the experimental bromeliads. After one day, we added a larva of the damselfly L. andromache (average body length = 11.3 mm ± 0.21 mm) and/or an adult female of the spider C. demersa according to predator treatments. We simulated the natural colonization of detritivores on the 18th day of the experiment by adding a further five chironomids, two culicids, a limonid, two scirtids, and a trichopteran. The abundances used in this simulated colonization were smaller than the initial quantity to avoid competition within each detritivore group in the phytotelmata. Thus, the inclusion of new individuals maintains the abundance of organisms that died or emerged during the experiments with other living organisms in each group of detritivores (Bernabe et al. 2018). We inspected the bottle traps daily for emerging adults, collecting and fixing the adults in 70% ethanol.

At the end of the experiment, we removed each plant from the pot, counted the final number of bromeliad leaves (including new leaves), and clipped and collected three bromeliad leaves for isotopic analyses of ¹⁵N. We then dissected the bromeliad, washing the leaves and contents on a white tray with water. We collected all the detritus, separating the dead leaves according to species and drying these in paper bags. Finally, we collected and recorded all of the macroinvertebrates. To verify whether the presence of the spider interferes with the growth of L. andromache larvae, we collected all damselfly exuviae, which indicates an increase in damselfly instar.

We quantified the survival of each group of detritivores as the ratio between the abundance of survivors (i.e., remaining larvae and captured adults) and the total number of larvae used in the experiment. Thus, the survival percentage of detritivores considers the permanence of living organisms throughout the experiment, being an important approach to evaluate the role that detritivores play in the ecosystem. We estimated detritus decomposition as the percentage of original biomass lost of the two litter species, O. pulchella and E. uniflora. We measured the transfer of ¹⁵N from E. uniflora debris to the bromeliad by quantifying the δ¹⁵N of Q. arvensis leaves at the end of the experiment (Appendix B). The vegetative growth of the bromeliads was approximated as the number of new leaves produced, that is, the difference between the initial marked leaves and final number of leaves.