Ants often interact aggressively for resources (e.g., nest sites and food) with members of their own or another species. In these competitive interactions, dominant ant species exert a strong influence on ant species coexistence and plant-associated arthropod community structure. However, few studies have experimentally manipulated the relative abundance of dominant ant species on plants, preventing a mechanistic understanding of the effects of ant competitive interactions on ant community structure as well as on their interactions with other insects, particularly mutualistic hemipterans. In this study, we performed a field experiment in a tropical dry forest in Brazil to investigate the effects of two dominant ant species (Camponotus crassus and Cephalotes pusillus) on the structure of ant communities and the abundance of the ant-tended hemipteran Enchenopa brasiliensis in the tropical shrub Solanum lycocarpum. For this, we identified and quantified all ant species foraging on S. lycocarpum plants and estimated the number of egg masses, nymphs and adults of the mutualistic hemipteran before and after experimentally removing nests of both dominant ant species. Our results showed that removal of C. pusillus nests significantly changed the community structure of ants foraging on S. lycocarpum plants, whereas removal of C. crassus nests did not. We also found that nest removal of both dominant ant species had significant effects on hemipteran abundance. In particular, plants generally hosted more hemipteran eggs, nymphs and adults after (vs. before) nest removal of both dominant ant species. Overall, this study demonstrates that dominant ant species can play a pivotal role in structuring ant communities and the interactions between ants and honeydew-producing hemipteran insects.

Study area

This study was carried out in a transitional area from cerrado sensu stricto to open areas in the Private Natural Heritage Reserve of Clube de Caça e Pesca Itororó de Uberlândia, Uberlândia, Brazil (48º17'27.0'' W; 18º58' 30.6'' S). The study area had ~230 ha of cerrado sensu stricto vegetation (Oliveira and Marquis 2002). Open areas consisted of shrubs and small trees, whereas more enclosed areas consisted of trees reaching up to 15 meters in height (Del-Claro et al. 2019). The climate is markedly seasonal with the rainy season from October to March, and the dry season from April to September. The annual mean temperature varies from 18 º C to 28º C and the rainfall from 800 to 2,000 mm (Calixto et al. 2021a, b).

Study species

Solanum lycocarpum is a woody shrub commonly found in open areas of the cerrado with environmental disturbances, reaching 1 to 3 m in height (Fig. 1A). This species has hollow trunks, continuous flowering, a berry-like fruit, absence of extrafloral nectaries, and is usually infested by the sap-sucking hemipteran E. brasiliensis (Fig. 1D-E). This hemipteran is absent or in low abundance on S. lycocarpum plants between the months of December and August, and exhibits high abundance between September and November (Stefani and Del-Claro, 2000) (see Fig. S1 in the Supplementary Material). Because this hemipteran insect produces honeydew, it is commonly tended by different ant species such as Cephalotes pusillus (Fig. 1B) and Camponotus crassus (Fig. 1C). Camponotus crassus (worker size: 5–10 mm) is indigenous to many forested parts of the Neotropics (e.g., Argentina, Brazil, Colombia, Paraguay, and Peru). This ant species nests in the ground, often building several small satellite nests (Lange et al. 2019). The distribution of nests is aggregate, with density of 450 nests/ha and average distance of c. 4 m between nests (Lange et al. 2019). Workers forage for up to 150 min out of the nest, covering a route of up to 9 m from the entrance of the nest (Lange et al. 2019). On the other hand, C. pusillus (worker size: 3–7 mm) is sympatric with C. crassus throughout most of its distribution range. This ant species nests in sticks, dry branches or hollow trunks of many tree species, including S. lycocarpum (Powell 2008). Mature colonies have only one queen, 170 workers, and 25 soldiers (Byk and Del-Claro 2011). Foraging trails extend from trees with nests to trees with food sources, covering a route of up to 8 m from the entrance of the nest (Powell 2008). Because they occur at large proportions in plants, both ant species are usually classified as numerically dominant species in the cerrado (Powell 2008, Lange and Del-Claro 2014, Fagundes et al. 2017, Camarota et al. 2020, Calixto et al. 2021a). These species do not coexist in abundance on the same S. lycocarpum individual plants when aggregations of E. brasiliensis are present (Costa-Silva, personal observation).

Experimental design

On October 9, 2017, we selected 20 adult S. lycocarpum plants with similar phenological status (1–2 meters tall, 50% young leaves, and 50% adult leaves) and a minimum distance of 15 meters from each other. We selected this distance to avoid plant clones. Ten of the selected plants were dominated by C. crassus, and the other 10 plants were dominated by C. pusillus. All plant individuals had aggregations of the ant-tended hemipteran E. brasiliensis. Immediately after selecting experimental plants (“before ant nest removal” hereafter), we identified and counted all ants on each plant, and estimated the total number of eggs, nymphs, and adults of the ant-tended hemipteran E. brasiliensis. For this, we carried out observations for 30 minutes on each plant from 8:00 am to 12:00 am because it was the time of the day with the highest activity for both ant species (Lange and Del-Claro 2014, Lange et al. 2017, Lange et al. 2019). 

On October 16, 2017, we removed nests of C. crassus and C. pusillus on experimental plants (“after ant nest removal” hereafter). In the case of C. crassus, we excavated 10 ground nests (the most distant nest was 5 m away from the focal plant) and transferred them 200 m away from their original location. Before the excavation of C. crassus nests, we monitored individuals of this species that foraged on host plants and their nests to verify satellite nests. To ensure that dominant ants were not from the same colony, we checked that the queen and/or chambers with immatures (eggs, larvae, and pupae) were also removed. Chambers with immature nests are usually associated with the presence of the queen (Lange et al. 2019). In the case of C. pusillus, we pulled out S. lycocarpum branches with C. pusillus nests and deposited these branches 200 m away from their original location. For both nest removal treatments, we also manually removed all ants present on the plants. During three consecutive weeks after removing the nests of both dominant ant species (27^th October, 3^rd November, 11^th November), we carried out observations for 30 minutes on each plant to identify and count all ant species and estimate the total number of eggs, nymphs, and adults of the ant-tended hemipteran E. brasiliensis. 

As a subsidiary test, we performed another field experiment to estimate aggressiveness of both C. crassus and C. pusillus ants. On October 9, 2017, we selected another set of 20 plants at the same field site used for the previous experiment. Ten of the selected plants were dominated by C. crassus, and the other 10 plants were dominated by C. pusillus. Again, all plant individuals had aggregations of the ant-tended hemipteran E. brasiliensis. Immediately after selecting the plants, we recorded the number of times each C. crassus or C. pusillus worker bit or chased non-nestmate ants or other insects (“agonistic interactions” hereafter) on plants. Again, we performed these observations for 30 minutes on each plant from 8:00 am to 12:00 am. Because >95% of the agonistic interactions were between ant species, we only included such ant-ant interactions in the statistical analyses. 

Statistical analyses

To investigate the effect of dominant ant species on ant community structure, we tested for the effects of ant nest removal treatment (four levels: before ant nest removal, and one, two and three weeks after ant nest removal) on the relative abundance (i.e., percentage) of each ant species, separately for plants dominated by C. crassus or C. pusillus. For this, we performed non-metric multidimensional scaling (NMDS), and Similarity Analysis (ANOSIM) with Euclidean distance and 999 permutations including ant nest removal treatment as a fixed factor. 

To investigate the effect of dominant ant species on the abundance of the hemipteran insect E. brasiliensis, we ran generalized linear mixed models (GLMMs) with a Poisson distribution testing for the effects of ant nest removal treatment (four levels: before ant nest removal, and one, two and three weeks after ant nest removal) on the number of eggs, nymphs and adults of E. brasiliensis, separately for plants dominated by C. crassus or C. pusillus. In all these models, we included the total number of ants in each plant as a covariate to control that changes in hemipteran abundance were attributable to shifts in ant community composition, rather than a reduction in attendant ants. In addition, we also included individual plant as a random factor to account for repeated measures taken from each plant throughout the sampling period (i.e., before and after ant nest removal treatments). 

To investigate aggressiveness of both C. crassus and C. pusillus ants, we ran a generalized linear model (GLM) with a Poisson distribution testing for the effect of the identity of dominant ant species (two levels: C. crassus or C. pusillus) on the number of agonistic interactions between the dominant ant species and the other species. In this model, we included the total number of ants in each plant as a covariate because antagonistic interactions between ants depend on how many ants are present on the plant.

We performed all analyses using R software version 4.0.1 (R Core Team 2020). We ran NMDS and ANOSIM using the “vegan” package (Oksanen et al. 2016). We implemented GLMMs and GLMs using the lmer and lm functions, respectively, from the lme4 package (Kuznetsova et al. 2017). For these GLMMs and GLMs, we reported back-transformed least-square means and standard errors from these models using the lsmeans function from the lsmeans package (Lenth 2016). In all the above models, if the treatment effect was significant we conducted post-hoc comparisons (using emmeans package) to test for pairwise differences between treatments.

We performed all analyses using R software version 4.0.1 (R Core Team 2020).