Chemical camouflage: a key process in shaping an ant-treehopper and fig-fig wasp mutualistic network

Different types of mutualisms may interact, co-evolve and form complex networks of interdependences, but how species interact in networks of a mutualistic community and maintain its stability remains unclear. In a mutualistic network between treehoppers-weaver ants and fig-pollinating wasps, we found that the cuticular hydrocarbons of the treehoppers are more similar to the surface chemical profiles of fig inflorescence branches (FIB) than the cuticular hydrocarbons of the fig wasps. Behavioral assays showed that the cuticular hydrocarbons from both treehoppers and FIBs reduce the propensity of weaver ants to attack treehoppers even in the absence of honeydew rewards, suggesting that chemical camouflage helps enforce the mutualism between weaver ants and treehoppers. High levels of weaver ant and treehopper abundances help maintain the dominance of pollinating fig wasps in the fig wasp community and also increase fig seed production, as a result of discriminative predation and disturbance by weaver ants of ovipositing non-pollinating fig wasps (NPFWs). Ants therefore help preserve this fig-pollinating wasp mutualism from over exploitation by NPFWs. Our results imply that in this mutualistic network chemical camouflage plays a decisive role in regulating the behavior of a key species and indirectly shaping the architecture of complex arthropod-plant interactions.

them less likely to be preyed on by the ants 14,16 . However, studies on the role of chemical signals in maintaining the stability of mutualisms and regulation predation behavior of one species are uncommon.
Weaver ants (Oecophylla smaragdina) often build leaf nests on fig trees 4 . These ants are voracious predators of other small insects 17  smaragdina can help prevent some NPFWs from out-competing pollinators by selectively preying upon those species that are independent gall makers 4 , and may also reduce the likelihood of the larvae of pollinating fig wasps from being attacked by NPFW parasitoids 18 . The aggressive, predatory behavior of O. smaragdina towards some NPFWs differs from that often directed at honeydew producing hemipterans, which concurs with the classical food-for-protection ant-hemipteran model (Digital appendices 1 and 2).
It has been frequently reported that chemical resemblance (mimicry or camouflage) can manipulate ant behavior to favour many species of phytophagous insect mutualists, making them less likely to be preyed on by the ants 14,16 . Here, we used a weaver ant-treehopper mutualism associated with the fig tree species Ficus racemosa in southwestern China to test if the behaviors of the ants are affected by chemical signals given by treehopper mutualisms. We hypothesized that chemical signals produced by the treehoppers may manipulate the behavior of the ants in their favor and thus help maintain the stability of this system. We attempted to answer five questions (1)

Discussion
Ants can play important roles in terrestrial ecosystems because they often interact closely with multiple species. Ant behavior is sensitive to chemical signals from con-specifics, symbionts, prey, predators, and the wider environment in which they forage and build nests 19 . Behavior responses of ants to one or more of these factors may indirectly affect the interspecific interactions with other species of a community of which the ants are a component. Our results show that the cuticular hydrocarbons of treehoppers on one hand attract mutualistic weaver ants, but on the other hand help protect the treehoppers from ant predation even in the absence of honeydew rewards. Our data suggest that this reflects the similarity between treehopper cuticular hydrocarbon profiles and  Camouflage is the simulation of the general background, which is a common strategy to reduce ant predation in phytophagous insects 21   demonstrated "tending behavior" to F. racemosa branches. Because fig inflorescence branches are a background object for O. smaragdina, we suggest that chemical camouflage explains to a large degree the behaviour of O. smaragdina towards Tricentrus nymphs.
Although the surface chemicals of Tricentrus and small fig tree branches have the smallest pair-wise NMDS distance for all the five species we measured, the difference of in overall chemical composition between Tricentrus and fig tree branches is obvious. However, only a subset of chemicals is responsible for insect inter/intra specific recognition and similarity in these info-chemicals will achieve chemical camouflage 22 . Moreover, we also demonstrated experimentally that Tricentrus cuticular hydrocarbons and surface chemicals from fig trees can independently protect Tricentrus from ant attack in the absence of any honeydew rewards, and therefore reduce punishment by ants for non-cooperative behavior. However, the precise mechanism for this camouflage remains unknown. Whether Tricentrus treehoppers sequester compounds from the plants or biosynthesize them independently requires further investigation, as does the function of the individual chemicals responsible for affecting ant behavior. Some chemicals present in both treehoppers and fig inflorescence branches may function as camouflage chemicals but others may attract ants, and both quantitative and qualitative reactions may be involved.
In general, mutualisms between ants and hemipterans are not species specific 12 . Ants can learn to associate honeydew production with the chemical profiles of hemipteran (aphid) cuticles through chemically-mediated associative learning 23 . Such an association can be established in short time periods in the laboratory, with some ant species even showing tending behavior towards dead hemipterans which fail to respond to any stimuli from the antennae of ants, as long as the hemipteran cuticular chemical profiles remain intact 13 . We found that O. smaragdina showed tending behavior to filter papers treated with chemicals from either Tricentrus or fig inflorescence branches, which indicated that chemical signals can help enforce the ants to behave cooperatively towards their treehopper mutualists. Our data therefore suggest an association between treehopper chemical cues and the production of honeydew, which may reduce any costs of association incurred by the treehoppers. Ant associative learning may contribute to the persistence of this mutualism 24 , with chemical camouflage being used by treehoppers to avoid costs inflicted by ants.
One factor that contributes to stability in mutualisms is actively rewarding cooperation and/or punishing less mutualistic behavior 25 . In some cases, sanctions against less mutualistic partners occur 25,26 , although the implementation of sanctions may be costly and mutualistic partners that provide reduced benefits to hosts still provide net benefits 26 . In the weaver ant-treehopper mutualism, ants may prey on hemiptera when food resources are limited 12 , which may act as an indirect sanction by ants to less cooperative hemipterans. However, predation on hemiptera may result in a net cost to the ants due to less food in the long term, which may be detrimental to the mutualism. Sakata 11 reported that if aphids provided potentially aggressive ants honeydew, and the honeydew was consumed, predation rates were lower than when the aphids failed to provide honeydew. Individual treehoppers might temporarily be unable to pay honeydew to weaver ant partners, because of a lack of resources, time constraints, or poor physical condition. Chemical camouflage may reduce any errors in the application of sanctions by ants against non-or less cooperative treehoppers, and may thus be important in helping maintain the stability of this system. Although a possible consequence of chemically mediated behavioral manipulation in this system is to reduce the effectiveness of resource/service exchange processes, it may also underpin the mutualism between weaver ants and treehoppers. Contrary to previous work 27 , we suggest that chemical camouflage may not only be a consequence, but also a possible contributory factor enabling a stable inter-specific mutualism. Chemical signaling is thus crucial in regulating ant behaviour and shaping this mutualistic network (Fig. 6).

Materials and Methods
Location and study system. The study was carried out in the Xishuangbanna Tropical Botanic Garden ants also frequently tap on the bracts, which suggest that ants may mistake tree branches for treehoppers. In contrast to their tending of honeydew-producing Tricentrus, O. smaragdina is an aggressive predator of other insects closely associated with F. racemosa trees, including fig-pollinating wasps and NPFWs (Digital appendix 2). pers, and from the external surfaces of small F. racemosa twigs. All insects and twigs were collected from the same three F. racemosa trees as previously used for the behavioural association assessments. Samples were analyzed with a Hewlett-Packard 7890 gas chromatograph-mass spectral (GC-MS) detector.

Behavioral associations between ants and treehoppers.
Figs in the male floral phase (mature figs) were collected and put into a mesh bag to enable the female wasps to be collected after emergence. Thirty wasps per species were immobilized by cooling, placed into a 4 ml Agilent sample vial, and then immersed in 200 µl hexane (chromatographic purity, Merck) for 10 min to dissolve their cuticular hydrocarbons. Only two wasp species were used, the pollinator C. fusciceps and the galler P. mayri, because these are the two wasp species associated with F. racemosa most likely to be preyed upon by O. smaragdina 4 . All weaver ants and Tricentrus sp. were also anesthetized by cooling. For each species, three O. smaragdina or three Tricentrus sp. were immersed in 200 µl hexane in a 4 ml vial for 10 min. Small F. racemosa inflorescence branch tips (including both the branch and inflorescences buds), each approximately 5 cm long and 0.5 cm in diameter, were removed from a branch. The cut end of each branch was wrapped with cotton wool soaked in tap water to prevent dehydration. In turn, five twigs were sequentially immersed in 4 ml of hexane with each given 10 min to dissolve the surface chemicals. The solvent was then concentrated to 400 µl, using a flow of purified nitrogen. All of the resulting solutions were then filtered with anhydrous sodium sulfate in a glass tube, to remove any water. Each sample (species) had three replicates (three vials).
A sample of each solution (0.2-0.8 µl) was injected into a Hewlett-Packard 7890 gas chromatograph-mass spectral (GC-MS) detector, equipped with an HP-5MS column (5% Phenyl methyl silox, 30 m × 250 µm × 0.25 µm). The temperature program started at 80 °C, and increased at 10 °Cmin-1 until 160 °C, then at 4 °Cmin-1 until 280 °C, and a further 10 °Cmin-1 until 300 °C was reached, and then held for 10 min. The flow of the nitrogen carrier gas in the column was 1.0mlmin-1. Aliquots of extracts were injected splitless at 250 °C. The areas of regularly occurring peaks (those detected at least twice in all of the samples from each source) were included in the subsequent identification and analysis. Chemical profiles were evaluated using the software package Enhanced ChemStation G1701DA Version D.00.01.27. Peaks with the same retention time and mass spectra were considered as the same chemical substances 33 . Each chemical was identified using mass spectra matching with standard mass spectra in the NIST 02 library, and using arithmetic indices (AI) which were calculated based on previously published data on n-alkanes 33 . The relative percentage of each chemical of the total compounds in each sample was calculated according to its percentage peak area of the total peak area of all compounds. of fig wasps, Tricentrus, and F. racemosa? We measured the behavioral responses of ants to filter papers that had been treated with the crude hydrocarbon extracts from the cuticles of treehoppers, fig wasps, or ants, or the external surfaces of small fig branches. For each of two wasp species (C. fusciceps and P. mayri), 60 individuals were placed into a 4 ml sample vial. We then added to each vial 300 µl of hexane, which was left for 10 min to extract the cuticular hydrocarbons of the insects. For the plant extract, five small tree branches were rinsed sequentially with 4 ml hexane for 10 min each, and the resulting solution concentrated to 400 µl. For the ant and leafhopper extracts, three individuals per species were put into a 4 ml sample vial. We then added 300 µl hexane, which was left for 10 min. All insects and branches were removed from the vials after extraction.

Are ants attracted to the chemical bouquets
For the behavioural assays, five microlitres of the crude extracts from the wasps, ants, treehoppers, or inflorescence branches were each placed onto one 5 mm × 5 mm square of filter paper. All papers were left to dry for 1 min. The five filter papers were then stapled to one large F. racemosa branch. Specific positions of filter papers with chemicals were randomly assigned in order for each replications by pseudorandom number. The number of ants that each made contact with each paper was recorded every minute for 15 minutes and all counts were summed for total scores.
A preliminary experiment showed that a hexane-only treated filter paper had no significant effect on ant behaviour compared with a dry filter paper. In a dual choice assay, ants simply removed with equal likelihood dry or hexane treated filter papers (60 replicates). However, in the main experiment the ants were more likely to carry the filter papers with fig wasp cues back to their nest. We thus used only a dry filter paper of equal size to the treated papers as a control in each replicate. The assay had 11 replicates on three different F. racemosa branches. Each branch was on a separate tree, with each tree having at least three replicates. replicates spread over three different F. racemosa trees. Several treehoppers (4th or 5th instar nymphs) were killed by being placed into a freezer at −80 °C for 5 min. Each nymph was then randomly assigned to one of two groups.
The aim of our first assay was to test if the cuticular chemical profiles of Tricentrus sp. offer protection from ant predation. For controls, nymphs were un-manipulated; for a treatment group, each nymph was rinsed in 100 μl hexane for 10 min to remove its cuticular hydrocarbons and then dried with purified nitrogen gas.
To measure the effects of the solvent (hexane) application, we performed a second assay, again involving two groups. For the first group, nymphs were rinsed with 100 μl hexane for 10 min. The hexane in each vial was then evaporated with purified nitrogen gas, so the cuticular hydrocarbons could re-adhere to the outer surface of the insects. For the second group, nymphs were rinsed in 100 μl hexane for 10 min to remove their cuticular hydrocarbons and then removed from the solvent and dried with purified nitrogen gas 14 .
To test if surface chemicals from F. racemosa inflorescence branches elicit similar behaviors from ants as do Tricentrus sp. cuticular hydrocarbons, we performed a third assay. For one group, the cuticular hydrocarbons of Tricentrus sp. nymph were removed as previously described; for the second group, cuticular hydrocarbons of Tricentrus sp. nymphs were again removed, but replaced with surface chemicals from F. racemosa branches. These F. racemosa surface chemicals were obtained by first rinsing in 4 ml hexane for 10 min sequentially five small inflorescence branches. Three randomly selected Tricentrus sp. nymphs were then rinsed with the resulting solution, and the solvent removed by drying with purified nitrogen gas 14 .
For the behavioural assays, each dead nymph was stuck to a 5 mm × 5 mm square filter paper using 0.5 μl of liquid glue, which was left to dry for 15 min. One large branch on a mature F. racemosa tree was selected and two areas were specified on the branch. Two papers, one for each nymph group, were pinned to each area. Test areas were reversed between treatment groups between repeats. Each assay lasted for up to 15 min. Within this time period, we recorded which of the two Tricentrus sp. was removed first by the ants 14,15,34 . The effects of ant-treehopper mutualists on the fig wasp community. We conducted a field experiment to test the effects of the ant-treehopper mutualism on the fig wasp community of F. racemosa. We recorded ant and treehopper abundances for each fruit branch, then categorized them into four levels of ant-treehopper abundance: (1) an ant-free treatment for which no ants and treehoppers were observed during the period (5 weeks); (2) a 'few ants' treatment, for which about 6.17 ± 3.76 (mean ± SD) ants and 1.00 ± 0.63 (mean ± SD) treehoppers were observed during the period, and (3) a 'many ants' treatment, for which about 20.20 ± 5.07 (mean ± SD) ants and 5.80 ± 1.64 (mean ± SD) treehoppers were observed. (4) To test the absolute effect of ant-treehopper presence on the fig-fig wasp mutualism, we also used a fourth treatment, which excluded ants by the use of flavorless, colorless rat-stop glue (Yiwu Xinqi Super Glue Product Factory, Zhejiang, China) 4 and all treehoppers were physically removed at the start of the experiment.
After approximately four weeks, the near-mature figs were collected and put into fine mesh bags for 24 h to collect the dispersing wasps. All wasps were then killed and stored at −20 °C. All wasps were sorted by species and sex, identified to species and counted. After wasp emergence, each fig was cut longitudinally into four equal parts. Seeds from one quarter (randomly chosen) were then counted. The total number of seeds per fig was estimated by multiplying the quarter segment count by four 4 .

Data analyses. The relationship between Tricentrus and O. smaragdina abundances were quantified with
Pearson's correlation.
Cluster analysis was used to classify chemical profiles of each sample into different groups based on Horn-Morisita distances. The clustering strategy used was the un-weighted pair-group method, using arithmetic averages (UPGMA). We then used non-metric multidimensional scaling (NMDS) to visualize chemical profile differences (Horn-Morisita distance) in two dimensions. Before analysis, we performed a square-root transformation and then used Wisconsin double standardization to normalize the data.
In the attractiveness of chemicals to ants experiment, we used a generalized linear mixed model (GLMM), with 'tree' as random factor, 'chemical treatment' as a fixed factor, and the number of ants as the response. Ant number data were modeled using a Poisson distribution family using the log link function. The pairwise comparisons between 'chemical treatment' using Tukey's HSD (honest significant difference).
For the dual choice behavioral assays, chi-square tests were used to test the difference in the frequencies of Tricentrus removed by the ants. The null hypothesis was that ants were equally likely to remove Tricentrus regardless of treatment.
We used analysis of similarity (ANOSIM) to examine how ant-treehopper abundances affect wasp community structure between different levels of ant-treehopper abundance based on Bray-Curtis distances and each test ran for 999 permutations. The effects of ant-treehopper abundances on the abundances of six species of fig wasps and seeds were quantified with generalized linear models. For each model, ant-treehopper abundance levels were fixed factors. 'Tree' was included in each model as a random factor to control for any effects of unmeasured variation among trees. Interactions between explanatory variables were not included in any model. Multiple pair-wise-comparisons between different levels of ant-treehopper abundances were performed using the Student-Newman-Keuls method. All analyses were conducted with R 3.0.3 35 . Data and materials availability. All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials (https://doi.org/10.6084/m9.figshare.1025771). Additional data related to this paper may be requested from the authors.