Owing to complex direct and indirect effects, impacts of higher trophic levels on plants is poorly understood. In tropical agroecosystems, ants interact with crop mutualists and antagonists, but little is known about how this integrates into the final ecosystem service, crop yield. We combined ant exclusion and introduction of invasive and native-dominant species in cacao agroecosystems to test whether (i) ant exclusion reduces yield, (ii) dominant species maximize certain intermediate ecosystem services (e.g. control of specific pests) rather than yield, which depends on several, cascading intermediate services and (iii) even, species-rich ant communities result in highest yields. Ants provided services, including reduced leaf herbivory and fruit pest damage and indirect pollination facilitation, but also disservices, such as increased mealybug density, phytopathogen dissemination and indirect pest damage enhancement. Yields were highest with unmanipulated, species-rich, even communities, whereas ant exclusion decreased yield by 27%. Introduction of an invasive-dominant ant decreased species density and evenness and resulted in 34% lower yields, whereas introduction of a non-invasive-dominant species resulted in similar species density and yields as in the unmanipulated control. Species traits and ant community structure affect services and disservices for agriculture in surprisingly complex ways, with species-rich and even communities promoting highest yield.
Ecosystem services and disservices depend on a complex suite of direct and indirect interactions  involving multiple functional groups such as herbivores , pathogens , predators [4,5] and pollinators . As anthropogenic disturbances decrease richness and modify dominance structures within communities, biodiversity–ecosystem function studies are shedding light on the effect of animal community characteristics in the delivery of individual ecosystem services such as pest control or pollination . However, few of these studies account for the fact that many animal groups simultaneously interact with many different components of the ecosystem via trophic and non-trophic interactions and can thereby impact multiple intermediate ecosystem services and disservices [8,9].
In tropical agroecosystems, where crop production depends on chains of intermediate ecosystem services, such as pollination and natural control of multiple pest and disease species, ants play such a key role [10,11]. Anthropogenic disturbance has been shown to favour strong numerical and ecological dominance of native or invasive ant species [12,13]. Dominant ant species often build multiple (polydomous) nests and form particularly intensive mutualisms with hemipterans (so-called trophobionts), which provide honeydew as sugar source in exchange for protection against potential predators . Some of these species reduce species richness and evenness of ant communities by aggressively excluding other species from their territory and from food sources , whereas others are more tolerant towards co-occuring ants and do not reduce species richness.
Predicting differences in ecosystem services between ant communities with evenly distributed species abundances and those dominated by single-species is not trivial . In terms of predatory function, both diversity and dominant ants have been heralded as being predictive for higher top-down effects on herbivorous arthropods [11,16]. Ant communities with low diversity but high abundance (e.g. those dominated by an aggressive species) may result in reduced top-down effects on the arthropod fauna , including agricultural pest species [18,19], as a consequence of a reduction in complementarity of traits and functional diversity . But ant communities with dominance of a single ant species may be equally or more effective in suppressing pest species than more evenly structured ant communities owing to higher worker abundances and patrolling activity . One can expect that the traits of the dominant ant species, such as spatio-temporal activity patterns and foraging behaviour, will determine whether dominance results in increased or reduced predation.
Crop production depends not only on pest predation, but also on a broad suite of ecological interactions, and effects of ants on crop yield also encompass ecosystem services for example enhancement of pollination success , and of ecosystem disservices for example pathogen spore dissemination , increases in herbivory via effects on pest–pest interactions  and mutualism with crop-damaging hemipterans . Little is known to date about effects of dominant ants and ant community properties on the outcome of such a suite of interactions which can be measured through final yield [1,24].
In grassland plant communities, the importance of species richness for ecosystem functioning increases with the number of functions considered . For ants, this relationship is less straightforward because they use different resources, the exploitation of some of which may be an ecosystem service (predation of pests), while that of others can cause a disservice (exploitation of plant sap via trophobionts). Nevertheless, we may expect ant communities with high species richness and high evenness to be beneficial by maintaining services  and diluting potential disservices associated with ant dominance , because most ants are predatory to some extent, but only a subset of the species are likely to transmit spores of pathogens  or to engage in mutualisms with economically important sap-sucking pests .
Here, we present a highly replicated ant fauna manipulation experiment in smallholder cacao in Sulawesi, Indonesia, in which we used ant exclusion treatments to confirm the economic importance of the presence of ants, and two further treatments in each of which we experimentally introduced one of two regionally common dominant ant species to compare their effects with those of the naturally occuring ant fauna. This resulted in four distinctive ant communities: (i) unmanipulated ant fauna with intermediate abundances, high evenness and species richness. (ii) Dominance by the native Dolichoderus cf. thoracicus with high abundances, species density and evenness. (iii) Dominance by the aggressive invasive Philidris cf. cordata with high abundances, but low species density/evenness. (iv) Exclusion of all ants (extremely low abundances and low species density/evenness).
Our hypotheses were the following:
— Ant exclusion negatively affects crop productivity and yield.
— Dominant species maximize individual intermediate ecosystem services, but not the final, integrated ecosystem service, which is agricultural yield.
— Species-rich ant communities with high evenness result in the highest yield, because high functional diversity maintains major ecosystem services while buffering potential disservices of single species.
Our experimental approach reduced confounding effects of environmental variables or land-use practices on ant community structure. For 16 months, every two weeks we monitored ant communities, flowers, fruits, incidence of fruit damage owing to pests and diseases and final yield. To test the hypotheses above, we analyse the effects of ant community structure on pests and diseases of cacao and how different ant communities affect the balance between services and disservices, and drive final crop outcome. We found that traits of dominant species and ant community structure affect cacao trees via a complex crop–antagonist–mutualist interaction network. Even and species-rich communities provided the best services for agriculture leading to highest yields.
2. Material and methods
(a) Study area and study plots
All sites were situated in Palolo Valley, Central Sulawesi, Indonesia. In May 2009, we selected 15 cacao plots (50 × 50 m) without insecticide use for at least 1 year, differing in shade intensity and with absence of Philidris cf. cordata and Dolichoderus cf. thoracicus (hereafter called Philidris and Dolichoderus, respectively). In each plot, we placed four subplots (10 × 10 m) with a minimum distance of 8 m containing nine neighbouring cacao trees each.
(b) Ant treatments
In cacao plantations of our study region, the dolichoderine ant Philidris cf. cordata has recently become invasive and ecologically dominant. It reduces ant species richness  and can displace other dominant species, for example the native Dolichoderus cf. thoracicus, which might be beneficial as an effective predator of cacao pests [27,28]. Both species are mutualistically associated with the cacao mealybug Cataenococcus hispidus Morrison (Hemiptera: Pseudococcidae), with whom they exchange protection against honeydew. Both species form large colonies and can be numerically dominant, but they differ in their ecological traits: While Philidris workers are rarely active in the cacao canopy and they aggressively expel other ants from baits , Dolichoderus workers are active in the whole cacao tree including the foliage and are relatively tolerant towards other ant species .
The ant fauna manipulation treatments were assigned randomly to four subplots: (i) undisturbed naturally occurring ant fauna as control; (ii) establishment of Philidris or (iii) Dolichoderus as ecologically dominant ant species on the test trees using artificial nest sites and (iv) exclusion of ants from test trees using nest destruction and insect glue barriers on the stem base (see electronic supplementary material, figure S1). The treatments were installed in August 2009 and maintained until July 2011 (for details, see the electronic supplementary material). Treatments were maintained stable in all nine trees per subplot during that time.
(c) Data collecting
Every two weeks from March 2010 to July 2011, on all test trees, flowers were counted, cacao fruits were classified by size and pest/disease incidence, ripe fruits were harvested, beans of fruits of one subplot were pooled and dry weight of marketable beans was recorded. To account for yield quality, defective beans were separated and weighed separately from marketable dry beans. To avoid confounding effects of differences between the subplots, we recorded tree characteristics, temperature and shade cover (based on hemispheric pictures). We determined the herbivory rates two times on 20 fully developed, mature leaves on each of three randomly selected trees per subplot using digital scans and pixel counting. All ant-attended mealybug aggregations and their location were recorded in all test trees in May 2011. Ants were surveyed and identified to morphospecies one time before and three times after treatment installation in each subplot using standardized tuna and sugar baits on the nine cacao trees and at four ground locations per subplot (for details, see the electronic supplementary material).
(d) Inoculation experiment
The phytopathogen Phytophthora palmivora Butler (Peronosporales: Pythiaceae) causes the most severe cacao disease in Indonesia. Beside transmission through wind and rain, invertebrate vectors, especially tent-building ants, are the most important ways of dissemination . To check for contamination, we inoculated each 50 fruits with 0.5 cm³ of dead Philidris and Dolichoderus workers and material of the typical detritus tents of Philidris, which are built on cacao fruits (see electronic supplementary material, figure S3). As control for each group, we used the same inocula, but disinfected with 5% bleach. As a baseline, we used control 50 fruits without inocula. After 8 days, the area of typical P. palmivora lesions was measured. We compared each experiment group with its according control and the baseline control using Welch two sample t-tests. Identity of P. palmivora was confirmed in the laboratory (for details, see the electronic supplementary material).
(e) Data analysis
(i) Cacao tree characteristics
To detect possible confounding variables, we fitted linear mixed-effects models with canopy cover, cacao tree height, diameter at breast height and crown volume as dependent variable, treatment as explanatory variable and plot as random effect.
(ii) Ant community characteristics
To detect differences in ant community structure, we fitted linear mixed-effects models by restricted maximum likelihood (REML) for mean ant abundances at the baits (per subplot, pooled across all species) and ant species richness per subplot with plot and survey round as random effects. Then, we conducted a Tukey's contrast test for multiple comparisons of means.
(iii) Insect pests and diseases
To predict ant treatment effects on the incidence of the main pest, Conopomorpha cramerella Snellen (Lepidoptera: Gracillaridae) and the incidence of a minor pest, Helopeltis sulawesi Stonedahl (Hemiptera: Miridae), we fitted binomial generalized linear mixed models (GLMMs) by the Laplace approximation over the amount of infected harvested fruits versus the number of healthy harvested fruits per subplot, plot and harvest run. To detect differences in leaf herbivory, we fitted linear mixed-effects models by REML for percentage of leaf loss (mean per subplot) with plot and subplot as random factors, and we fitted linear mixed models by REML to check whether leaf loss was correlated with evenness. We fitted linear mixed-effects models by REML for total number of mealybug aggregations (log-transformed) per subplot with plot, subplot as random factors. To compare differences in location of trophobiont aggregations, we calculated the percentage of total aggregations found at leaves and shoots and fitted linear mixed-effects models by REML, with subplot and plot as random factors followed by a Tukey's contrast test for multiple comparisons of means. We fitted a binomial GLMM by the Laplace approximation over the total amount of medium and large fruits lost owing the phytopathogen P. palmivora during the experiment time versus the total number of harvested non-infected fruits per subplot, with plot as random effect.
(iv) Fruit set
To detect differences in fruit set rates, we aggregated open flower data from April 2010 to May 2011, but small fruit counts from June 2010 to July 2011 per subplot to avoid bias, because two months were the mean development time from flowers to small fruits. We then fitted binomial GLMMs by the Laplace approximation over the amount of small fruits versus the number of open flowers per subplot with plot and subplot as random factors.
(v) Early fruit abortion
To detect differences in the rates of early fruit abortion, we fitted binomial GLMMs by the Laplace approximation over the number of aborted fruits versus the number of survived fruits per subplot with plot and subplot as random factors. To assess possible influence of pests on early fruit abortion, we repeated the analysis with leaf loss, number of mealybug aggregations and percentage of Helopeltis damage as explanatory variables.
(vi) Fruit development
We aggregated flower and fruit data of all observation runs and of the nine trees per subplot and used linear mixed-effects models fit by REML with plot and subplot as random effects to test for ant treatment effects on the number of flowers, total number of young cacao fruits, number of fruits which survived the early fruit abortion and number of harvested fruits.
(vii) Yield and revenue
For dry yield and total marketable dry cacao beans, we used the first 12 months of data to have estimates for one complete harvesting season. Extrapolating the ant community effects on yield observed in our experiment to larger areas is legitimate, since Philidris, when naturally present in a cacao plantation can dominate more than 80% of the trees with similar high abundances as in our experimental treatments , and as well Dolichoderus, can be similarly dominant over larger cacao plantations, when provided with nesting sites and trophobionts [29,30]. The mean cacao world market price during this time was 3.14 US$ kg−1 . For each ant community treatment, we calculated total harvest value per year and hectare using the first 12 months of yield data and assuming 1111 trees per hectare (3 m planting distance).
(a) Cacao tree characteristics
Cacao tree height, stem diameter at breast height, crown volume and shade cover did not vary significantly among treatments in each plot (see electronic supplementary material, table S1).
(b) Ant communities
Before establishment of the treatments, species density in all subplot was similar (see electronic supplementary material, table S2). With treatments, ant abundances in subplots with experimentally established single species dominances were higher than in the control treatments (figure 1a, Philidris: t = 3.61, p < 0.01; Dolichoderus: t = 3.32, p < 0.01). In ant exclusion trees, the mean ant abundance was reduced to less than 10% of the control (t = −5.55, p < 0.01). While species density in Dolichoderus treatments was similar to the control (figure 1b, t = −0.55, p = 0.59), it was approximately 50% lower in the other two treatments (figure 1b, Philidris: t = −3.59, p < 0.01, ant exclusion: t = −4.14, p < 0.01). Similarly, the evenness in Dolichoderus treatments was comparable to the unmanipulated communities (figure 1c, t = −0.47, p = 0.64), while presence of the aggressive Philidris reduced evenness (t = −3.59, p < 0.01). Composition, abundances and species density of ant communities with Philidris or Dolichoderus dominance were similar to natural occurrences of these species found in cacao plantations of the same region [26,32] (see the electronic supplementary material, table S2 and figure S4a–d for model details and for ant communities characteristics, respectively).
(c) Effects on fruit set and initial number of fruits
There were no differences in number of flowers between the treatments (see electronic supplementary material, table S3). In ant exclusion treatments, a similar proportion of flowers set fruit as in the control (figure 2; electronic supplementary material, table S4; z = −26.1, p < 0.01), whereas in Dolichoderus (z = 46.5, p < 0.01) and Philidris (z = 29.0, p < 0.01) treatments, the rates were higher. Fruit set correlated positively with ant abundances (see electronic supplementary material, figure S2 and table S4; z = 26.1, p < 0.01). The number of young fruits were similar in the control subplot and in subplots dominated by Dolichoderus or Philidris, but lower in subplots with ant exclusion (figure 3; electronic supplementary material, table S3; Dolichoderus: t = −0.74, p = 0.46; Philidris: t = 1.05, p = 0.30; ant exclusion: t = −2.31, p = 0.026).
(d) Effects on early fruit abortion
Cacao trees can abort fruits in early development stages because of environmental factors, diseases and pests. The abortion rates were lowest in non-manipulated control subplots, followed by Dolichoderus-dominated ant communities (figure 2; electronic supplementary material, table S4; z = 2.2, p = 0.03) and ant exclusion (z = 7.2, p < 0.01), whereas Philidris-dominated ant communities had highest abortion rates (z = 2.2, p < 0.01). In control and Dolichoderus-dominated subplots, there were similar amounts of developing fruits (figure 3; electronic supplementary material, table S3; t = −1.03, p = 0.30). In ant exclusion subplots, the number of surviving fruits was lower than in control (t = −3.4, p < 0.01) and in Philidris subplots it was higher on average, but not significantly so (t = −1.03, p = 0.10).
(e) Indirect ant community effects on early fruit abortion
The percentage of aborted fruits was positively correlated with fruit damage by the sap-sucking mirid Helopeltis sulawesi (see electronic supplementary material, table S4; t = 4.30, p < 0.01), leaf loss owing to herbivores (t = 5.82, p < 0.01) and amount of mealybugs (t = −4.48, p < 0.01), all of which were affected by the ant fauna. The two Dolichoderinae-dominated ant communities reduced fruit damage owing to H. sulawesi (figure 2; electronic supplementary material, table S4; Dolichoderus: z = −15.35, p < 0.01; Philidris: z = −7.71, p < 0.01) compared with control subplots, while in ant exclusion subplots there was no difference in fruit damage by Helopeltis (z = −0.22, p = 0.83). In comparison with control treatments, herbivory rates were reduced in Dolichoderus-dominated trees (figure 2; electronic supplementary material, table S1; t = −4.14, p < 0.01), but elevated in ant exclusion treatments (t = 3.85, p < 0.01) and was higher (with marginal significance) in Philidris treatments (t = 2.68, p = 0.09; electronic supplementary material, table S1). Ant communities with dominance of single species attended more mealybug aggregations per subplot than the communities of the control (see electronic supplementary material, table S2; Dolichoderus: t = 1.02, p ≤ 0.01; Philidris: t = 5.56, p ≤ 0.01). In ant exclusion plots, numbers of trophobiont aggregations were lower (t = −1.37, p ≤ 0.01). A major difference between the ant communities was the distribution of mealybug aggregations within trees, which may influence the spatial ant activity patterns . While Philidris attended mealybugs preferably at fruits and flowers and only with a lower percentage on leaves and shoots (see electronic supplementary material, table S2; t = −2.67, p ≤ 0.01) relative to control, Dolichoderus preferred to transfer them to young leaves and shoots (t = 2.28, p ≤ 0.01; electronic supplementary material, table S2).
(f) Effects on fruit-infesting pathogens
In Philidris subplots, a higher percentage of fruits were lost owing to the phytopathogen P. palmivora than in all other treatments (figure 2; electronic supplementary material, table S4; z = 12.52, p < 0.01). In the inoculation experiment, we showed that Philidris workers and their tent material were infested with spores of P. palmivora (Philidris ants: t = 3.63; p < 0.01; Philidris tent material: t = 6.18; p < 0.01; electronic supplementary material, table S5). Dolichoderus workers had also a higher infection potential than the control (t = 5.67, p < 0.01), but significantly lower than Philidris workers (t = −3.3, p < 0.01; electronic supplementary material, table S5).
(g) Harvested fruits
In control and Dolichoderus treatment, the number of harvested fruits was similar, while in Philidris and ant exclusion treatments around 25% fewer fruits were harvested (figure 3; electronic supplementary material, table S3).
(h) Effects on herbivores reducing bean quality and quantity
The larvae of C. cramerella mine into the fruits and reduce bean numbers and quality . In control and ant exclusion treatments, the percentage of harvested fruits affected by Conopomorpha was similar (z = −1.51, p = 0.13). In trees dominated by Philidris infestation rates were higher than in control treatments (z = 4.7, p < 0.01), whereas in Dolichoderus-dominated trees they were reduced (z = −2.95, p < 0.01; electronic supplementary material, table S4; figure 3). Conopomorpha damage rates were negatively correlated with ant community evenness (z = 6.8; p < 0.01).
(i) Sum of all interactions: ant community effects on marketable yield and revenue
Compared with the control, total marketable yield was 27% lower in ant exclusion treatments, which equals a loss of 875 US$ ha−1 yr−1 (t = −3.34, p < 0.01) and 34% (1109 US$ ha−1 yr−1) lower in Philidris treatments (t = −4.21, p < 0.01, figure 3; electronic supplementary material, table S3). Yields in Dolichoderus and control did not differ significantly (t = −1.27, p = 0.21). Total provided services for agriculture in terms of marketable yield correlated positively with evenness of the associated ant communities (figure 4; electronic supplementary material, table S3; t = 2.7, p = 0.013).
Ants contributed to agricultural productivity by driving a surprisingly complex network of direct and indirect interactions with the crop plant, its pests, pollinators and pathogens, with ant exclusion resulting in substantially lower yields. Their overall impact was higher than in comparable studies from other systems . As expected, ants drove both ecosystem services and disservices, with the balance depending on the ant community composition. We show that traits of dominant species, such as high abundances and spatial activity patterns (fruit versus foliage-oriented foraging) can cause uneven and species poor ant communities to protect crop plants better against some pests, as predicted by Gove . However, prevalence of services over disservices and therefore higher marketable yield and revenue was achieved in subplots with naturally occurring, non-manipulated, species-rich and even control ant communities (figure 4) which provided services to the farmers of 875 US$ ha−1 yr−1 compared with the ant exclusion treatment, where 27% less marketable yield was harvested. In ant communities dominated by single ant species, the effect strongly depended on the identity of the dominant species. We observed (relative to the unmanipulated control) yield losses of 34% (−1109 US$ ha−1 yr−1) in presence of the invasive species Philidris, forming ant communities with very low evenness, and no significant yield losses with the native Dolichoderus, which allowed relatively high ant community evenness. As discussed below, differences in marketable yield can be ascribed not only to changes in pest and disease incidence, but also to more complex effects on fruit set, fruit abortion and interference among herbivores, which were all directly or indirectly influenced by ant community structure and dominant species traits (figure 5).
Our results support the importance of evenness as a driver of biological control , and suggest the benefits of evenness could extend to cases where more complex systems with multiple intermediate ecosystem service are affected by the focal predator group. While the combination of species and interactions we describe is system-specific, it is noteworthy that similarly complex interactions between ants and other organisms have been described from a neotropical coffee agroforest , with evidence that several ant species, in interaction with other organisms, accomplish distinct functions within the system, which is reminiscent of both the complexity of our system and the importance of ant species richness and evenness in our study. Lach  suggests that invasive ants will have idiosyncratic effects on plants, dependent on the traits of the ant species, the system and the season. Our study not only confirms the importance of foraging traits in determining ant effects on predation, but also suggests that in systems which can support diverse ant communities, such as cacao or coffee agroforests, species-rich and even communities may protect the plants from these idiosyncrasies. Such synergistic effects of diversity on ecosystem functions are also reported from other plant animal systems: in diverse pollinator communities, increased pollinator diversity synergistically increase pollination services through complementarity effects as well as species interactions that alter the behaviour and resulting niche shift of pollinator species [35,36].
(a) Ant community characteristics and species traits
We successfully manipulated ant communities over a time period of nearly 2 years. The ant exclusions lowered worker abundances by more than 90%. This is comparable to the reduction rates of other ant exclusion experiments [37,38]. The experimental establishment of Dolichoderus and Philidris led to an ecological and numerical dominance of these species providing 71% (Dolichoderus) and 98% (Philidris) of the workers recruited to the baits. Total worker abundances of both species were similar so that differences in treatment responses are not due to unequal establishment success. The structure, composition and spatial distribution of worker activity of the ant communities are affected by presence and identity of dominant ants [26,34]. The non-invasive ant species Dolichoderus, is relatively tolerant towards other ant species . Therefore, presence of Dolichoderus resulted in ant communities with relatively high species density and evenness, which was comparable to our unmanipulated control ant communities, but Dolichoderus promoted much higher total worker abundances. Distribution of honeydew sources influences the spatial activity patterns of ants . Dolichoderus placed its mealybug aggregations not only at fruits, but also to a high percentage (75%) at young shoots and leaves and built nests between leaves in the foliage, leading to highworker activity in the whole tree. By contrast, the invasive-dominant ant Philidris aggressively displaces most other ant species from the trees . Leading to an uneven and species poor ant community. Because of the mealybug distribution and nesting behaviour, Philidris workers were active mainly on fruits, stem and big branches, but less in the foliage and at small branches of the cacao trees.
(b) Ant community effects on fruit set
Fruit set of cacao flowers was positively related to ant abundance, being lowest in ant exclusion treatments, intermediate on the unmanipulated control trees (intermediate ant abundances) and highest in single species-dominated trees (high worker abundances and activity; electronic supplementary material, table S4). Florivory by herbivores has been observed only rarely in the field, and direct cacao pollination by ants has been under discussion since beginning of the twentieth century , but has not been convincingly demonstrated . While ants are frequently present near the flowers, tending mealybugs at the flower buds, they would have to transfer pollen to neighbouring trees, which is unlikely, thus preventing successful cross pollination. A plausible explanation is the pollinator-disturbance hypothesis [21,41]: nuisance by ants may cause pollinators to switch between flowers and trees more frequently, and thereby enhance pollen transfer, pollen load and number of pollen donors. In self-incompatible plants like cacao, this should cause an increased pollination success and fruit set.
(c) Indirect ant community effects on early fruit abortion
Percentage of young fruits being aborted by the tree increased with amount of mealybug aggregations, damage by the sap-sucking pest Helopeltis and leaf herbivores (see electronic supplementary material, table S4). The ant communities with dominant species facilitated much higher numbers of trophobiont aggregations than control communities (see electronic supplementary material, table S2). Such intensive trophobiont facilitation may negatively affect plant fitness via cryptic herbivory . In Dolichoderus and Philidris-dominated ant communities, we found lower rates of fruits damaged by Helopeltis (figure 2), which spends most of its life cycle at the fruit surface . This can be explained by the high worker abundances and increased worker activity on the fruits in these ant communities. Similarly, the reduced leaf herbivory in the Dolichoderus- and increased leaf herbivory in the Philidris-treatments (figure 2) is presumably the result of differences in spatial activity of the workers, mainly foraging in the tree foliage (Dolichoderus) or on fruits (Philidris). These ant-mediated effects of herbivory on early cacao fruit abortion resulted in lowest abortion rates of young fruits in the unmanipulated control communities, followed by Dolichoderus-dominated ant communities, ant exclusion and Philidris-dominated ant communities (figure 2).
(d) Ant community effects on fruit losses owing to diseases
In Philidris-dominated subplots, about 65% more fruits were lost owing to the plant disease P. palmivora relative to the mean of the other subplots (figure 2). Philidris uses detritus material from infected rotten fruits to build nests and tent structures for protecting trophobiont aggregations at healthy cacao fruits (see electronic supplementary material, figure S3). Dolichoderus builds its nests between leaves without the use of detritus. In inoculation tests, we showed that Philidris workers and their nest material are highly infested with spores of P. palmivora. Dolichoderus workers had a certain infection potential, but it was much lower (see electronic supplementary material, table S5). This suggests that Philidris is an effective vector of the disease, leading to serious yield losses (figure 3). Invertebrate vectors of Phytophthora have been observed before  and Dolichoderus ants were even discussed as possible vector . But here we show that the transmission efficiency is highly associated to specific behavioural traits and may differ widely even between related species.
(e) Ant community effects on pests reducing bean quality and quantity
Philidris-dominated ant communities, with lower evenness and species richness, were associated with high damage by the major cacao pest (figure 2), which significantly reduced bean quality, and thus quantity of marketable beans (figure 3). Philidris showed little activity in the foliage where Conopomorpha hides during daytime . The characteristics of a Dolichoderus-dominated ant community (higher worker abundances activity in the foliage, combined with relatively high evenness and species richness) enable it to suppress Conopomorpha more effectively than control ant communities (figure 2). There is a plant-mediated indirect effect between the two common pests , with the major pest Conopomorpha showing a clear oviposition non-preference towards Helopeltis damaged fruits. Therefore, ant communities particularly efficient in preventing damage by Helopeltis, such as those dominated by Philidris or Dolichoderus, may indirectly facilitate damage owing to the major pest Conopomorpha. Our data confirm this pattern in the case of Philidris, but in ant communities with presence of Dolichoderus, this indirect effect appeared to be partly compensated by direct predation and disturbance of Conopomorpha.
(f) Dominant species traits and ant community properties affect multiple crop plant–antagonist–mutualist interactions and associated services
We show that traits of the dominant, invasive Philidris species such as high abundances and spatial activity patterns (fruit versus foliage-oriented foraging) enhanced pest suppression, as predicted by Gove , However, this was only the case with a minor pest (Helopeltis) and this did not result in increased yield. On the contrary, specific traits of this dominant ant species increased disservices, such as increased fruit pathogen dissemination and reduced leaf herbivore and Conopomorpha suppression, which finally lead to serious yield losses. Ant communities dominated by Dolichoderus still showed high species richness and evenness, possibly due to its relatively low aggressiveness compared with Philidris. Although their high worker density reduced damage by leaf herbivores and the fruit-affecting pests Conopomorpha and Helopeltis, these services were counterbalanced by disservices for example increased herbivory by mealybugs (causing higher abortion of young fruits; figure 2) so that Dolichoderus dominance did not propagate to increased yield (compared with the control; figure 3). Crop deficits under ant exclusion can be ascribed mainly to reduced fruit set and indirectly enhanced young fruit abortion.
High evenness of species-rich predator communities has been assumed to be positively related to functional diversity, and hence biological pest control [11,46,47]. In our experiment, evenness of the ant communities was correlated positively with amount of marketable yield (figure 4; electronic supplementary material, table S3). It is difficult to separate evenness effects from effects of dominant ant species traits in our study, because traits of single species like high aggressivity and dominance both lead to changes in community properties such as low evenness and higher abundances, and were correlated with other traits impacting yield. In this particular case, high evenness counterbalanced disservices associated with single species.
Our results have implications for agricultural management: insecticide use in the tropics, including our study region, has been increasing dramatically . This has no or minimal effect on cacao herbivores , but appears to negatively affect evenness and diversity of ant communities and can favour dominant or invasive species [50,51]. For example, in our study area, special nest-building behaviour (see electronic supplementary material, figure S3) makes the invasive ant Philidris less vulnerable to insecticide use than other ant species, which may have favoured the spread of this detrimental species in the recent past. We show that such changes in community composition can lead to a disruption of the service–disservice ratio and cause substantial crop losses. Our results also suggest that taking a ‘magic bullet’ approach , for example by introducing a single dominant ant as biocontrol agent (e.g. in Dolichoderus sp. [27,44,52]), or promoting just one presumed key pollinator species [35,41], is risky and may not result in yield increases compared with communities of less-conspicuous, yet more evenly distributed species [5,15,35].
In our study, we use yield as a benchmark ecosystem service for evaluating the impact of ant communities differing in species richness and dominance structure and show that ant exclusion negatively affects productivity and yield, and that even, species-rich ant communities result in the highest yield. An invasive, dominant species maximized the control of a particular pest species, but not agricultural yield. While this is not a demonstration that predator diversity will maximize ecosystem functioning in general , these results suggest that taking into account multiple interactions between predators and ecosystems can lead to different conclusions than focusing on individual interactions or single ecosystem services (i.e. predation of pests). From a practical perspective, this suggests that managing predator communities for ecosystem services in agroecosystems should be based on a broad assessment of the relevant final services for agriculture, for example marketable yield, as these integrate the effect of all relevant direct and indirect interactions. Such studies should cover long time periods because effects associated with traits of species or community properties may accumulate in the long term  and vary with the years or season .
This study was supported by the German Academic Exchange service DAAD and the Collaborative Research Centre 990 ‘EFForTS’. A.W., Y.C., S.S. and T.T. designed research; A.W., H.S. and S.S. performed research; A.W., Y.C. analysed output data, A.W. wrote the first draft of the manuscript and A.W., Y.C., B.F. and T.T. contributed substantially to revisions.
We thank our Indonesian representatives in Tadulako University Abdul Rauf, the field and laboratory assistant Nining Winarti and team, village heads and villagers in the study region; the Indonesian and German project coordination team. We thank Ivette Perfecto for comments on a previous version of the manuscript.
- Received October 7, 2013.
- Accepted November 12, 2013.
- © 2013 The Author(s) Published by the Royal Society. All rights reserved.