Trees were selected only if ant nests were noticeably active. In each site, we quantified ant activity on the nest tree as the number of ants crossing a single point on the main trunk during 1 min. This methodology has been used in previous studies to measure overall ant activity of a nest . We then selected the six coffee plants nearest to the nesting tree, making sure they were not directly touching each other or the tree by removing branches and vines . We then randomly assigned three of the coffee plants at each site to a connection treatment and three as controls without connections, then measured ant activity on the plants by counting the number of ants passing a point on the central trunk for 1 min. Strings remained in the field for 3 days to allow for ant acclimation to disturbance and for ants to establish new foraging pathways. After 3 days, we returned to the sites and re-measured ant activity on the nesting tree and coffee plants. Observations took place between 10 a.m. and 1 p.m. and were immediately stopped as soon as it started raining, as this drastically decreases ant activity. To test how connectivity impacts potential biological control provided by ants, we added dead adult CBB onto connected and control coffee plants to directly assess ant removal rates. We collected CBB-infested coffee berries from the field, dissected them, plastic pots for planting extracted female adult CBB individuals , and placed them in the freezer for up to 24 hr, after which beetles were dead.

Three days after placing strings and after re-assessing ant activity, we placed 10 dead CBB adults on a small piece of white card on each coffee plant near the center of the trunk, left cards for 30 min, and then counted the number of CBB remaining. Cards were balanced on coffee branches and were bent slightly to keep the CBB from falling. Restricting movement of sentinel prey, either by gluing them to observation sites or by freezing them, is a common technique for assessing predator behavior . We used frozen sentinel prey to increase the availability and similarity of beetles on cards and to reduce the potential for live prey to escape from the arena. To assess whether CBB removal was due to ant activity, we monitored cards across the plot over a period of 30 min and recorded any arthropods present. Only ants were observed on the cards, indicating that these were responsible for removing the CBB. Although we acknowledge that the use of dead prey may alter ant behavior, it is already well established that A. sericeasur both antagonizes and predates live CBB in the field, and reduces CBB infestation on plants . We used dead prey in this experiment to more readily assess ant removal rates and infer that these changes translate to changes in the bio-control efficiency of this ant on live prey. Immediately following each experiment, we characterized the vegetation in each site because several different environmental factors are known to influence ant foraging in coffee systems . We measured the percentage of canopy cover , coffee plant height, and distance from each coffee plant to the central Inga nest tree.To test for statistical differences in ant activity on nest trees before and after connecting trees to coffee plants, we fit our data to a generalized linear mixed model .

We included time , canopy cover, and their interaction as fixed effects . We also modeled nest tree identity as a random effect. To assess count data , we originally fit our model to a Poisson distribution with a log link function. However, to correct for observed over-dispersion, we modified our model to a Poisson-lognormal distribution by adding a per-observation random effect term .To test for statistical differences in ant activity on coffee plants before and after establishing connections, we used a GLMM. We included time , treatment , coffee plant distance to nest tree, the interaction between time and treatment, and the interaction between time and distance as fixed effects . We also included coffee plant height and ant activity on nest tree as covariates. Random effects were modeled with plant identity nested within site to account for the block design of the experiment and to control for variation between our sites. To model count data and to correct for over dispersion, we used a Poisson-lognormal model with a log link function by including a per-observation random effect as described above .We modeled CBB removal by ants using a GLMM. We included treatment , coffee plant distance to nest tree, ant activity on coffee plants after string placement, the interaction between treatment and distance, and the interaction between treatment and ant activity on plants as fixed effects . We also included ant activity on nest tree and coffee plant height as covariates. Random effects were modeled with plant identity nested within site to account for the block design of the experiment and to control for variation between our sites. To model count data in our response variable, we used a Poisson distribution with a log link function.We constrained model selection to include biologically pertinent terms for inference and to aid in model interpretation. A full model of these terms was tested, along with subsequent models of different covariate combinations and a null intercept-only model of random effects .

The best fit model was determined via backwards model selection compared to the full model, where the model that resulted in the lowest AIC was selected, if ∆AIC > +2 when the best fitted model was not the full model. Overall significance in models was assessed using Wald type II Chi-squared tests. Statistical differences among treatments were compared by Wald Z tests . In all cases, fixed effect parameters and the variance of random effects were estimated by maximum likelihood with Laplace approximation using the “glmer” function in the “lme4” package in R . To aid in data interpretation, we removed one coffee plant replicate from our analysis where measured ant activity was more than double that of any other plant measured and may have resulted from an unusually high buildup of scale insects which are tended by A. sericeasur on coffee. Additionally, one nest tree replicate was not included in the tree activity analysis because the data were not collected at that site. Finally, coffee plant height and distance to nest tree were centered and scaled to aid model interpretation. All analyses were performed in R .Although there was an 18.6% increase in ant activity on nest trees after experimental set up , including time in our model did not improve its explanatory power. Canopy cover also varied from 53% to 94% among sites; however, including it as a factor did not improve model fit. The GLMM that best explained ant activity on trees was our null intercept-only model . Thus, we did not further assess statistical significance for our model of ant activity on nest trees.The GLMM that best explained CBB removal was our full model, which included treatment , coffee plant distance to tree, ant activity on coffee plants after string placement, the interaction between treatment and ant activity on plants, and the interaction between treatment and coffee plant distance as fixed effects . This model also included coffee plant height and ant activity on the nest tree as covariates.Coffee berry borer removal was nearly three times higher on connected coffee plants than on control coffee plants . Overall, the effect of coffee plant ant activity on CBB removal was significant ; however, this effect was significant on control plants , but only marginally significant on connected plants in our model . Despite this, the interaction between treatment and ant activity on coffee plants was not significant , drainage for plants in pots indicating that ant activity on coffee plants and treatment independently drive CBB removal. CBB removal rate was not significantly affected by distance to the nesting tree . Additionally, neither coffee plant height nor ant activity on nest tree was significant covariates in our model . Although we chose the full model based on the lowest AIC value as explained in our methods, it should be noted that the second best model does not include coffee height, which suggests that it may not be a very important variable for determining the removal rate of CBB.Our experiment demonstrates that the addition of string to connect shade trees and coffee plants in coffee agroecosystems facilitates movement for A. sericeasur and potentially increases ant recruitment rates. Studies in natural systems have reported increases in ant activity with arboreal connections across the arboreal stratum , possibly driven by the easy access these pathways provide to resources . Other ants, such as Pogonomyrmex spp., prefer linear arboreal substrates and switch to cleared routes as a mechanism to reduce the energetic costs of ant foraging , and in some cases to decrease the risk of encountering predators . The observed increase in ant activity on connected coffee plants after the placement of strings suggests that structural connectivity can increase ant recruitment rates to foraging areas in coffee and may enhance the efficiency of movement for A. sericeasur. This may lead to increased foraging efficiency for ants and enhanced resource capture rates on coffee. However, this could also reflect other benefits associated with using linear arboreal substrates, such as avoiding predators, a behavior that is known to occur in A. sericeasur .

Using more efficient foraging pathways and thereby avoiding the leaf litter as a primary foraging substrate may potentially protect A. sericeasur workers from the attack of the phorid fly parasitoid Pseudacteon spp. . While ant activity only significantly increased after string placement on connected coffee plants, we also observed lesser increases in ant activity on control coffee plants and nest trees . This unexpected result could mean that strings, a novel element in the environment, acted as a form of habitat modification or disturbance, which increased overall ant activity in the local area. However, if our manipulation was the cause, we would have expected the ants to attack the jute strings , a behavior that we did not observe during the experiment. Experiments in tropical forests have shown that the long-term removal of lianas can influence ant richness on trees , and therefore may possibly also affect overall ant abundance and activity when promoted. It is also possible that other factors could potentially explain this result in control plants, such as changes in local abiotic factors that we did not measure systematically in our experiment. Future research which expands on the temporal scope of this study may be useful in assessing the long-term effects of artificial connectivity in this system. Ant activity after string placement was negatively affected by distance to the nesting tree . This result is consistent with previous studies suggesting that within 5 m A. sericeasur dominance in the leaf litter decreases with distance to the nesting tree . However, in our study, the effect of distance after string placement was significant only on control plants, but not on connected plants. This suggests that connections could buffer the negative effects that larger distances from the nesting tree pose to ant activity and potentially increase antprovided biological control services in these plants. Connected coffee plants also had significantly higher CBB removal than control plants . Overall, greater ant activity on coffee plants was associated with higher CBB removal rates , suggesting that ant activity directly influenced CBB removal rates. However, while this effect was significant on control coffee plants, it was only marginally significant on connected plants. While we believe that these results support the hypothesis that connectivity enhances ant foraging and bio-control services on coffee, the use of dead CBB in this experiment as a proxy to measure bio-control may explain the only marginally significant effect of ant activity on CBB removal in connected plants. It is possible that dead prey exhibit more variable recruitment responses from ants than live prey. Despite this, it is likely that strings facilitated ant movement to coffee plants by providing a smooth, linear substrate and indirectly increased CBB removal . In other systems, the leaf-cutting ant Atta cephalotes uses fallen branches to rapidly move between areas and thereby quickly discover new food resources .