Bananas (Musa spp.,Musaceae, Figure 1) are the largest export crop in Costa Rica and are the most popular fruit in the United States (reviewed by [7]). In 2009, banana imports exceeded 14 million tons in the USA-Canada, European Union, and Japan; and Costa Rica is the third top exporter of bananas world-wide behind Ecuador and Philippines [40].Commercial banana cultivation practices currently depend upon regular applications of a variety of fungicides, herbicides, insecticides and nematicides to maintain productivity (see [5,6,7] for complete list of pesticides used). We focused our caterpillar collection on two farms near Puerto Viejo, Sarapiqui, Costa Rica. Rebusca is a conventional input farm, while Penjamo is a moderate-input farm (Table 1).

We investigated how exposure to agro-chemicals (herbicides, fungicides, nematicides, and insecticides) affected the encapsulation and melanization response of Caligo memnon (Nymphalidae, Brassolinae) caterpillars. These caterpillars are commonly found in banana plantations and feed on plants in the genus Musa (Musaceae) and Heliconia (Heliconiaceae). During June and July 2004, C. memnon caterpillars were collected from both Rebusca and Penjamo and brought back to La Selva Biology Station for immune challenge. For each sampling trip, half of the collected individuals were changed to feed on pesticide-free leaves from the forest at La Selva. To challenge the immune response, sephadex beads, 40–120 µm, (Sigma-Aldrich) were injected into the hemoceol of 3rd–5th instar caterpillars [9,41]. Before injected, the beads were dyed with Congo Red. Ten beads in Ringers solution were injected into the hemolymph of each caterpillar using a glass Pasteur pipette fashioned into an injection syringe [13]. The injection site was covered with Second Skin^® adhesive to repair any damage. The caterpillars were allowed to continue feeding on their respective diets for 24 hr after injection and were freeze-killed. Beads were recovered by dissection and photographed using a camera mounted on a dissection microscope focused at 80×. Since the beads were dyed red before injecting them into the caterpillars, we quantified melanization by measuring the red value (r-value, Adobe Photoshop ver. 6.0), a scale ranging from 0–255, where 0 = pure gray, and 255 = pure red, for each bead. The mean r-value for all the beads from each caterpillar was statistically compared between treatments. The r-value was transformed into percent melanization (1-(r-value/maximum r-value)) where the maximum r-value is 255 for a non-injected, unmelanized bead.

A Bayesian approach was used to analyze these data using informative priors from previous experiments with Caligo memnon [9]. Bayesian analysis produces statistics that estimate the probability of the hypothesis given the data [P(hypothesis)|data] and incorporates prior parameter estimates to create posterior probability distributions. The mean and variance of melanization from prior encapsulation experiments with C. memnon were used as informative priors for the caterpillars fed La Selva leaves (mean = 74, variance = 474). For the caterpillars fed pesticide exposed leaves, we utilized an effect size from immune response experiments that examined responses to leaves treated with secondary metabolites versus non-treated leaves [13] to calculate the pesticide prior (mean = 61, variance = 474). To estimate model parameters, we used Markov chain Monte Carlo (MCMC) simulation where each step in the chain estimates melanization based upon the prior data. We used a 2000 step burn-in period followed by the 10,000 step MCMC to generate the posterior density distribution for the melanization of pesticide and pesticide-free individuals. We calculated posterior probabilities of the null hypothesis (no difference in encapsulation) by quantifying the frequency that the difference was equal to or greater than zero in each step of the post burn-in MCMC. For example, if pesticide-free (F) individuals have a higher melanization than the pesticide (P) individuals (F-P > 0) across 99% of the post-burn-in MCMC steps, we can conclude that the probability that melanization in pesticide-free caterpillars is equal to melanization in pesticide caterpillars is 0.01 [42,43]. Bayesian analyses were performed in SAS v.9.3 using PROC GENMOD.

From 2004 to 2006, we collected and reared caterpillars from banana plantations to obtain parasitism rates. The method used by our laboratory in the past [7,6,44] for collecting caterpillars in agriculture were utilized. Plantation managers were given a brief fact sheet about the project, and asked to fill out a very simple sheet with information about management practices at their farm. Caterpillars were collected in plastic bags and taken back to La Selva for rearing. Caterpillars were housed in an ambient lab and checked daily for parasitoids or adults. Fresh leaves from the banana plantations were added to the bags as needed. Once parasitoids emerged, they were stored in the freezer until they could be pinned. Parasitism frequency per species was calculated as the total number of parasitized individuals divided by the total number of results: (total # parasitized)/((total #adults + total #parasitized)).

MCMC simulations for the melanization response produced non-overlapping posterior probability distributions between the pesticide (banana) caterpillars (95% HPD (highest probability density: 1.94–2.61) and the pesticide-free (La Selva) caterpillars (95% HPD: 4.66–11.4) (Figure 2), indicating a difference in the melanization response (MCMC for the null, P < 0.0001). The 95% HPD interval or credibility interval is similar to 95% confidence intervals, but in this case indicates that there is a 95% chance that the population mean actually occurs within the calculated interval. Here, we found that the 95% HPD intervals for the banana and La Selva populations do not overlap, indicating a support for our hypothesis and a significant difference in the mean melanization response between groups.

Caterpillars feeding on the pesticide-free leaves (mean = 85.0, SD = 7.2, n = 48) had a higher melanization response compared to the caterpillars feeding on the pesticide leaves (mean = 77.5, SD = 10.9, n = 159) (Figure 3). When we examined the change in melanization over instars, we found that early instars had a slightly higher melanization score compared to later instars (2^nd instar mean = 82.5, SD = 9.2; 3^rd instar mean = 81.2, SD = 9.6; 4^th instar mean = 77.9, SD = 9.8; 5^th instar mean = 77.4, SD = 11.9) (Figure 4).

Parasitism rates were highest at La Selva, the pesticide-free location, followed by Penjamo and Rebusca (Table 2), the moderate and conventional pesticide plantations. At each location, parasitism was dominated by tachinid flies. This was especially true at both plantation sites, where tachinids accounted for 78% (Penjamo) and 81% (Rebusca) of total parasitism (Table 2). The species with the highest parasitism rate was the saddleback caterpillar, Acharia nesea (Limacodidae) (Table 3). These results corroborate earlier studies that predict highest levels of parasitism by tachinids, followed by braconids then other parasitic hymenoptera [6,11]. Models by Dyer and Gentry [11] also predict highest levels of parasitism by tachinids for generalists, such as Acharia and lowest levels of parasitism by tachinids for specialists, such as Caligo.

Results from the immune assay support our hypothesis that pesticides have a detrimental effect on the immune response of caterpillars. In this case, our inference is limited to the encapsulation and melanziation in Caligo memnon from two specific ecosystems in Costa Rica. Nevertheless, these results may be relevant to other insect species in similar ecosystems, as the evidence for the immunosuppressant activity of pesticides accumulates [45]. Since injections were performed within 24 h of C. memnon collection, pesticide residues on leaves most likely affected the inducible components of the immune response. That is, since individuals fed pesticide-free leaves after injection were initially collected from the banana plantations, it appears that the negative immune effects are reversible in a short period of time, with the immune suppression persisting on individuals that continued feeding on leaves with pesticide.

The most common pesticides found on banana plantation leaves are fungicides and a nematicide [5]. The fungicide is aerially deposited on leaves, thus herbivores are exposed both through the cuticle and during feeding. The nematicide is granular and applied at the base of the banana stem, where it seeps into the soil and is absorbed by the roots. The nematicide then becomes systemic in the plant and herbivores are again exposed during feeding (Dyer, unpublished data). Since treatment individuals were feeding on leaves collected from plantations, they were ingesting not just one pesticide, but likely two or more. This gives rise to the possibility that there could be a synergistic effect of the banana pesticides on the immune response. Recent studies show that synergistic effects of secondary metabolites may be a common mode of action by which plants deter herbivores [46,47,48]. While we did not explicitly test for synergy in pesticides, the possibility cannot be ruled out that there may be a combinatorial effect of pesticides on the insect immune response.

The effects of banana plantation leaves on the caterpillar immune response suggest that parasitic Hymenoptera might be more successful when their hosts are feeding on plantation leaves. However, in plantations not only will herbivores be exposed to pesticides, but also parasitic flies and wasps by the aerial spray and through nectar feeding. Results from the current study and in Dyer 2005 [6] demonstrated that parasitism by Hymenoptera increases as levels of pesticide decrease (natural forest > moderate input > conventional input; Table 2), and the increase is accompanied by a concurrent increase in proportion of parasitism by tachinid flies. In addition, the decrease in abundance of parasitoids at the banana plantations compared to forest sites indicates that although caterpillars are immune compromised, it is not benefiting parasitoid populations (as measured by percent parasitism). The first possible mechanism to explain these patterns is a potential increase in host availability for tachinids because of lower numbers of hymenopterans. Empirical [49] and theoretical [50] studies as well as reviews [51] have documented the importance of interspecific competition between parasitoids utilized for biological control. The second possible explanation, which could act in concert with the first, is that tachinids are released from control by hyperparasitoids. Several species of Brachymeria (Chalcididae), an important parasitoid genus in our banana plantations [6], are facultative hyperparasitoids of tachinids [52,53,54]. Although Brachymeria have not been reported to parasitize the Tachinidae observed in the study, they have been reared from other Lepidoptera parasitized by Lespesia aletiae, the numerically dominant tachinid observed in this study [55,56]. However, our collection and rearing techniques did not allow us to determine whether any of the parasitic Hymenoptera reared were hyperparasitoids. Finally, Matlock and de la Cruz (2002) [5] found apparent negative effects of chlorpyrifos, organophosphates, and carbamate nematicides on hymenopteran parasitoid communities in banana plantations. Our results clearly corroborate this negative association between high pesticide use and hymenopteran parasitism rates of banana caterpillars, but both the effects on encapsulation and the positive association between tachinid parasitism and pesticide use demands further study.