The role of breeding system on ant ecological dominance: genetic analysis of Ectatomma tuberculatum

Abstract

Social insects exhibit a great variability in their social organization, and this affects colony kin structure, relatedness among nest mates, and population genetic structure. In the mosaic of arboreal ants of neotropical habitats, mutually exclusive dominant ant species occupy different territories, and their nest distribution is spatially aggregated in patches influencing patterns of population genetic structure. In this study, we performed an analysis of the population and colony genetic structure of the facultative polygynous ant Ectatomma tuberculatum to investigate how the particular breeding and social system of this species can explain its ecological dominance in the mosaic. Within-nest genetic analysis revealed that relatedness between nest mate workers was significantly greater than zero (r = 0.30) with an effective number of queens per nest of Ne = 2.5–3, indicating that polygyny is functional in this species. Moreover, we found that queen number was highly variable, probably due to queen adoption events, leading to the prevalence of polygyny over monogyny. Finally, the strong population genetic structure and the significant isolation by distance suggested that both budding and polydomy take place in this species. The respective role of secondary polygyny, budding, and polydomy are then discussed in the context of the mosaic of arboreal ants, and we propose that this particular social organization ensures the ecological dominance of E. tuberculatum by optimizing the colonization of new available nesting sites and by increasing territory size.

Hamilton's (1964) kin selection theory provides a coherent framework to explain the evolution of altruistic behaviors that emphasizes the importance of genetic relatedness between altruistic and recipient individuals. In social insects, the pattern of relatedness between individuals of the colony plays a crucial role for predicting social conflicts and their outcomes (Keller and Chapuisat 2001; Tarpy et al. 2004; Ratnieks et al. 2006). Not surprisingly, with the development of highly variable markers such as microsatellites (Queller et al. 1993; Goldstein and Schlötterer 1999; Ross et al. 1999), a large number of molecular studies have been conducted to estimate this parameter (Bourke and Franks 1995; Crozier and Pamilo 1996; Pamilo et al. 1997; Ross 2001). In ants, the relatedness between nest mate workers appears to be highly variable ranging from full-sister relationships (r = 0.75) to completely unrelated nest mate workers (r = 0) (Ross and Keller 1995; Giraud et al. 2002; Van der Hammen et al. 2002).

The pattern of relatedness within a social group and the population genetic structure are affected by the breeding system and the colony mode of foundation (Bourke and Franks 1995; Ross 2001). The breeding system corresponds to the number of breeders in the social group, their genetic relationships (in association with their dispersal abilities), and the distribution of reproduction among them (Ross 2001). Polygyny (i.e., several queens coexisting in the same colony) and polyandry (i.e., queen mating with several males) are the 2 main reproductive strategies known to decrease the relatedness among nest mate workers (Pamilo 1991). The dispersal abilities of reproductive individuals are closely linked to the mode of colony foundation (Bourke and Franks 1995). Independent colony foundation by one or several queens without the help of workers is generally associated with nuptial flights with long-range dispersal and mating away from the natal nest. This high dispersal ability results in random mating with no genetic structure within a population. In contrast, when the queen founds a new colony with the help of workers (dependent colony foundation), the dispersal ability is restricted to the walking distance of workers that can result in population genetic viscosity with neighboring colonies being genetically more similar than distant ones (Pamilo et al. 1997).

Natural habitat characteristics (e.g., fragmented or continuous) can also affect the pattern of population genetic structure through effects on dispersal of sexuals (Clémencet et al. 2005). Moreover, in seemingly continuous habitats, species distribution can be influenced by ecological factors such as food resources, environmental conditions, and interspecific competition (Leston 1978; Davidson 1997). Typically in Neotropical environments, the characteristic “patchwork” or “mosaic” distribution of the arboreal ants results from such ecological features (Room 1971; Leston 1978; Majer 1993). Thus, in these habitats, mutually exclusive dominant ant species occupy different groups of trees and their nests are spatially aggregated in patches, influencing in turn patterns of population genetic structure.

In the mosaic of arboreal ants of cocoa plantations in Bahia State, Brazil, 3 main species are considered as ecologically dominant: Wasmannia auropunctata, Azteca chartifex spiriti, and Ectatomma tuberculatum (Majer et al. 1994; de Medeiros et al. 1995). Interestingly, different breeding systems and social organizations characterize these species. Wasmannia auropunctata (Myrmicinae) is a highly polygynous and polydomous species, known to be unicolonial with no aggressive behavior occurring between individuals from different colonies. Moreover, new colonies are thought to be founded by budding (Hölldobler and Wilson 1977, 1990; Passera 1994). In contrast, A. chartifex spiriti (Dolichoderinae) is a monogynous and polydomous species highly territorial and aggressive (de Medeiros 1992). Ectatomma tuberculatum (Ectatomminae) is a facultatively polygynous species with half of the nests containing 2–26 queens (Hora, Vilela, et al. 2005). However, nothing is known about the colony mode of foundation or the dispersal strategies of sexuals in this region. Polydomy has been suggested to occur in populations from Mexico (Garcia-Perez et al. 1991), French Guiana (Richard FJ, personal communication) and Brazil (Hora, Vilela, et al. 2005). However, polydomy remains less clear in E. tuberculatum than in A. chartifex spiriti, for example, because of low level of intraspecific aggression in Brazilian population of E. tuberculatum (Fénéron et al. 1999).

The aim of our work was to further investigate the breeding system and population genetic structure of the ecological dominant ant E. tuberculatum, which seems to show a particularly complex social organization. We thus perform a genetic analysis using microsatellite markers to answer 3 specific questions. 1) what is the level of relatedness among nest mates in this facultative polygynous ant species? 2) what are the dispersal abilities and the mode of colony foundation? and 3) how the breeding system and the social organization of this ant can explain its ecological dominance?

METHODS

Sampling and genetic analysis

To investigate the aggregated nest distribution of E. tuberculatum, 3 hierarchical levels were distinguished: the nest, the patch, and the locality levels. The nest level refers to the nest itself, built at the basis of a tree, along the main root, and containing one or several queens (Hora, Vilela, et al. 2005). A visible chimney of 10–30 cm long constructed at the nest entrance (Delabie 1990) allows the location, marking, and mapping of nests in the field. Nest distribution is heterogeneous and forms areas of tens to hundreds of square meters with clusters of nests (Delabie 1989, 1990) that we identified as patches. Nests within a same patch were typically distant from a few meters from one another, whatever the number of nests per patch, which appeared to be highly variable (ranging from 16 to 306 with mean ± standard error [SE] = 57.8 ± 69.7). Between 2 patches not a single nest of E. tuberculatum can be found, and the minimum distance between nests used in this study to consider that they belonged to 2 different patches was of 100 m. The locality level corresponds to a cluster of patches grouped in a few square kilometers area. In this study, 3 localities in Bahia, Brazil, were studied: Itabuna, Uruçuca, and Buerarema, distant from 17 to 34 km (Figure 1).

Figure 1

Map of the 3 localities (Itabuna, Uruçuca, and Buerarema) and the 15 patches (1–15) studied in Bahia, Brazil. The 3 surrounded patches in Itabuna locality (patch 1–3) are those studied for within-nest genetic analysis.

The genetic structure at a large spatial scale was investigated by sampling workers from Itabuna, Uruçuca, and Buerarema localities, in 5 patches per locality and 10 nests per patch (n = 150) (Figure 1). A single worker per nest was analyzed to avoid use of nonindependent genotypes caused by sampling genetically related workers. Sampled patches within a locality were distant from a mean of 1.6 ± 0.2 km (range 0.1–3.8 km for the most distant ones, see Figure 1).

To study genetic structure at a fine spatial scale and to reliably estimate the number of queens per nest, 20 workers per nest were sampled in 3 patches of Itabuna locality (Patch 1–3 surrounded in Figure 1) (n = 1080). Within each patch, nests were sampled along a transect. These 3 transects were 131 m, 74 m, and 49 m long (transect length was depending on patch size) and consisted of 29, 13, and 12 nests, respectively, with an average distance between nests of 4.8 m (±0.7).

DNA was extracted from the head and thorax of workers using a standard 10% Chelex protocol. Polymerase chain reactions (PCR) were performed as described in Hora, Doums, et al. (2005) with 6 fluorescent-labeled microsatellite primers: L17, L84, L90, L92, L134, and L164 (Poteaux et al. 2003). We assessed PCR products length with an ABI 310 automated sequencer and scored allele sizes with the GENESCAN corresponding software.

Population genetic analysis

For the large-scale sampling, genetic diversity was estimated for each locality by the number of alleles and the expected heterozygosity using the web implementation (version 3.1c) of the program GENEPOP (Raymond and Rousset 1995). The inbreeding coefficients (Fis) were estimated using the locality or the patch as the reference population with the F statistics (SmallF) of Weir and Cockerham (1984) using FSTAT 2.9.3.2 (Goudet 2001). The SEs were obtained by jackknifing over loci. Exact tests for Hardy–Weinberg proportions were tested by using 15 000 randomizations of alleles within samples with the same software. Genetic differentiation (Fst) was estimated at 2 hierarchical levels: between patches within each locality (i.e., patches considered as populations) and between localities (i.e., localities considered as populations), and the SEs were obtained by jackknifing over loci with FSTAT 2.9.3.2. Genetic differentiations were tested using the GENEPOP program, and a Bonferroni correction was performed over the obtained P values. To evaluate the amount of genetic variance at the different hierarchical levels (among localities and among patches within locality), we performed an analysis of molecular variance (AMOVA) using the program ARLEQUIN (Schneider et al. 2000). Tests for differences among localities and patches for allelic richness, observed heterozygosity, gene diversity, inbreeding coefficient, and genetic differentiation were performed with patches considered as populations, using comparison among groups of samples option of FSTAT 2.9.3.2 software, 2-sided P values being obtained after 10 000 permutations. A pattern of isolation by distance (IBD) between patches over all localities was tested using a Mantel test with 10 000 permutations performed with the GENEPOP web implementation program of Raymond and Rousset (1995). Spearman rank correlation coefficients were calculated between the transformed matrix of pairwise Fst (i.e., Fst/(1 − Fst)) and the matrix of ln-transformed geographical distances.

For the fine-scale sampling, genetic differentiation (Fst) was estimated as described above, F statistics being estimated between nests (i.e., nests considered as populations), and genetic differentiations were tested between nests with the GENEPOP program. A second AMOVA was performed with this genetic data set to evaluate the amount of genetic variance at a fine spatial scale among patches and among nests using the program ARLEQUIN. A pattern of IBD between nests within each transect was tested by plotting the modified Fst (i.e., Fst/(1-Fst)) against the ln-transformed geographical distance and using a Mantel test with 10 000 permutations as described above to estimate the level of significance of the obtained Spearman rank correlation coefficients. Moreover, to analyze the spatial genetic structure within patch, spatial autocorrelation analysis was performed using the program SPAGEDI 1.2 (Hardy and Vekemans 2002). Moran's I statistics were calculated for diploid multilocus worker genotypes for class of geographical distances of 25 m. P values on Moran's I statistics were obtained by performing 10 000 permutations of spatial locations of individuals within each class.

Within-nest genetic analysis

The inbreeding coefficient (Fit) of nest mate workers were estimated with the F statistics (CapF) using FSTAT 2.9.3.2 (Goudet 2001) and the genetic data of the 3 transects (i.e., 20 workers per nest) using the transect as the reference population. SEs were obtained by jackknifing over loci. The significance of the inbreeding estimates was evaluated with exact tests of Hardy–Weinberg proportions over all nests at the transect level using 15 000 randomizations (Goudet et al. 1996). Regression relatedness coefficients among nest mate workers were calculated using the same genetic data set and the 3 transects being labeled as different deme with the program RELATEDNESS 4.2 (Queller and Goodnight 1989). SEs of r were obtained by jackknifing over colony, and colonies were weighted equally. Student's t-tests were performed to compare inbreeding coefficients to the reference value of zero and relatedness estimates to the theoretical value of r = 0.75 (Sokal and Rohlf 1995).

We looked for the presence of multiple queens per nest using the program MATESOFT 1.0 (Moilanen et al. 2004), which performed parentage analysis based on worker genotypes found in the 54 nests sampled in the 3 patches. Allele frequencies for computations in MATESOFT were calculated for each patch using the program RELATEDNESS 4.2 (Queller and Goodnight 1989).

The effective number of reproductive queens (Ne) was estimated from the relatedness among nest mate workers (r_w–w) with the formula Ne₍₁₎ = 3/(4r_w–w) assuming 1) unrelated equally fecund queens, 2) no inbreeding, and 3) monoandry. Alternatively, assuming equal relatedness among queens and workers (i.e., queens stay in their natal nests and mate with unrelated males), the effective number of related queens was estimated as Ne₍₂₎ = (3 − r_w–w)/(3r_w–w) (e.g., Pamilo 1991). We compared our results with the data already known about the number of queens per nests in E. tuberculatum (Hora, Vilela, et al. 2005). Considering both the monogynous (n = 40) and polygynous colonies (n = 48) collected in the same site of Itabuna, Hora, Vilela, et al. (2005) found a harmonic mean number of queens per nest equal to Nh = 1.5 (±3.3).

RESULTS

Population genetic analysis

Considering the entire sample, the average number of alleles per locus was 6.8, ranging from 3 to 12 alleles with a mean gene diversity of 0.56 (0.23–0.79) (Table 1).

Table 1

Genetic diversity of microsatellite markers in Ectatomma tuberculatum

	Itabuna				Uruçuca				Buerarema				All
	Na	He	Fis		Na	He	Fis		Na	He	Fis		Na	He	Fis
L134	3	0.23	0.06	ns	3	0.51	−0.10	ns	2	0.10	−0.04	ns	3	0.28	−0.07
L90	6	0.64	−0.09	ns	6	0.59	−0.02	ns	8	0.69	−0.07	ns	8	0.64	−0.05
L84	11	0.77	−0.20	ns	10	0.79	−0.09	ns	11	0.82	−0.15	ns	12	0.79	−0.17
L17	6	0.78	−0.18	ns	8	0.79	−0.16	ns	6	0.79	−0.19	ns	9	0.79	−0.18
L92	4	0.58	−0.48	ns	4	0.71	−0.30	ns	5	0.59	−0.50	ns	5	0.63	−0.43
L162	3	0.34	0.07	ns	2	0.10	−0.05	ns	4	0.30	−0.01	ns	4	0.24	0.021
All	5.50	0.56	−0.18	ns	5.50	0.58	−0.14	ns	6	0.55	−0.19	ns	6.83	0.56	−0.18

Na, number of allele; He, expected heterozygosity; Fis, inbreeding coefficient. Fis level of significance: ns means not significant.

The level of gene diversity was very similar for the 3 localities studied (Table 1). The 3 localities did not differ by their allelic richness (range 5.5–6, P = 0.611), observed heterozygosity (P = 0.710), gene diversity (0.55–0.58, P = 0.236), or inbreeding coefficient (Fis) (P = 0.637). Even if inbreeding coefficient values for each locality (Fis) were found to be negative (range −0.18 to −0.14, Table 1), they did not differ significantly from zero (Itabuna: t = 0.94, P = 0.45; Uruçuca: t = 1.25, P = 0.34; Buerarema: t = 1.18, P = 0.36; and all localities: t = 3.00, P = 0.10), and Hardy–Weinberg proportions were found to be respected within each locality. Moreover, at the patch level, Fis were either positive or negative depending on the locus and the patch but similarly no patch at any locus departed from Hardy–Weinberg's equilibrium (Supplementary data). This indicates that random mating seems to occur both at the locality and the patch level.

At the large spatial scale, the estimates of genetic differentiation between localities and between patches within a locality were all highly significant although they showed low Fst values (ranging from 0.029 to 0.073, Table 2). Furthermore, the 3 localities did not differ significantly by their level of genetic differentiation Fst (P = 0.399). The AMOVA showed that 97.17% of genetic variance was due to genetic differentiation within patch and that most of the remaining genetic variance was observed between patches rather than between localities (Table 3). When examining pairwise genetic differentiation between patches, 51 out of the 105 pairs of patches showed a significant genetic differentiation after a Bonferroni correction. Moreover, a significant positive correlation was found between pairwise genetic differentiation of patches and their geographical distances (Rs = 0.012, P < 10⁻⁵), indicating that IBD occurs at a large spatial scale.

Table 2

Genetic differentiation between localities and between patches

Population level	Genetic differentiation	Fst ± SE	P
Locality	Between localities	0.038 ± 0.017	<0.0001
Patch	Between Itabuna patches	0.054 ± 0.024	<0.0001
	Between Uruçuca patches	0.029 ± 0.017	0.001
	Between Buerarema patches	0.073 ± 0.035	<0.0001

P values correspond to the exact test of genetic differentiation.

Table 3

Large-scale sampling AMOVA

Level of variation	df	% Variance	Fixation index	P
Among localities	2	0.05	F = 0.0005	0.3341
Among patches within locality	12	2.78	F = 0.0278	<0.001
Within patches	285	97.17

AMOVA based on microsatellite genotypes of 150 individuals with a genetic structure corresponding to 15 patches distributed among 3 localities. df, degrees of freedom.

At the fine spatial scale, genetic differentiation between nests within each patch was strong with a mean Fst (±SE) = 0.128 ± 0.02. The 3 focal patches did not differ significantly by their level of genetic differentiation (P = 0.493). Genetic differentiation between nests was equal to Fst = 0.147 ± 0.02 in patch 1; Fst = 0.137 ± 0.02 in patch 2; and Fst = 0.100 ± 0.01 in patch 3, all the exact tests for genetic differentiation being highly significant (P < 0.0001). The AMOVA revealed that genetic variance was due to between nests and between patches differences, both variances being significantly different from zero (Table 4). Nevertheless, a larger part of the genetic variance was explained by differentiation among nests. A significant positive correlation between pairwise genetic differentiation of nests and their geographical distances was found in 2 patches (patch 1, Rs = 0.048, P < 0.0001 and patch 3, Rs = 0.022, P = 0.004), whereas a similar pattern of IBD was marginally significant in patch 2 (Rs = 0.021, P = 0.075). Similarly, along each transect, Moran's indexes that reflect genetic similarities among individuals were found to decrease linearly with geographical distance, confirming that neighboring nests are more genetically related than more distant ones. Spatial autocorrelation analysis revealed a similar shape of spatial structuring in the 3 transects even if they were of different lengths (Figure 2), indicating that patterns of genetic viscosity were comparable among patches.

Figure 2

Patterns of spatial autocorrelation observed in each transect. Levels of significance are indicated by * if P < 0.05, ** if P < 0.01, *** if P < 0.001, and ns if they were not significant.

Figure 3

Distribution of nest mate worker relatedness estimates in Ectatomma tuberculatum.

Table 4

Fine-scale sampling AMOVA

Level of variation	df	% Variance	Fixation index	P
Among patches	2	1.46	F = 0.0146	<0.0001
Among nests within patch	51	6.86	F = 0.0696	<0.0001
Within nests	2106	91.68

AMOVA based on microsatellite genotypes of 1080 individuals with a genetic structure corresponding to 54 nests divided in 3 patches. df, degrees of freedom.

Within-nest genetic analysis

For the 3 focal patches, the inbreeding coefficients of workers (Fit) did not differ significantly from zero (Table 5). Moreover, the exact tests of randomizations for testing Hardy–Weinberg proportions were not significantly different from zero (Table 5).

Table 5

Population and nest structure of Ectatomma tuberculatum

Transect	n	Fit ± SE	HW P	rw–w ± SE	P	Ne(1)	Ne(2)
Patch 1	29	−0.05 ± 0.02	ns	0.31 ± 0.04	0.009	2.42	2.90
Patch 2	13	−0.14 ± 0.04	ns	0.32 ± 0.06	0.021	2.34	2.79
Patch 3	12	−0.18 ± 0.04	ns	0.25 ± 0.07	0.018	3.03	3.70
All	54	−0.10 ± 0.02	ns	0.30 ± 0.03	0.005	2.50	2.99

Fit, inbreeding coefficient with the patch as the reference population; n, number of nests; HW P, P values corresponding to the exact tests for Hardy–Weinberg proportions (ns, not significant); r_w–w, genetic relatedness among nest mate workers; P, P values corresponding to 2-tailed t-test comparing r_w–w with 0.75; Ne, effective number of reproducing queens per nest assuming unrelated queens Ne₍₁₎ or queens as related as workers Ne₍₂₎.

The average within-nest genetic relatedness among nest mate workers was significantly greater than zero with r_w–w (±SE) = 0.30 ± 0.03 (t = 9.53, P = 0.011) and lower than 0.75 (t = 14.26, P = 0.005). Genetic relatedness among nest mate workers did not differ significantly between patches (t_{patch 1/patch 2} = 0.10, P = 0.93; t_{patch 1/patch 3} = 0.55, P = 0.64; t_{patch 2/patch 3} = 0.54, P = 0.64) (Table 5). The breeding system varied greatly among nests because within-nest genetic relatedness varied from 0 to 0.74 (Figure 3). The distribution of nest mate worker relatedness estimates was not bimodal in any studied patch but rather continuous with intermediate values of relatedness (Figure 3). Most of the nests likely had a polygynous social organization with low intranest relatedness. Matesoft analysis of parentage gave similar results with 35 nests (64.8%) detected as polygynous and only 5 nests (9.3%) corresponding to full-sib genotypes that is monogynous and monoandrous nests. The mean relatedness within nest r_w–w = 0.30 ± 0.03 corresponds to an effective queen number Ne₍₁₎ = 2.5 (95% CI: 2.1–3.1), if queens are not related, and Ne ₍₂₎ = 3.0 (95% CI: 2.5–3.8), if queens are assumed to be as related as workers (Table 5). These values were indeed consistent with the harmonic mean number of queens per nest Nh = 1.5 (±3.3) found by Hora, Vilela, et al. (2005) by collecting 88 colonies in this same locality of Itabuna between 1996 and 2001.

DISCUSSION

The genetic analysis of nest mate workers of the ant E. tuberculatum established that polygyny is functional in this species, several mated queens being involved in the reproduction within a colony, as previously shown by behavioral data (Hora, Vilela, et al. 2005). The relatedness between nest mate workers was indeed found to be significantly higher than zero and lower than 0.75 with an average of 0.30. Moreover, the effective number of reproductive queens estimated from this relatedness value (2.5 unrelated queens or 3 related queens per nest) was not significantly different from the harmonic mean number of queens collected per nest (Nh = 1.5 ± 3.3; Hora, Vilela, et al. 2005), indicating that polygyny alone can explain within-nest genetic relatedness. Thus, neither polyandry (i.e., several mates per queen) nor reproductive skew among queens is likely to occur in this species. The absence of polyandry is not surprising because it is rather rare in social insects and particularly in polygynous species (Keller and Reeve 1994; Crozier and Fjerdingstad 2001; Strassmann 2001). However, reproductive skew among queens is often found in polygynous colonies (Bourke and Franks 1995), and our results reveal that all queens of E. tuberculatum actually contribute equally to the production of workers. This is in agreement with behavioral observations that showed that all queens were egg layers and that no apparent dominance hierarchy or agonistic behavior took place in these Brazilian polygynous colonies (Hora, Vilela, et al. 2005).

The large variation of relatedness values showed that the social organization varied greatly among nests in queen number and/or in the degree of relatedness among them. The continuous distribution of nest mate worker relatedness within patch suggested that monogyny and polygyny are not 2 distinct alternative strategies. Moreover, because only 5% of the studied nests were found to be monogynous, polygyny clearly prevails in E. tuberculatum, and it likely results from queen adoptions (Hora, Vilela, et al. 2005) and queen turnover (i.e., queen loss and/or queen replacement) as in most polygynous species (Keller 1995). Monogyny therefore corresponds to a transitory state of the colony, which is likely to be succeeded by polygyny.

Population genetic analysis of E. tuberculatum revealed strong genetic structuring both at fine and large spatial scales. Limited dispersal of sexuals (female and/or male) and polydomy (i.e., several nests constituting a single colony unit) can both explain the pattern of IBD found within a given patch. Genetic IBD from limited dispersal of females generally corresponds to dependent colony foundation (i.e., queens found new colonies with the help of workers in the proximity of their natal nests) (Hölldobler and Wilson 1977; Pamilo 1991; Rosengren et al. 1993; Bourke and Franks 1995; Keller 1995). In contrast, limited dispersal of flying queens followed by independent colony foundation is unlikely to give rise to such genetic viscosity. Dependent colony foundation thus seems to occur in E. tuberculatum, and the absence of inbreeding suggests that male dispersal is not as restricted as female dispersal. Differential dispersal strategies between males and females may exist in E. tuberculatum, but mitochondrial genetic analysis would be required to confirm this hypothesis. In this case, mating in E. tuberculatum could therefore take place as in other polygynous species (Bourke and Franks 1995), near the female natal nest, between nonrelatives.

In addition to the low female dispersal, in polydomous species, nests can stay interconnected after budding through worker exchanges (Pamilo and Rosengren 1984; Boomsma et al. 1990; Chapuisat et al. 1997; Pedersen and Boomsma 1999). Given that worker exchanges between nests are likely to decrease with geographical distance, polydomy also results in a pattern of genetic viscosity with microgeographic genetic structure. Pattern of IBD has been detected in many polygynous and often also polydomous species (Pamilo 1983; Crozier et al. 1984; Pamilo and Rosengren 1984; Seppä and Pamilo 1995; Chapuisat et al. 1997; Tay et al. 1997; Pedersen and Boomsma 1999). In E. tuberculatum, polydomy has already been suggested to occur (e.g., Garcia-Perez et al. 1991; Hora, Vilela, et al. 2005) and our population genetic analysis tends to confirm that. The similar pattern of spatial autocorrelation found in the 3 patches for the same classes of geographical distances suggested that worker exchanges between neighboring nests, occurring over fixed distances of the order of colony boundaries, likely play an important role in the observed genetic viscosity. Moreover, around a given distance (20–30 m, Figure 2), the pattern of IBD does not seem to hold any more suggesting that nests could not belong to the same colony over this distance. The absence of aggression between workers coming from nests distant from 3 m (Zinck et al. submitted) also strongly supports the occurrence of polydomy in this species.

Secondary polygyny, budding, and polydomy are biological traits often linked to each other in ants. Indeed, polygyny resulting from queen adoptions typically leads to larger colony size (e.g., Janzen 1973; Satoh 1989; Hora, Vilela, et al. 2005), and it also favors reproduction by budding (Keller 1991). Moreover, budding can easily produce multinest organization if buds stay connected through worker exchanges (Rosengren, et al. 1993). Considering our genetic analysis and previous behavioral study (Hora, Vilela, et al. 2005), we assume first that larger colony size is achieved in E. tuberculatum by polygyny and polydomy; second that variable queen number results from queen adoptions and budding.

In the mosaic of arboreal ants, given that 90–97% of the trees are occupied by dominant ants (Majer 1993), living space is one of the resources in the shortest supply (Leston 1973; Davidson 1997). In these habitats where available space without dominant ants is rare, competition can take place at 2 stages: either between adjacent mature colonies or between young queens and recently established colonies (Majer 1976, 1993). As the success rate to found new colonies is generally higher by budding than by independent colony foundation (Rosengren 1993; Peeters and Ito 2001), and especially in the mosaic of arboreal ants (Majer 1976, 1993), dependent colony foundation appears to be particularly advantageous in these habitats. Colonization of new nesting site therefore corresponds to lateral spread of existing colonies, either in space between colonies or after the death of neighboring colonies (Majer 1976). Polygyny also plays an important role in the colonization process because larger sizes of multiple-queen colonies in comparison with monogynous colonies (Hora, Vilela, et al. 2005) lead to increased colony productivity (Herbers 1984; Walin et al. 2001). Moreover, polygyny and polydomy, in increasing the ability to monopolize resources (McIver 1991; Human and Gordon 1996; Davidson 1997; Holway and Case 2000) and the protection of the colony against predators and competitors (Nonacs 1988; Herbers 1993; McGlynn et al. 2004; Denis et al. 2006), ensure colony longevity and as a consequence, E. tuberculatum occurrence.

Three main species are characterized as ecologically dominant in the mosaic of cocoa agrosystems of Bahia, Brazil: E. tuberculatum, W. auropunctata, and A. chartifex spiriti (de Medeiros 1995). Ecological dominance is considered to result from both numerical and behavioral dominance (i.e., dominance in interspecific competition due to superior fighting and/or recruitment abilities) (Adams 1994; Davidson 1998). In W. auropunctata and A. chartifex spiriti, dominance seems to be achieved through unicoloniality and opportunistic strategy for W. auropunctata (Wetterer and Porter 2003) and high levels of aggression and territoriality for A. chartifex spiriti (de Medeiros 1992; Adams 1994). In E. tuberculatum, reduced queen fecundity (Hora, Vilela, et al. 2005) is likely compensated by polygyny. The flexibility of queen number may also be a key factor responsible of the ecological dominance of this ant. The regulation of queen number and colony size by queen adoptions and worker exchanges could indeed allow budding to occur at any time. Such plasticity could thus assure E. tuberculatum to take advantage of any nesting site opportunities, particularly at the edge of the patches. Colony size is indeed determinant in between-colonies competition at territory edges (Adams 1990), and environmental factors are known to affect reproductive allocation among polydomous nests (Sundström 1995; Walin et al. 2001). Moreover, variation in queen number can be due to queen transport between nests as observed in other polygynous polydomous species (e.g., Wilson 1971; Mabelis 1979). In addition, E. tuberculatum workers could play a direct role in queen number regulation because they are directly involved in queen adoption and actively bring back new queens for adoption inside the nest (Hora, Vilela, et al. 2005). They could therefore choose on the basis of the information they get about the different polydomous nests, the one requiring a new queen most.

To conclude, dominant species in the mosaic of arboreal ants share common properties such as being broad-spectrum predators, tending Homoptera and collecting honeydews and extrafloral nectaries (Leston 1973, 1978). However, different breeding systems and social organizations can be found in association with these general traits and result in ecological dominance. In E. tuberculatum, secondary polygyny with flexible queen number and budding result in a social organization within a given patch that ensures to maintain its ecological dominance in the mosaic of cocoa agrosystems. Further investigations about the relationship between nests might be interesting to do to determine in which extent a patch of E. tuberculatum could correspond to a supercolony. Indeed, if no aggression occurs between nonnest mate workers within a given patch, the definition of unicoloniality would be fulfilled (Hölldobler and Wilson 1990) and unicoloniality could contribute to E. tuberculatum ecological dominance as in W. auropunctata (Ulloa Chacón and Cherix 1990).

SUPPLEMENTARY MATERIAL

Supplementary data can be found at http://www.beheco.oxfordjournals.org/.

We thank N. Châline, J. Clémencet, C. Tirrard, and T. Monnin for their helpful comments on the manuscript, and J. H. C. Delabie and the CEPLAC staff for their great help during our field trips. This work was supported by a French research grant “Action Concertée Incitative jeunes chercheurs 2001” ACI n°5183 and the “Bureau des Relations Internationales de l'Université Paris 13”. R.R.H. received financial support from CNPq, Brazil (3098552003-9). Research was permitted by the Brazilian Minister of Science and Technology (licence n°0107/2004).