After release, bags were checked for feces, which were removed with ethanol-flamed tweezers and placed in ready glass vials or 2 ml Eppendorf tubes filled with ethanol. They were refrigerated until transport to the United States and then stored at −20°C. Bags were always immersed in a 10% bleach solution, sun-dried, and washed after use. We conducted feeding trials with three common, small insectivores that frequent coffee plantations and were expected to consume the borer: Rufous-capped Warbler , Rufousbreasted Wren , and Plain Wren . Individuals were fed 0, 2, 4, or 8 borers collected in nearby plantations . Specifically, we held each bird’s mouth open and placed the borers inside with a sterilized tweezers. We then used a syringe to inject water into the bird’s mouth, inducing it to swallow the insects. Birds were then placed in mesh cages over sterilized cotton floors. Cages were checked for fecal samples every 15 minutes for 1.5 hours; stressed birds were released earlier. Though referenced in Karp et al. , feeding trial data were not previously analyzed.We poured off ethanol and weighed samples prior to DNA extraction. For all species that were strictly frugivorous, we combined samples derived from different individuals of the same species captured at the same plantation to reduce processing time and cost. The combined sample was homogenized and a <0.25 g subset was extracted. Because many more individuals were captured, black flower bucket in the second year we additionally combined samples from multiple individuals of the same species at the same plantation for all omnivores and large insectivores. Again, samples were homogenized and a <0.25 g subset was extracted. Samples from feeding trials and from small insectivores were always extracted individually. Extraction was performed with kits , modified to increase product yield .

All extractions were accompanied with negative controls with no fecal material added so that we could identify any possible sources of contamination. Following extraction, we ampliftied borer DNA through PCR with borer-specific primers . Each PCR reaction was accompanied with negative and positive controls, derived from feeding trial samples. Though primers were designed to be borer specific, we sent all amplicons of the expected size range for sequencing because many PCR cycles can result in ampliftication of non-target DNA. We used Sequencher to form consensus alignments of DNA reads from forward and reverse primers that were then compared to a borer reference sequence. Only sequences with clean, discernable peaks at target base pairs were analyzed. Those sequences with >98% similarity to borer reference sequences were deemed successful borer identifications. The next most similar sequence from another species in Genbank at the target amplicon was 85% similar.We accidentally contaminated several samples with PCR amplicons, necessitating the development of alternate primers. We ampliftied an 113 bp segment of COI, outside the previous ampliftication region, with forward and reverse primers. Reactions were carried out using the same reagents and protocols, apart from the annealing step . Products were visualized on gels, and negative controls confirmed that the contamination was previous PCR product. Because primers were not borer-specific, all products of the expected size range were sequenced and compared to reference sequences. After the borer, the next most similar sequence in Genbank was 86% similar.We assessed whether confirmed borer predators shared functional traits through compiling a trait database for birds in our study area, focusing on resource and acquisition traits that may affect pest-control provision . We used measurements from birds we captured, and a bird population dynamics dataset collected at 18 nearby sites . Wing chord length and mass were obtained from the population dynamics dataset. We also calculated the total number of captures for each species.

We collected bill width , bill length , and tarsus length from species that we trapped during fecal sample collection. Body lengths were obtained from literature . We gathered behavioral traits from literature . We translated foraging stratum into an ordinal scale , and calculated the average foraging stratum for each species. We quantified diet breadth as the number of food categories consumed . From literature and conversations with local ornithologists , we also identified species that consumed insects and the subset that specialized strictly on insects.We used Generalized Linear Mixed Models to identify variables that significantly influenced the probability of borer DNA detection in feeding trials . The model contained a logit link and binomial error structure, and the feeding trial as a random effect. Species identity, elapsed time since feeding, number of borers fed, fecal sample mass, and 2-way interactions were included as predictors. We then used backwards model selection, iteratively dropping predictor variables and comparing nested models with Aikaike Information Criteria and likelihood-ratio tests . Next, we determined whether species confirmed as borer predators shared traits. Because very few of the birds that were not involved feeding trials tested positive for borer DNA, it was impossible to use logistic regression to associate bird traits with borer predation. Instead, we created a randomization procedure in which six species were drawn at random 1000 times from a species pool , and the average trait value for these species was calculated each time. This procedure generated a null distribution for each trait that could then be compared to the average trait value of confirmed borer predators. If the observed trait value fell outside the 95% confidence interval, then we determined the trait was a significant predictor of borer-predator identity.We ampliftied and sequenced borer DNA from the feces of six species: Buff-throated Foliage-Gleaner , Common Tody-Flycatcher , Rufous-breasted Wren , Rufous-capped Warbler , White-tailed Emerald , and Yellow Warbler . The majority of detections were derived from surveys conducted in 2010 and reported in Karp et al. , even though fewer samples were collected in 2010 . In total, 30 and 27 samples yielded PCR products of the expected size range, 4 and 6 of which were >98% sequence similar to borers.

Though detections rates were low, feeding trials confirmed the efficacy of our approach. Fifteen samples yielded PCR products of the expected size range, all of which were confirmed as borers through sequencing . Additionally, detection probability increased with elapsed time since feeding. For example, models predicted that, for birds that were fed 4 borers and defecated a 0.05 g fecal pellet, detection probability increased from 10% at 20 minutes after feeding to 50% at 80 minutes after feeding. Species identity of feeding trial birds did not influence detection probability. While it is possible that positive borer detections in feeding trials could have resulted from prior consumption, no detections occurred when birds were not fed borers. Moreover, the low detection rates in non-feeding trial birds further reduce the likelihood that positive detections were the result of prior foraging. We found that functional traits differed between confirmed borer predators and other sampled species . Borer predators had narrower bills and shorter wing chords than expected. Though not significant, predators also tended to be smaller, both in mass and length . Diets were specialized , and insectivores were overly represented — only the White-tailed Emerald was not a specialized insectivore. Borer predators were not species of general conservation concern. Predators were equally abundant to other species in our study system, and were neither endemic nor listed on the IUCN red list. Leveraging traits that were over-represented in confirmed borer predators, we predicted other species that may consume the borer but no pest DNA was found in their fecal samples, likely the result of low detection rates .Ecosystem-service management necessitates identifying service providers, especially in the many agricultural systems that are rapidly expanding and intensifying . Our analysis of ~1500 fecal samples documented that six Costa Rican bird species consume coffee’s most damaging insect pest. Still, detection rates were very low: only 0.7% of analyzed samples contained borer DNA. We offer several explanations for low detection. First, we sampled the entire bird community, including frugivores which do not likely consume the borer. Second, borer abundance is low in our study system. Only 2.5% of berries across plantations are currently infested with borers, square black flower bucket whereas infestation has soared above 90% in other countries . Third, detection windows may be narrow. We detected borer DNA in only one sample defecated within 30 min of feeding. Insect DNA could be detected in Carrion Crow feces 30 minutes to 4 hours after consumption . Borers disperse most often and hence are most vulnerable to predation in the afternoon .

Because tropical weather constraints precluded afternoon sampling, a mismatch in sampling and consumption could have depressed detections. Finally, feeding trials demonstrated that false negatives are regular. Models predicted that a positive detection was ~20 times more likely when birds were fed 8 borers and defecated 0.1 g versus 2 borers and 0.01 g. In addition to DNA degradation in the gut, our extraction and PCR procedures may be prone to false negatives. First, PCR inhibitors can persist through extraction and impede DNA ampliftication from fecal pellets . Second, unlike the primers developed by Jaramillo et al. , the primers that we developed were not specific to the berry borer, meaning the primers could have ampliftied DNA from any one of the many species of insects that a bird had recently consumed. Moreover, iterant non-specific PCR binding of either primer set could generate chimeric sequences of multiple species. Accordingly, only 10 of the 57 samples that yielded PCR products of the expected size range were identified as borer DNA after sequencing. Future work could utilize a post-PCR sorting method such as next generation sequencing or cloning to help reduce the frequency of false negatives . Low detection rates suggest that there are other species that consume the borer that we did not identify. The species we did identify, however, shared traits that may be characteristic of these other predators. All identified borer predators except the nectarivorous White-tailed Emerald were strict insectivores. Unsurprising given the borer’s size , borer predators had narrow bills. Additionally, these species had short wings, ideal for navigating the dense coffee understory . It is possible that functional traits would change with a larger sample of predators; however, confirmed borer predators in Jamaican coffee plantations shared many of these traits , supporting our hypothesis that they may help predict other predators . A key difference between our studies, however, is that only one of the species that we identified as a borer predator is migratory . Wecollected our fecal samples during the period of maximum borer dispersal , a time when most migratory species are absent from Costa Rica. Because migratory species could consume borers during their secondary dispersals that occur throughout the year, future work should temporally expand sampling effort to ensure that migratory species are well represented. Our work yielded the critical management insight that managing the predators of crop pests may require looking beyond traditional conservation targets. The six documented borer predators were not rare, endemic, or listed on the IUCN red list. Traditional conservation efforts for threatened species often center on delineating large protected areas. Focusing conservation explicitly in agricultural landscapes could benefit species involved in providing critical ecosystem services to farmers . By confirming that birds consume pests, our work could thus help change attitudes towards biodiversity in human-dominated landscapes by fostering greater recognition of its role in supporting human well being. Species interactions play a pivotal role in many ecologically and economically important ecosystem processes. Uncovering the basic relationships between animals and their food is critical for managing pest control, pollination, seed dispersal, and sanitation . Molecular methods can provide us with a window into these interactions, in some instances for the very first time. Our results demonstrate how identifying just a few key interactions between predators and their prey can yield potential insights for management. Indeed, managing nature to enhance both biodiversity and human well being requires diverse approaches. Techniques and practices have already been borrowed from fields as diverse as agronomy, economics, hydrology, psychology, and sociology. Our results indicate that molecular biology offers ecologists the ability to expand their toolkit in key dimensions and, in turn, advance ecosystem service management.