Dominance in a ground‐dwelling ant community of banana agroecosystem

Abstract In tropical ecosystems, ants represent a substantial portion of the animal biomass and contribute to various ecosystem services, including pest regulation and pollination. Dominant ant species are known to determine the structure of ant communities by interfering in the foraging of other ant species. Using bait and pitfall trapping experiments, we performed a pattern analysis at a fine spatial scale of an ant community in a very simplified and homogeneous agroecosystem, that is, a single‐crop banana field in Martinique (French West Indies). We found that the community structure was driven by three dominant species (Solenopsis geminata, Nylanderia guatemalensis, and Monomorium ebeninum) and two subdominant species (Pheidole fallax and Brachymyrmex patagonicus). Our results showed that dominant and subdominant species generally maintained numerical dominance at baits across time, although S. geminata, M. ebeninum, and B. patagonicus displayed better abilities to maintain dominance than P. fallax and N. guatemalensis. Almost all interspecific correlations between species abundances, except those between B. patagonicus and N. guatemalensis, were symmetrically negative, suggesting that interference competition prevails in this ground‐dwelling ant community. However, we observed variations in the diurnal and nocturnal foraging activity and in the daily occurrence at baits, which may mitigate the effect of interference competition through the induction of spatial and temporal niche partitioning. This may explain the coexistence of dominant, subdominant, and subordinate species in this very simplified agroecosystem, limited in habitat structure and diversity.

However, intra-and interspecific interactions also play an important role in the structuring of ant communities (Fellers, 1987(Fellers, , 1989.
This may be particularly true in homogeneous environments such as agroecosystems.
Intra-and interspecific competition can be divided into two types: interference competition, which includes all direct interactions involving aggressive encounters between ants (Fellers, 1987;Kenne et al., 2005), and exploitative competition, when the consumption of a limiting resource by one species reduces the availability of that resource for another species (Fellers, 1987;Kenne et al., 2005). Researchers have hypothesized that the coexistence of ants results from a tradeoff between traits linked to interference competition and those linked to exploitative competition. For example, Fellers (1987) hypothesized that there is a trade-off between bait discovery and bait dominance, that is, those ant species adept at finding resources have poor interference competitive abilities while those species adept at dominating a resource have good interference competitive abilities but poor resource discovery abilities. However, this trade-off seems to be the exception rather than the rule (Castracani, Spotti, Grasso, Fanfani, & Mori, 2014;Parr & Gibb, 2012), and positive correlations between discovery and dominance have been reported (Parr & Gibb, 2012).
Interspecific interactions may determine which species are members of particular ant communities. In tropical ant communities, apart from the major arboreal and terrestrial guilds that have specialized foraging habitats, some species forage both in tree canopies and on the ground and may therefore compete (Bluthgen & Feldhaar, 2010). Ant species differ in competitive ability because of differences in foraging activity, colony size, or body size (Hölldobler & Wilson, 1990). In particular, dominant ants can alter the structure of ant communities by interfering in the foraging activity of other ant species (Savolainen & Vepsalainen, 1988). Dominant ants often have mutualistic associations with nonant herbivores that provide honeydew as a sugar source in exchange for protection against predators (Bluthgen, Stork, & Fiedler, 2004). Such mutualisms enable ants to build large colonies with many nests (Richard, Fabre, & Dejean, 2001). Dominant ants achieve superiority because of their aggressiveness, numerical dominance, superior interference behavior, and superior ability to participate in exploitative competition (Parr & Gibb, 2010); such ants are frequently found in disturbed habitats including intensive agroecosystems (King & Tschinkel, 2006). The spatiotemporal dynamics of such dominant ants greatly affect ant community structure (Zakharov, 2002).
In the humid tropics, bananas (Musa AAA genome) are mostly grown on bare soil and as a single crop. These semi-perennial agroecosystems contain regularly spaced banana plants and are extremely simple and homogeneous and, therefore, are well suited for studying ant community structure. In the current study, we performed a pattern analysis at a fine spatial scale and provided information on temporal and spatial dynamics of ants foraging in a single-crop banana agroecosystem: (i) We assessed the diurnal and nocturnal foraging activity of these species; (ii) determined which species are dominant, subdominant, and subordinate; (iii) assessed how numerical dominance at an impermanent resource (i.e., a bait) evolved through time; and (iv) assessed how abundance of species, at baits and in the neighborhood of the baits, were correlated.
The sampling area or plot was 44 m long and 20 m wide. The banana crop (Cavendish Grande Naine) was in its first cycle when the data were collected to ensure homogeneity across the plot. The age of the banana plantation was 8 months. The climate at the study site is humid tropical with a mean (± SE) monthly temperature of 26.5 ± 0.3°C and a mean monthly rainfall of 174.6 ± 21.2 mm. Within the main plot, we defined 60 regularly spaced subplots (14.7 m 2 each; Figure 1).
F I G U R E 1 Schema of the experimental design. Gray squares: ceramic tile with baits; black circles: pitfall traps. Subplots correspond to areas delimited by dashed lines

| Ant sampling: day and night pitfall trapping
To assess the general foraging activity of the ant community, we carried out diurnal and nocturnal pitfall trapping. Pitfall traps were situated between each subplot, with a total of 54 pitfall traps (Figure 1).
Pitfall traps contained 50 ml of water with a drop of dishwashing liquid. Pitfall traps were deployed at nightfall at 6:00 p.m., and the trapped ants were collected at 6:00 a.m. and conserved in 70% alcohol for identification. Next, pitfall traps were washed with water and refilled. Trapped ants were then collected at 6.00 p.m. and conserved in 70% alcohol for identification. This sampling method was replicated three times. For day and night period, we calculated the proportion of pitfall traps where the species were recorded as well as their mean abundance per pitfall trap. Next, we used Kruskal-Wallis tests to analyze the day-night variation in foraging activity.

| Ant sampling: baiting
One 30 × 30 cm white ceramic tile was placed at the center of each subplot, such that the tiles were regularly spaced across the main plot.
In each subplot, we measured ant abundance using canned tunahoney baits. One bait, which had a diameter of 4 cm, was placed in the center of each ceramic tile on the 23 of April 2012, which we called hereafter date 1. Each subplot was sampled 30, 90, and 180 min (sampling times 1, 2, and 3, respectively) after the baits were deployed.
At each sampling time, we identified and counted the individuals of different species present on the tile. The ants were also recorded according to a six-point abundance scale (following Andersen, 1997;Baccaro, Ketelhut, & De Morais, 2010;Parr, Sinclair, Andersen, Gaston, & Chown, 2005). We used the percentage of bait controlled as a measure of dominance (following Bestelmeyer, 2000;Baccaro et al., 2010) rather than strict bait monopolization (Andersen, 1992). Baits were considered controlled by a species (i) if the number of individuals is >20 and no other ant was present (monopolization) or (ii) if one species was at least twice as numerous as the second numerous taxon when several species were present and the total number of individuals was >20). Samples of all species were collected and conserved in 70% alcohol, then we performed identification to genus according to the Bolton key (Bolton, 1994), and all ants were sent to J. Delabie (Laboratory of Myrmecology, UESC/CEPLAC, Brazil), who identified the ants to species. Sampling was performed between 8:00 and 11:00 in the morning and between 2:00 and 5:00 in the afternoon and repeated the 12 of May 2012, the 31 of May 2012, and the 19 of June 2012, which we called, respectively, hereafter dates 2, 3, and 4. For each species, we used Kruskal-Wallis tests to analyze the daily variation (morning vs. afternoon) in abundances.

| Dominant analysis
Following Baccaro et al. (2010), we used a combination of numerical and behavioral criteria of dominance to determine dominant, subdominant, and subordinate ants. An ant species was considered as a dominant species when (i) it occurred at a large proportion of baits; (ii) it controlled a large proportion of baits whenever it was present; and (iii) it had a high mean abundance score (Andersen, 1992;Baccaro et al., 2010;Parr, 2008). The dominant (respectively, subdominant) species were classified as those that were recorded in >10% of all baits, controlled >25% (respectively, >10%) of baits where they occurred, and with a mean abundance score (calculated by dividing the sum of the abundance scores for the species at all baits by the number of baits at which the species was present) of >3 (respectively, >2.5).
All other species that did not meet all these criteria were considered as subordinate species.

| Data mining and probability matrices of transition
We first performed a principal component analysis and a hierarchical cluster analysis, based on Ward's method (Ward, 1963), to classify the observations at baits according to their similarities in species distribution and abundance. We thus obtained a hierarchical classification of observations for each replicate tile at one sampling time. Once the partition of observations into groups on the basis of the Ward's minimum variance agglomerative clustering was obtained, we used that partition as the initial value for K-means partitioning (Murtagh & Legendre, 2014). This procedure was used to identify the type of "momentary" communities that we observed at baits and to established transition matrices. Moreover, the type of "momentary" community (which we called hereafter a group) observed at sampling time 1 are likely to be indicative of the colonies that was spatially close to the subplot (see below in Subplot scale dynamics and correlations between ant species abundances).
Once groups of observations were identified, we named these groups according to the numerically dominant species of each group.
We next used the Markov chain approach to define the probability (in the form of probability matrices) that a numerically dominant species will maintain its numerical dominance on the resource or lose it to the benefit of another dominant or subdominant species from sampling time 1 to sampling time 2 and from sampling time 2 to sampling time 3. We then used 10,000 Markov chain simulations to obtain probabilities that each species will maintain numerical dominance from sampling time 1 to sampling time 3. Probability matrices of transition were obtained using the "markovchain" R-package (Spedicato, 2015).

| Autocorrelation
It is important to consider autocorrelation in studies of species interactions (Dormann et al., 2007), and the omission of spatial autocorrelation in analyses may lead to false conclusions (Kuehn, 2007). For the abundance of each of the identified dominant and subdominant species, we tested the spatial autocorrelation using the C index of Geary (1954). Low values of this index indicate that two locations are positively correlated, that is, that they are more likely to resemble each other. High values of this index indicate the absence of correlation between two locations. The C index of Geary was calculated for the abundance of each dominant and subdominant species for each replicate tile at each sampling time and for distances d ranging from 2.5 to 6.0 m. The effects of distance on the Geary C values were assessed with linear models (LMs) that took the form of C = α + β 1 d + β 2 d 2 + ε where α was the model intercept, β were regression coefficients, and ε was the normally distributed error term.

| Subplot scale dynamics and correlations between ant species abundances
We used Poisson generalized mixed-effect models (GLMMs) to assess correlations between abundances of dominant and subdominant species, which may reflect the outcome of species interactions at baits.
Following Zuur, Ieno, Saveliev, and Smith (2009), we first tested for collinearity (i.e., correlation) between covariates using the variance inflation factor (VIF) method. Once the set of explanatory variables (fixed effects) was determined, we tested the random effect structure. Here, for each species i, we defined as the fixed effect, the local abundances of species j (i.e., abundances in subplots), the mean abundances of conspecifics of i and of species j in the neighborhood (i.e., abundances in the surrounding subplots situate at a distance defined for each species by the spatial autocorrelation analysis) of the considered subplot, and the sampling time.
Humidity, temperature, and other factors may cause the abundance of a species to be similar between subplots at the same sampling time on one date. To consider the nonindependence between data from subplots at the same sampling time on one date, we introduced this latter variable as a random intercept effect. Moreover, the ants observed at the sampling time 1 are likely to be indicative of the community that was spatially close to the subplot and the data from the three sampling times at a specific subplot are likely to be correlated. Therefore, to consider the nonindependence between sampling times at a same subplot, we introduced the group at sampling time 1, obtained by the data mining procedure, for each subplot as a random slope effect on sampling time. Following Zuur et al. (2009), we tested random effect structures by comparing nested GLMMs comprising all fixed effects.
We used Akaike information criteria (AIC) and likelihood ratio tests (LRTs) to select the best random effect structure of the model for each ant species (Bolker et al., 2009).
After using the GLMMs to determine the best random structure for each species, we selected the best model by removing nonsignificant fixed-effect parameters in a backwards-stepwise process using LRTs. The selection procedure was continued until a model was found in which all effects were significant (Zuur et al., 2009).
All LMs and GLMMs were estimated using the "glmer" function in the "lme4" package (Bates, Maechler, & Bolker, 2012), in which the maximum likelihood of parameters is approximated by the Laplace method (Bolker et al., 2009). All statistical analyses were performed with R 2.15.0 (R Development Core Team, 2014) and with an alpha level of 0.05.

| Diurnal and nocturnal foraging activity
A total of 6,229 ants belonging to 11 species were collected with pitfall traps. Among the 11 species trapped in pitfall traps, we found only two specimens of Tetramorium bicarinatum (Nylander), which species was not included in the analysis.  Table 2).
Maps of spatial distribution of species are provided in the Appendix ( Figure A1).

| Dominant, subdominant, and subordinate species
A total of 10 species were recorded during bait experiments. These species were the same as recorded in pitfall traps with the exception of T. bicarinatum, which was never recorded at baits. At baits, the presence of all species was in relatively high proportion, ranging from 10% to 77% (Table 3). S. geminata, N. guatemalensis, and M. ebeninum were identified as dominant species because they controlled a large proportion of baits at which they were present and have a high mean score abundance (Table 3). P. fallax and B. patagonicus were identified as subdominant species because they controlled a moderate proportion of baits at which they were present and have a moderate mean score abundance (Table 3). All other species were identified as subordinate species (Table 3). Maps of spatial distribution of species are provided in the Appendix ( Figure A2).
The abundance of S. geminata, M. ebeninum was greater in the afternoon than in the morning, while the abundance N. guatemalensis was greater in the morning than in the afternoon (Table 4). We found no difference between the abundances at baits in the morning and in the afternoon for P. fallax and B. patagonicus (Table 4).

| Probability of transition matrices
Based on hierarchical classification and clustering methods, only the three dominant and the two subdominant species contributed to the community structure at the subplot scale. We characterized seven typical groups corresponding to different community structures at the subplot scale (Table A1). Five of the groups were numerically dominated by one species, that is, groups 1-5 were numerically dominated  (Table 5). For each species, most of the subplots where numerical dominance was not maintained were taken over by group 7, that is, the group that lacked a numerically dominant species (Figure 2a,b).
This was particularly true for P. fallax and N. guatemalensis, two species that displayed low probabilities of maintaining numerical dominance on subplots (Figure 2a,b). We also observed that subplots categorized as group 7 (which lacked a numerically dominant species) tended to stay in the group 7 at the following sampling time (Figure 2a,b).

| Autocorrelation
For the abundance of each species, we found a positive spatial autocorrelation among neighboring subplots that were separated by as much as 6.0 m, and that values for Geary's C (according to LMs) were positively associated with distance ( Figure 3)

| Subplot scale dynamics and correlations between ant species abundances
Statistical analyses of the relationships between the abundance of each ant taxon (interspecific interactions) are provided in the supporting information (Tables A2-A6), and the results are summarized in   (Tables A2-A6).

| DISCUSSION
We performed bait and pitfall trapping experiments in a very simplified and homogeneous agroecosystem for studying ant community structure. Doing so minimized the effects of habitat-related factors F I G U R E 2 Probability of transition matrices. These figures display the probability that a numerically dominant species (listed on X-axis) will maintain its numerical dominance on the resource or lose it to the benefit of another dominant or subdominant species (listed on Y-axis) from: (a) sampling time 1 to sampling time 2; (b) sampling time 2 to sampling time 3. In (a) and (b) (Trager, 1991).
When S. geminata did not maintain numerical dominance, it was usually not to the benefit of a particular species. A similar pattern of maintaining numerical dominance was observed for M. ebeninum, which is consistent with Hanson and Gauld (1995), who reported that this species behaves like Solenopis in the field. However, when M. ebeninum did not maintain numerical dominance, it was mostly to the benefit of S. geminata, suggesting that the two species may have similar ecological niches. This is consistent with the similarity that we observed in their diurnal and nocturnal foraging activity and with the daily variation of occurrence at baits. Brachymyrmex patagonicus displayed high probabilities of maintaining numerical dominance at baits, and the decrease in its probability of maintaining numerical dominance over time was lower for this species than for the other species in our study. MacGown, Hill, and Deyrup (2007) reported that Brachymyrmex spp. in general and B. patagonicus in particular have the ability to coexist with a variety of other dominant species. These authors also suggest that B. patagonicus may be protected by potent chemicals. N. guatemalensis displayed the lowest probability of maintaining numerical dominance at baits. This is consistent with LaPolla, Brady, and Shattuck (2011), who described Nylanderia spp. as efficient foragers that rapidly find and recruit to resources but that rarely can defend the resources against other ants that arrive later. P. fallax displayed low probabilities of maintaining numerical dominance at baits. Perfecto and Vandermeer (2011) found that Pheidole subarmata tends to lose dominance at baits against S. geminata. In our study, P. fallax was apparently kept low at baits not only by S. geminata, which is a strong interference competitor (Perfecto & Vandermeer, 2011), but also by all of the other dominant and subdominant ant species. Nylanderia spp. are known to be good exploitation competitors (LaPolla et al., 2011), but our results suggest that, depending on the context and the competitors that they face, Nylanderia ants may also be interference competitors.
However, the patterns we observed may also depend, at least in part, on other factors such as daily variation in foraging activity because of thermal constraints. Thermal constraints have been shown to disrupt hierarchies in ant communities (Bestelmeyer, 2000;Cerdá, Retana, & Cros, 1997). We found that two of the three dominant spe-  Sol, S. geminata; Mon, M. ebeninum; Phe, P. fallax; Nyl, N. guatemalensis; Bra, B. patagonicus interactions, however, were symmetrically positive between B. patagonicus and N. guatemalensis. Thus, competitive exclusion at an ephemeral food source does not occur between these two species of Formicinae, which seem to tolerate each other.
The positive correlation between the local abundance and neighborhood abundance of a considered species may reflect the absence of intraspecific competition between colonies. Indeed, S. geminata is known to be, at least temporarily, polygynous (Trager, 1991). Some Nylanderia spp. may also be polygynous (Arcila, Ulloa-Chacon, & Gomez, 2002), and MacGown et al. (2007)  Indeed, our data support this hypothesis as P. fallax foraging activity is greater during the day while the foraging activity of N. guatemalensis greater at night. This should lead to the temporal sharing of resources.
The positive spatial association could also result from a similarity in foraging behavior. We found the lowest probabilities of maintaining dominance in these two species, both of which are good exploitative competitors that rapidly discover resources (Itzkowitz & Haley, 1983;LaPolla et al., 2011), had a similar probability to losing dominance at baits to the benefit of the other. This should lead to the spatial sharing of resources at such fine spatial scale. We also found a positive correlation between the local abundance of P. fallax and the neighborhood abundance of B. patagonicus, whose local abundance was not correlated with the neighborhood abundance of P. fallax. Because B. patagonicus was highly numerically dominant at baits and had a high probability of maintaining its numerical dominance at baits, it is not surprising that the presence of the P. fallax in the neighborhood did not apparently affect B. patagonicus.
To our knowledge, this is the first study reporting information on ant community in a banana crop agrosystem in Martinique. In summary, we found that three dominant and two subdominant species structure the ant community in a very simplified banana agrosystem.
All these species generally maintain numerical dominance at a "momentary" resource and that interference competition probably prevails in this ground-dwelling community. However, temporal and spatial niche partitioning may also explain, at least partly, the observed pattern. Moreover, the presence of invasive species, such as the fire ant S. geminata, the rover ant B. patagonicus, the little fire ant W. auropunctata, and the crazy ant P. longicornis, might greatly affect ecological processes and ecosystems (Lach & Hooper-Bui, 2010