A minor pest reduces yield losses by a major pest: plant-mediated herbivore interactions in Indonesian cacao


  • Arno Wielgoss,

    Corresponding author
    1. Agroecology, Department of Crop Sciences, University of Göttingen, Grisebachstr. 6, D-37077 Göttingen, Germany
    2. Animal Ecology and Tropical Biology (Zoology III), Biocenter, University of Würzburg, Am Hubland, D-97074 Würzburg, Germany
      Correspondence author. E-mail: awielgo@gwdg.de
    Search for more papers by this author
  • Yann Clough,

    1. Agroecology, Department of Crop Sciences, University of Göttingen, Grisebachstr. 6, D-37077 Göttingen, Germany
    Search for more papers by this author
  • Brigitte Fiala,

    1. Animal Ecology and Tropical Biology (Zoology III), Biocenter, University of Würzburg, Am Hubland, D-97074 Würzburg, Germany
    Search for more papers by this author
  • Alfianus Rumede,

    1. Department of Agriculture, Klabat University, Airmadidi No. 1, ID-95371 Manado, Indonesia
    Search for more papers by this author
  • Teja Tscharntke

    1. Agroecology, Department of Crop Sciences, University of Göttingen, Grisebachstr. 6, D-37077 Göttingen, Germany
    Search for more papers by this author

Correspondence author. E-mail: awielgo@gwdg.de


1. Herbivores inducing host-plant trait changes can indirectly affect the performance, distribution, abundance and behaviour of other herbivores, even when they are temporally or spatially separated. However, it is unclear whether this occurs at scales relevant for applied purposes such as pest control in agricultural crops.

2. We studied the indirect effects of a minor pest, the mirid bug Helopeltis sulawesi, on the major pest of cacao Theobroma cacao in Southeast Asia, the pod-boring moth Conopomorpha cramerella. For 2 years, we surveyed herbivore damage and yield in 10 focal trees in each of the 43 cacao plantations and analysed patterns of co-occurrence of the two herbivores. In a two-choice experiment, we tested whether gravid females of C. cramerella searching for oviposition sites discriminate against pods damaged by H. sulawesi.

3. The proportion of pods affected by both pest species was significantly lower than expected. This pattern could not be ascribed to differential responses to environmental or management variables, but was because of avoidance of H. sulawesi damage by ovipositing C. cramerella females as shown in a two-choice experiment. The reduction in co-occurrence of damage by the two herbivores was found at tree and at plot scale and held across three harvest seasons.

4. The differential yield impacts by the two herbivores and the avoidance by C. cramerella of pods damaged by H. sulawesi lead to a yield optimum at a H. sulawesi incidence of 51%.

5.Synthesis and application. Plant-mediated indirect interactions between minor and major pest insects can be important drivers of yield loss at agriculturally relevant spatial and temporal scales. In cacao, the mirid bug H. sulawesi, a minor pest, generates conspicuous damage which often triggers pest control with insecticides. This practice may be counterproductive, because decreasing H. sulawesi damage benefits the main pest, the cacao pod borer C. cramerella resulting in a marketable yield optimum at intermediate densities of the minor pest. Pest control recommendations should take into account the relative effect of control measures on interacting herbivores to avoid replacing one pest problem with another, potentially more serious one, during the course of a fruiting season.


Indirect interactions between herbivores mediated by herbivore-induced plant trait changes can be more frequent and have a larger impact on biodiversity and community structure than direct competition between herbivores (Ohgushi 2005; Anderson, Inouye & Underwood 2009). Plants respond to herbivore damage by changes in allelochemistry, cell structure and growth, physiology, morphology, phenology (Karban & Baldwin 1997; Denno & Kaplan 2007) or nutritional content (Denno et al. 2000; Sandström, Telang & Moran 2000). These herbivore-induced plant responses may not only have multiple indirect effects on the performance and abundance of other herbivores (Van Zandt & Agrawal 2004; Poveda et al. 2005), but also influence their behaviour and spatial distribution (Van Dam, Hadwich & Baldwin 2000). Plant-mediated indirect interactions may involve herbivores that are separated spatially (Bezemer et al. 2003), temporally (Johnson et al. 2002) and/or taxonomically (Bailey & Whitham 2006).

The applied relevance of plant-mediated interactions in terms of yield losses to invertebrate pests in agricultural crops is largely unknown. This is surprising given the large potential for such interactions, with most crops being affected by a sequence of different herbivore species with contrasting ecologies and host-plant effects. The quantitative importance of effects depends on the spatial and temporal scales at which these effects play out. While quantitative and long-term studies on plant-mediated interactions have often been called for (Bolker et al. 2003; Werner & Peacor 2003), few have gone beyond qualitative snapshot studies (Hougen-Eitzman & Karban 1995; Utsumi, Ando & Miki 2010).

Here, we investigate the potential of plant-mediated indirect interactions between insect herbivore species to affect yield losses in cacao. Cacao Theobroma cacao L. is one of the most important cash crops worldwide, but also a species severely affected by pests and diseases. These have been estimated to be responsible for up to 30% losses in global production (Ploetz 2007) and can play a major role in cacao boom and bust cycles (Clough, Faust & Tscharntke 2009). The cocoa pod borer Conopomorpha cramerella Snellen (Lepidoptera: Gracillariidae) is the major cacao pest in Southeast Asia, causing crop losses up to 50% (Day 1989). The larvae mine into medium-sized pods, causing quantitative and qualitative effects on the yield, and impede the separation of husk and pod contents (Fig. 1c). The mirid parasite Helopeltis sulawesi Stonedahl (Hemiptera: Miridae) feeds on pods of all ages and young shoots of cacao (Giesberger 1983). The surface of damaged pods is covered with scars and a thick sclerotic layer, and this may promote the abscission of young fruits (cherelles) (Muhamad & Way 1995; Fig. 1b). In Southeast Asia, Helopeltis spp. are considered as serious pests by farmers and agriculturalists (Muhamad & Way 1995, personal observation A. Wielgoss, Y. Clough). As such, Helopeltis spp. are often used as a trigger for insecticide application, especially because imagos and early stages of the main pest C. cramerella are difficult to detect. Helopeltis sulawesi and C. cramerella may co-occur in the same pod at the same time, but H. sulawesi often damages the pod surface from the cherelle stage onwards, that is, 3 months before the preferred time-window for oviposition by C. cramerella.

Figure 1.

 Cocoa pods (a) healthy (b) epidermis with scars because of feeding by Helopeltis sulawesi (c) bisected pod with feeding damage by Conopomorpha cramerella larvae.

We hypothesized that the biology and the sequence of attack by the two herbivore species can lead to indirect interactions between them. Sap-feeding by Helopeltis spp. induces changes in the pod texture early in the pod development, before the stage preferred by C. cramerella for oviposition, which suggests a potential for asymmetric effects of the mirid on the pod borer.

We conducted a large-scale, 2-year study in 43 smallholder cacao plots on the island of Sulawesi, the main cacao-producing region in Indonesia. We expected that the damage by the local mirid species, H. sulawesi, will be negatively associated with subsequent damage by C. cramerella. In a two-choice experiment, we tested whether C. cramerella females discriminate between healthy pods and pods damaged by H. sulawesi. The results are quantified both in terms of herbivore response, potential environmental response and in terms of the impact on cacao yield.

Materials and methods

Study area and study sites

The study was conducted in Palolo and Kulawi valleys at the margin of the Lore Lindu National Park, Central Sulawesi, Indonesia. The elevation of the plots ranged from 400 m to about 1000 m above sea level.

We chose 43 cacao plantations, 21 in Palolo, 22 in Kulawi (see Clough et al. 2011 for details), which had seen little or no insecticide applications in the previous years, based on the information given by the farmers. In each plantation, we established a plot of 40 × 40 m, which was managed by local assistants from November 2006 to August 2008. Half of the plots were assigned randomly to frequent manual weeding regime (every 2 months), the other half of the plots to infrequent manual weeding (every 6 months). Each plot was separated into two 20 × 40 m subplots, and one half was fertilized twice a year with urea fertilizer (46% N). Fertilizer was applied twice a year from December 2006 to June 2008 at a rate of 217 g urea (100 g N) per tree with aliquots placed into 10 concentric holes around each tree, which were subsequently covered with soil.

Five focal trees were selected in each of the two 20 × 40 m subplots, that is, 10 trees per plot, to be used for more detailed phenological and pest and disease surveys (see below). Trees were selected randomly, with apparently non-productive or damaged trees discarded until we obtained five trees per subplot.

The temperature was recorded hourly using Dallas Thermochron ibuttons® (DS1921G; Maxim/Dallas Semiconductor, Sunnyvale, CA, USA), and the age of the trees was obtained from the farmers.

Study organisms

Conopomorpha cramerella

Female cocoa pod borers oviposit eggs (orange, flat and oval in shape, ∼0·5 mm length) on the cacao pod surface (Day 1985; Keane & Putter 1992). The preferred age of pods at the time of oviposition is 3 months (Day 1989). Upon hatching, the first instar larvae (∼1 mm length) tunnel through the floor of the egg shell and through the epidermis until the sclerotic layer of the husk, where they seek a weak or thin point to penetrate it. The young larvae feed on the placenta until fully grown (12 mm length, 14–18 days, with 4–6 instars). Pods attacked by cocoa pod borer larvae often ripe immaturely and show uneven yellowing. An infestation often results in beans being malformed, undersized, clumped at the pod husk and significantly reduces quality and quantity of the harvest (Fig. 1c). The mature larvae tunnel out of the pod and descend to the ground for pupation.

Helopeltis sulawesi

The eggs of Helopeltis spp. are white, elongated (1·0–1·2 mm in length) and apically compressed (Ambika & Abraham 1979). Two unequal respiratory filaments arise from the anterior end of the egg, the longer being 0·4–0·5 mm long. On cacao, Helopeltis spp. prefer to lay the eggs on the pods, but occasionally oviposit on young shoots. The incubation period of the egg varies with locality and season, but it is generally in the range of 6–11 days, although longer durations are observed occasionally. Helopeltis sulawesi has five larval instars that vary in size, colour and development of body parts such as antennae and wings (Stonedahl 1991). Data for adult longevity and fecundity of H. sulawesi are not available. For the closely related H. theivora, a mean adult longevity of 30 days has been measured (Tan 1974a). On cocoa, Helopeltis spp. feed on young shoots, flowers and developing pods (Muhamad & Way 1995). Feeding damage on pods appears as dark, circular lesions hardening as scars on the husk (Fig. 1b). Heavy infestations can result in pod malformation and premature drop. Tan (1974b) reported that pod abscission of young cherelles is mostly restricted to pods smaller than 5 cm and that larger pods have a reduced dry bean weight. Quantification of total yield losses because of Helopeltis spp. is difficult, because Helopeltis-induced abscission of young cherelles is masked by natural abscission, and the reduction in bean weights is highly dependent on the pod age at the time of attack (Muhamad & Way 1995).

Co-ocurrence study

Data collection

All cacao pods on the test trees were counted fortnightly from November 2006 to October 2008. The pods were assigned to size and health categories (small or cherelles; medium; large unripe; large ripe, that is, harvested; damage by H. sulawesi or C. cramerella or both; infected by Phytophthora palmivora (black pod disease); rodent feeding). Pods infected by P. palmivora and pods damaged by rodent feeding are of no value for harvest and, as is the usual farmer’s practice, were removed from the trees. Ripe pods were harvested and the fresh weight of the beans was recorded. To quantify the canopy cover per plot, we used vertical digital canopy photography with a fisheye lens. For each test tree, one picture was taken above the cacao tree canopy (5·8 m above the ground). We calculated the mean shade cover per plot above the cacao canopy for each plot using the software ‘Winscanopy’ (Regent Instruments Inc.; http://www.regent.qc.ca/).

Environmental correlates of herbivore incidence

To investigate whether patterns of co-occurrence of C. cramerella and H. sulawesi are because of environmental parameters, we fitted a multi-level model for each of the two herbivores, using a Bayesian hierarchical framework to accommodate the aggregation of pod counts at tree, subplot and plot level, as well as the temporal structure given by the three main harvests (Gelman & Hill 2007). The model was fitted in WinBUGS (Lunn, Thomas & Best 2000) using Markov-chain-Monte-Carlo sampling (three chains, 50 000 iterations, first 2500 discarded, thinning rate: 75, see Appendix S1–S4 in Supporting Information). We tested for effects of the parameters: altitude above sea level (continues in 1000 m, centred around the mean), weeding frequency of the plot (0 = every 6 months; 1 = every 3 months), shade cover above the cacao canopy per tree (proportion 0–1; centred around the mean) and fertilizer treatment of the subplot (0 = unfertilized; 1 = fertilized).

Randomization tests: co-occurrence of H. sulawesi damage and C. cramerella infection

To test for a possible effect of H. sulawesi damage on the probability of successful C. cramerella attack, we calculated the sum of all harvested pods per harvesting season per tree according to their pest incidence classification. Trees with no harvested pods in the categories ‘healthy’, ‘damaged by H. sulawesi’ or ‘infected with C. cramerella larvae’ were discarded to be certain that C. cramerella females were present and had a choice between pods with and without H. sulawesi damage within each tested tree. We calculated the observed ratios of pods on which both C. cramerella and H. sulawesi damage was recorded divided by the total sum of pods with C. cramerella infection. We then simulated the outcomes expected if C. cramerella chooses the pods at random and repeated this 1000 times for each tree. After that, we compared the randomized simulated ratios with the observed ratios for each tree and checked for significant differences. We combined the multiple P-values using Fisher’s method, which makes it possible to combine the results of multiple independent tests bearing upon the same overall hypothesis in a single test statistic (Borenstein et al. 2009). Because we consider it likely that oviposition site choice by C. cramerella females depends on the density of conspecifics and C. cramerella incidence differed between seasons (see Results), we conducted this analysis separately for each harvesting period (see Appendix S5 and S6 in Supporting Information).

To test whether the patterns within trees hold at the plot scale, which is most relevant for the farmers, we repeated this simulation at the plot scale, using the data aggregated by plot and harvesting season (see Appendix S7 in Supporting Information).

Oviposition experiment

Cocoa pod borer rearing

To rear C. cramerella, we harvested medium-sized cacao pods from plantations with high incidence rates. We picked pods that showed the typical symptoms of uneven, premature ripening of a C. cramerella infection. Each pod was covered with a cacao leaf and the pods stacked in a shed to protect them from rain. Each morning the cacao leaves were searched for cocoons with pupating larvae. The cocoons were transferred separately in small plastic boxes closed with fine mesh, where they remained until they reached the imago state. We identified the sex of the adults by distinguishing morphology of the tip of abdomen (Posada et al. 2011; see Fig. S1 in Supporting Information).

Oviposition choice test

To test whether C. cramerella females show preferences between healthy and H. sulawesi-damaged pods, we conducted an oviposition choice test. We harvested medium-sized cacao pods from a plantation with low C. cramerella incidence. Half of the pods showed serious incidence of H. sulawesi damage incidence, while the others were healthy. Conopomorpha cramerella eggs were removed carefully from the pod surface using water and a brush. In each experimental box (50 × 50 × 40 cm, mesh-covered; Fig. 2), we hung one healthy pod and one pod with H. sulawesi damage. The side on which the healthy and affected pods were placed was randomized to avoid bias affecting the results. In each box, we inserted one male and one female C. cramerella (imago 2 days after emerging, reared in laboratory, see above). The boxes were stored in a dry, ant free place at outdoor temperatures. After 5 days and nights, we opened the boxes, checked if the imagos had survived until the end of the test and searched the pod’s surfaces for eggs using a binocular microscope. We repeated this two-choice oviposition test 75 times in total, with different C. cramerella individuals and cacao pods at each trial. To test for differences between the number of eggs on healthy and Helopeltis-affected pods, we conducted a paired Wilcoxon signed rank test with continuity correction (see Appendix S8 and S9 in Supporting Information).

Figure 2.

 Oviposition experiment box with one healthy cacao pod and one with scars of Helopeltis-incidence.

Effects on yield

We fitted a joint multi-level model for pods affected with C. cramerella only and those attacked by both C. cramerella and H. sulawesi using the total number of harvested pods and the proportion of harvested pods affected only by H. sulawesi as explanatory variables. The model was fitted in WinBUGS (Lunn, Thomas & Best 2000) using Markov-chain-Monte-Carlo sampling (three chains, 10 000 iterations, first 5000 discarded, thinning rate: 15, see Appendix S10–S12 in Supporting Information). Based on the posterior distribution of the fixed effects, we calculated (for a tree in an unfertilized, infrequently weeded subplot) the expected pod weights. In this study, the dry weight of marketable beans per test tree was not measured because of the small quantities of beans per tree and harvest run. Instead, we used observed dry bean data of a follow-up study in the same research area conducted by the first author (15 plots each with 26 trees, biweekly data collection with the same method from April 2010 until July 2011). The expected contribution to yield by individual pods in the different categories (undamaged, damaged by H. sulawesi, damaged by C. cramerella, damaged by both pests) was estimated using a generalized linear model with the total marketable yield as a response and the counts in each harvested pod category as explanatory variables, with the intercept removed (see Appendix S13 and S14 in Supporting Information).



We could distinguish three main harvesting periods (Season A: February–August 2007, B: September 2007–March 2008, C: April–October 2008; Fig. 3). There were no significant seasonal differences in the percentage of pods affected by H. sulawesi (mean Season A: 35·50%, B: 34·75%, C: 41·98%; Fig. 4a). The percentage of harvested pods damaged by C. cramerella differed significantly between the three harvesting periods (mean: Season A: 54·93%, B: 73·20%, C: 72·40%; Fig. 4b).

Figure 3.

 Harvested cocoa pods (sum of 43 study plots) from February 2007 until October 2008 with the pods damaged only by Conopomorpha cramerella or Helopeltis sulawesi or affected by both pests, separated by harvesting seasons.

Figure 4.

 Percentage of harvested pods in a plot (N = 43) from three harvesting seasons. (a) with Helopeltis sulawesi incidence. (b) infected with Conopomorpha cramerella.

Co-ocurrence study

Environmental correlates of herbivore damage

The multi-level model predicting H. sulawesi damage suggests that altitude, weeding frequency of the plot and shade cover above the cacao canopy per tree did not have significant influence. The probability of H. sulawesi incidence was significantly larger in trees with fertilizer treatment (Table 1). None of the other tested environmental parameters had a significant effect on the probability of C. cramerella damage (Table 1).

Table 1.   Summaries of the 1002 samples from the posterior distribution for each of the parameters of the multi-level models fitted using WinBUGS for damage by Helopeltis sulawesi or Conopomorpha cramerella (2·5–97·5% = 95% credible interval; α0 = global intercept; significant fixed effects in bold)
 H. sulawesi damageC. cramerella damage

Co-occurrence of H. sulawesi and C. cramerella damage

At the tree level, the observed ratio of number of pods attacked by both herbivores to the total number of pods attacked by C. cramerella (median = 0·33, 1st and 3rd quantiles = 0, 0·67) was significantly lower than would have been expected if C. cramerella females did not discriminate against damage caused by H. sulawesi (i.e. 0·5). This was true for all seasons (Season A: < 0·001, d.f. = 710, Fishers-χ2: 3693·16; Season B: < 0·0001, d.f. = 214, Fishers-χ2: 643·21; Season C: < 0·0001, d.f. = 346, Fishers-χ2: 1254·61).

The same results were found at plot level: for all seasons, the observed ratio of number of pods attacked by both herbivores to the total number of pods attacked by C. cramerella (median = 0·18, 1st and 3rd quantiles = 0·10, 0·48) was significantly lower than expected based on the assumption of non-discrimination by C. cramerella females (Season A: < 0·001, d.f. = 88, Fishers-χ2: 438·30; Season B: < 0·00002, d.f. = 79, Fishers-χ2: 129·77; Season C: < 0·0003, d.f. = 86, Fishers-χ2: 140·17). In Fig. 5a, we show the observed relationship between percentage of pods affected by H. sulawesi and C. cramerella.

Figure 5.

 (a) Posterior predictive values for the percentage of cacao pods damaged by Conopomorpha cramerella given the percentage of pods damaged by Helopeltis sulawesi (black) and 95% credible interval (grey). (b) Predicted mean weight of marketable dry beans per cocoa pod [g] as a function of percentage of H. sulawesi-affected pods modelled for plots with mean observed C. cramerella incidence and H. sulawesiC. cramerella-interaction (black) and 95% credible interval (grey).

Oviposition experiment

In total, we bred 402 imagos of C. cramerella. We identified 208 males and 194 females (sex ratio did not differ from 1 : 1; d.f. = 401, P = 0·486). Time from pupation to hatching was 8·96 ± 0·74 days (mean ± SD, N = 402). The longevity of adults in the laboratory was 4·46 ± 1·35 days. In 37 of the 75 oviposition choice tests, there was no successful oviposition. In these cases, the male or female pod borer did not survive until the end of the test. In the 38 successful tests, C. cramerella significantly preferred healthy pods instead of H. sulawesi-affected pods for oviposition (paired Wilcoxon signed rank test with continuity correction: < 0·0001; d.f. = 27). On cacao pods damaged by H. sulawesi, we found a mean of 3·97 C. cramerella eggs (SD:  ± 6·2; max: 24, total eggs on 38 pods: 151; number of pods with no eggs: 13), on healthy cacao pods a mean of 13·66 C. cramerella eggs (SD:  ± 26·14; max: 152, total: 519; number of pods with no eggs: 0).

Effects on yield

According to our generalized linear model predicting yield weights per pod, pods damaged by H. sulawesi contained a similar mass of marketable beans (32·3 ± 10·4 g; mean ± SE) as healthy pods (mean: 32·5 ± 6·9 g). The weight of marketable beans in pods infected by C. cramerella was 33·8% lower (mean: 21·5 ± 3·7 g) than in healthy pods, and the outcome per pod was reduced by 56% (mean: 14·0 ± 5·8 g) if a pod was infected by both pests.

In Fig. 5b, we show the predicted weight of healthy dry beans per cacao pod as a function of percentage of pods with H. sulawesi damage in the presence and absence of C. cramerella. In the absence of C. cramerella, the mean weight of marketable beans per pod maintains the same level, that is, around a mean of 32 g per fruit. When C. cramerella is present, our model predicts a maximum mean weight of dry marketable beans per pod when 51% of cacao pods are affected by H. sulawesi (24·8 g; 9·6% more than at 0%H. sulawesi incidence; with 0%H. sulawesi-affected pods: 22·4 g; 33%: 24·5 g; 66%: 24·6 g; 100%: 22·1 g).


Top-down and/or bottom-up control are commonly used to explain herbivore population dynamics, while indirect herbivore–herbivore interactions, which may lead to complex ecological consequences for arthropod community compositions, still only get minor attention. We reveal a plant-mediated indirect negative interaction between two important, taxonomically separated cacao pests in South-East Asia, the mirid bug Helopeltis sulawesi and the cocoa pod borer Conopomorpha cramerella.

Plant-mediated interactions between the two herbivores

In our field observations, the percentage of cacao pods damaged by both pests was lower than expected based on the expectation of non-discrimination by C. cramerella females between pods affected by the mirid H. sulawesi and healthy pods. These results were mirrored in two-choice oviposition tests, in which C. cramerella showed a clear discrimination against cacao pods affected by H. sulawesi. Determining the physiological mechanisms behind the avoidance behaviour was beyond the scope of this study. However, owing to the feeding scars that are inflicted from the early stage of pod development, H. sulawesi-affected cacao pods have a harder and thicker sclerotic layer than healthy pods (Muhamad & Way 1995; Stonedahl 1991; Fig. 1c). It has been reported previously that Lepidoptera are able to detect physical surface traits of host plants and choose the preferred morphological phenotype for oviposition (Thompson & Pellmyr 1991) and that plant surface texture appears to be more critical for moths than for butterflies in the evaluation of potential oviposition sites (Renwick & Chew 1994). It has been shown that the physical properties of the sclerotic layer of the pods are associated with larval mortality and performance of C. cramerella (Azhar & Long 1996), so the oviposition discrimination of pods with traits associated with H. sulawesi damage appears to be adaptive. There are few other examples of indirect plant-mediated interactions via altered oviposition choice behaviour (Kruess 2002; Wise & Weinberg 2002; Poelman et al. 2008).

We cannot eliminate the possibility of a reversed indirect effect of a C. cramerella infection on feeding or oviposition preferences of H. sulawesi. We expect the size of such an effect on oviposition behaviour to be rather small, because the dispersing abilities of H. sulawesi are weak (the first larval instars are wingless and the adults are poor flyers) and alternatively lays eggs on shoots of cocoa, if no suitable pod is found (Stonedahl 1991). As Helopeltis feeding only has a direct effect on bean quantity or quality when it affects very young pods (cherelles, Muhamad & Way 1995), it is likely that a possible indirect interaction of a C. cramerella infection on feeding preference would not be economically relevant.

Consistency across spatial scales and seasons

Local patterns need not translate to larger scales, that is, local host-plant level avoidance by a pest species could be assumed to cause increased incidences on neighbouring host plants. However, our results not only confirm plant-mediated indirect interactions between H. sulawesi and C. cramerella on single fruits or trees, but also show that H. sulawesi negatively affects the incidence of C. cramerella at the scale of a plot (40 × 40 m). The size of the plot was chosen to mirror the smallholder management unit as commonly found in Central Sulawesi, suggesting our results are relevant to the scale at which management decisions are made by the farmer. The mechanisms behind the larger-scale patterns were not studied directly, but it is known that when encountering low densities of host plants appropriate for oviposition, gravid Lepidoptera females can alter their search patterns (Thompson & Pellmyr 1991), for example, by flying longer distances between landings and doing fewer sharp turns (Odendaal, Turchin & Stermitz 1989), which causes an individual female to move faster between patches until it reaches a more rewarding area with more suitable oviposition sites. A similar change in searching behaviour combined with a reduced larval performance because of thicker sclerotic layers of H. sulawesi-affected cacao pods may be responsible for the reduced densities of C. cramerella-damaged pods in plots with higher H. sulawesi incidence rates. It may not be uncommon that plant-mediated indirect interactions are important for larger-scale insect herbivore distributions and dynamics. While relatively few studies addressed larger-scale patterns driven by plant-mediated indirect interactions, of the 90 studies on herbivore-induced plant trait-mediated interactions (71 studies reviewed by Ohgushi (2005) and 19 studies published since 2005; see Appendix S15 in Supporting Information), nine could show a propagation of plant level effects to larger spatial scale. For instance, Karban (1986) reported the case of a folivorous spittlebug that had lower persistence when feeding on leaves affected by a leaf-herbivorous moth, resulting in lower spittlebug densities in areas with higher moth abundances.

In addition to being valid across scales, our findings suggest that the ecological importance of plant-mediated indirect competition in our study holds across seasons. Field experiments have shown that because of variation in climate and/or herbivore development times, the outcome can differ dramatically among replicates or years (Van Zandt & Agrawal 2004). In our field study, even though C. cramerella incidence changed between seasons, we observed the same plant-mediated indirect interaction patterns in three consecutive harvesting seasons.

Plant-mediated indirect interactions affect yield losses

To our knowledge, there are no studies addressing economic relevance of plant-mediated indirect interactions in any crop species. Yield losses, or increases (Poveda, Jímenez & Kessler 2010), because of herbivory are usually studied for single pest species. Muhamad & Way (1995) hypothesized that abscission of pods because of early damage of H. theivora on cacao yield is unlikely to contribute importantly to ultimate crop losses because it is overlapped by natural cherelle wilt and later compensated by increased cherelle production. In their data, yield losses were closely related to the time of damage: H. sulawesi-affected medium-sized pods had ∼15% lower yield, but yield was unaffected when full-sized but not yet ripe pods were heavily damaged. In the simulation of our observation data, where no differentiation of pod age at attack could be made, there were only marginal direct crop losses on pods that were affected by H. sulawesi (<1%). In the study of Day (1989), the crop losses because of C. cramerella reached about 40% with 90% of pods attacked. In our study, the first to integrate the effects of both pest organisms, the estimated yield losses directly linked to C. cramerella at an infection rate of 90% and in the absence of H. sulawesi, was comparable (37·1%).

Our results show that in areas where C. cramerella is absent H. sulawesi does not lead to significant yield losses. But it is notable that pods affected by both pests contained significantly fewer marketable beans than a pod affected by C. cramerella alone. Nevertheless, in regions where C. cramerella is common, as in our study area (mean infection rate ∼67%), there is a yield optimum at an incidence rate of 51%H. sulawesi-damaged pods (Fig. 5b). At this level, the crop losses because of C. cramerella are minimized via the indirect negative interaction between the two pests, which overcompensates the direct crop losses because of H. sulawesi resulting in 9·6% more marketable cacao beans than if H. sulawesi is absent. We deliver an example of how indirect plant-mediated interactions between two agricultural pests result in minimized overall yield losses in co-occurrence of the two pest species.

Management implications

The cocoa pod borer C. cramerella is the most destructive pest of cacao plantations in Southeast Asia, but the fact that the Helopeltis spp. mirids are conspicuous and C. cramerella is rather cryptic can lead to a disproportionate role of the mirids in affecting management decisions such as whether to spray insecticides or not. The authors have observed that insecticide use is often triggered by early season H. sulawesi feeding scars. In this case, the insecticide is sprayed directly on the pods. The sap-feeding H. sulawesi nymphs are day-active, flightless and remain at the pods most of the time and, therefore, are much more affected by this management practice than C. cramerella, which hides for most of the day underneath branches and visits the pods only for short-time periods for oviposition at night (Day, Mumford & Hing 1995). In areas where C. cramerella is common, the described negative indirect interaction between the two pest organisms would make application of broad-spectrum insecticides on the cocoa fruits against H. sulawesi pointless or even counterproductive, because the reduction in losses because of H. sulawesi is likely to be exceeded by the increasing yield losses because of C. cramerella. The same might be true for integrated pest management practices against Helopeltis spp. by ants, as proposed by Graham (1991) and Way & Khoo (1989). Ant species that are effective predators of Helopeltis might indirectly increase the probability of a C. cramerella infection, given that the two pests have contrasting spatial and temporal activity patterns, and are, therefore, unlikely to be equally exposed to any particular predator species. The relatively immobile H. sulawesi is possibly a much easier prey for ants such as Dolichoderus spp. (Way & Khoo 1989, 1992; Stonedahl 1991) than the rather cryptic C. cramerella. In consideration of these facts, it might be preferable to use more specific C. cramerella control strategies such as complete regular harvesting to break the development cycle, pheromone traps or selection of resistant genotypes.

More generally, our results demonstrate that plant-mediated indirect interactions can be quantitatively relevant at spatial and temporal scales relevant for agricultural management. Given the large potential for such interactions in most other crops, we suggest that pest management strategies could benefit significantly from being adapted to account for these effects. In particular, this study underlines that efficiency of any pest control measure should not only be measured in impact on the target species alone, but should also consider effects on species which might directly alter host-plant traits and indirectly other species, which includes not only natural enemies and pollinators but also other herbivores.


This study was part of the DFG Project STORMA (“Stability of Rainforest Margins of Indonesia”; SFB 552; focus 2, subproject C3). The German Academic Exchange Service (DAAD) covered the personal funding of A.W. We thank S. Erasmi and D. Seidel for sharing the canopy cover data, the coordination of STORMA in Göttingen, Bogor and Palu, A. Anshari for support and M. Iqbal for assistance in the field.