Context-dependent benefits from ant–plant mutualism in three sympatric varieties of Chamaecrista desvauxii


  • Beatriz Baker-Méio,

    Corresponding author
    1. Department of Biology, University of Missouri – St. Louis, One University Boulevard, St. Louis, MO 63121-4499, USA
    2. Instituto de Biologia, Universidade Federal de Uberlândia, C.P. 593, Uberlândia MG 38400-902, Brazil
    Search for more papers by this author
  • Robert J. Marquis

    1. Department of Biology, University of Missouri – St. Louis, One University Boulevard, St. Louis, MO 63121-4499, USA
    Search for more papers by this author

Correspondence author. E-mail:


1. Mutualistic interactions are characterized by conditional outcomes that depend on both the biotic and the abiotic context. However, limited information is available on the factors that affect the strength of ant–plant interactions among sympatric congeneric species.

2. We compared the benefits gained from attracting ants via extrafloral nectaries – i.e. lowered herbivory and increased seed set – of three co-occurring varieties in the Chamaecrista desvauxii complex (Leguminosae) in a cerrado area in Uberlândia, Brazil. Using whole-individual exclusion experiments, we tested the hypotheses (i) that the relative strength of those benefits is higher in the variety with the largest extrafloral nectaries and (ii) that those benefits are conditional on the presence of predispersal seed predators.

3. Extrafloral nectaries are larger, produce more nectar and attract more ants in var. brevipes than in the other two varieties included in the study. Var. modesta has intermediate-sized nectaries, while a third, undescribed variety has small nectaries, and both attract relatively few ants.

4. For var. brevipes, extrafloral nectary (EFN) removal significantly increased folivory and attack on fruits by sucking insects, decreasing the relative number of flowers, fruits and seeds produced per individual. For the other two varieties, in contrast, ant effects were reduced, and ants did not significantly improve reproductive success. In addition, effects of EFN removal were less pronounced or absent when seed predators were excluded from fruits of var. brevipes.

5.Synthesis. We showed experimentally that benefits from interactions of three co-occurring varieties of Chamaecrista desvauxii with ants are context-dependent both within and among taxa. Variation in the strength of mutualisms among sympatric taxa may potentially reinforce ecological reproductive isolation and contribute to diversification in this group.


Numerous plant traits such as tissue toughness, trichomes and secondary compounds directly reduce the impact of herbivores (Coley 1983; Harborne 1993; Fernandes 1994; Diniz et al. 1999; Peeters 2002). Indirect defences, on the other hand, are based on the attraction, nourishment or housing of herbivore enemies (Heil 2008). Examples include protection provided by ants attracted to plants by extrafloral nectar, domatia and/or food bodies (Bentley 1977). Ants can increase plant fitness by preying on herbivores, chasing them away or simply by disrupting their behaviour (Schemske 1980; Costa, Oliveira-Filho & Oliveira 1992; Del-Claro, Berto & Réu 1996; Freitas & Oliveira 1996; de la Fuente & Marquis 1999), ultimately increasing the number of flowers, fruits and seeds produced by an individual (Horvitz & Schemske 1984; Del-Claro & Oliveira 1996; Rudgers 2004; Nascimento & Del-Claro 2010).

Within a plant species, benefits from mutualistic interactions with ants often depend on the biotic and abiotic contexts: Benefits can depend on the abundance of ants and herbivores (Rudgers & Strauss 2004), the identity of the ants attracted to the plant (Miller 2007; Palmer & Brody 2007) and resource levels (de la Fuente & Marquis 1999; Kersch & Fonseca 2005). This is especially true for facultative mutualisms, that is, when ants are attracted by extrafloral nectar or food bodies but do not nest on the plant (Heil 2008; Bronstein 2009); these associations are looser, and absent mutualists can be easily replaced by alternative species (Bronstein 1994). In addition, benefits from ant–plant mutualisms are likely to be a function of the abundance of the partners (Barton 1986), in contrast to other mutualistic interactions where one individual can fulfil the reward or service required (Bronstein 1994). Nevertheless, ant effects are consistently positive or at most neutral, regardless of context (Rico-Gray & Oliveira 2007; Chamberlain & Holland 2009), and only rarely negative (Palmer et al. 2008).

When studied, congeners have been shown to vary in benefits received from ant protectors, although in some cases no differences between plant species were detected (Koptur 1979, 1984). The level of protection is often related to the rate of ant attendance. For instance, in two sympatric Macaranga species, ants only reduced leaf damage and number of herbivores in the species with the highest level of ant activity (Whalen & Mackay 1988). Likewise, ant exclusion only decreased herbivore mortality in Passiflora individuals with naturally high levels of ant activity, regardless of species (Smiley 1986). However, while most P. vitifolia individuals frequently attracted ants, individuals of the sympatric P. quadrangularis had low ant activity (Smiley 1986). In these and other systems, differences in benefits from ant association may be related to interspecific differences in the production of extrafloral nectar. For example, among three co-occurring Passiflora species in Costa Rica, both ant activity and discovery of termite baits were lower in the species with the smallest extrafloral nectaries (Apple & Feener 2001).

To our knowledge, no studies have addressed how benefits to reproductive success derived from ant attraction vary among taxa that have facultative interactions with ants. As noted earlier, we might expect such variation because co-occurring, closely related species usually attract different assemblages of ant species (Schemske 1982; Koptur 1984) and vary in the level of extrafloral nectar expression (Koptur 1984; Blüthgen, Gottsberger & Fiedler 2004). Both the presence of extrafloral nectaries and the rate of extrafloral nectar production can be induced by leaf damage (e.g. Agrawal & Rutter 1998; Mondor & Addicot 2003; Wäckers & Bonifay 2004), suggesting that there may be a cost of producing nectar. Given that extrafloral nectar production can be a heritable trait (Rutter & Rausher 2004), these costs could select for permanent loss or decrease in nectar production in species for which herbivory rates are low and/or ant defence is ineffective. In this study, we address the question of context dependency in the outcome of ant–plant mutualisms by quantifying the relative strength of ants as a defensive mechanism in three co-occurring varieties of Chamaecrista desvauxii (Coll.) Killip (Leguminosae). Although all varieties have extrafloral nectaries, nectary sizes are consistently different among varieties, and different levels of ant activity are evident in the field. We examined both inter and intrataxon context dependency. First, we predicted that benefits from ant attraction would depend on the variety (intertaxon context), being highest in the variety with the largest nectaries and lower or negligible in the other two varieties. Second, we predicted that the outcome of the ant–plant mutualism within each variety would be more substantial in the presence of predispersal seed predators (intrataxon context), because seed predators have more direct impacts on plant reproduction than leaf herbivores.

Materials and methods

Study site

The work was conducted at the 130-ha cerrado reserve of Clube Caça & Pesca Itororó in Uberlândia (CCPIU), Brazil (18º55′ S, 48º17′ W). This reserve includes areas of seasonally flooded grasslands (vereda), open savannah areas with scattered trees (campo cerrado) and areas dominated by trees and shrubs (cerrado sensu stricto; see Oliveira-Filho & Ratter 2002). Temperatures are constant throughout the year, but rainfall is concentrated in the wet season, between October and April.

Plant species and varieties

Section Xerocalyx of genus Chamaecrista is a neotropical taxon that includes three species and several varieties distributed from Argentina to Mexico, except for the West Indies. Although the section is monophyletic, phylogenetic relationships among species and varieties within this section are not resolved (Conceição et al. 2009), and the number of recognized species has fluctuated from 16 to three, depending on the treatment (Irwin 1964; Irwin & Barneby 1982). The focal species of this paper, C. desvauxii, is currently subdivided into 17 varieties. All varieties have relatively macrophyllous leaves, each with four leaflets and one extrafloral nectary (EFN) on the petiole (Irwin & Barneby 1982). In some varieties, an additional, smaller EFN may occur on the rachis, between the two pairs of leaflets (BB, pers. obs.). The flowers have no nectar but produce large amounts of pollen, which is harvested by bees through vibrational movements (Gottsberger & Silberbauer-Gottsberger 1988; Costa et al. 2007). The pods open explosively to disperse the seeds. More than one variety may be found in the same area, often side by side; co-occurring varieties may have reproductive isolating mechanisms that prevent hybridization (Costa et al. 2007; B. Baker-Méio and R.J. Marquis unpubl. data).

Three sympatric varieties of the C. desvauxii complex were included in this study. Although a given variety may be more common in some areas of the study site, there were no clear differences in the distribution of the varieties, and they often occurred synoptically. Chamaecrista desvauxii var. brevipes (Benth.) Irwin & Barneby (Fig. 1a) is a subshrub found in savannah areas from Honduras to Central Brazil and Paraguay. Its flowers are large and supported by relatively short pedicels. Although pilosity is variable throughout the range of this variety, in CCPIU the relatively broad leaflets are covered with short, soft trichomes. Chamaecrista desvauxii var. modesta Irwin & Barneby (Fig. 1b) is restricted to Central Brazil and is characterized by a more erect habit, the narrow shape of its leaflets and the reduced size of its flowers. In CCPIU, the flowers have extremely short pedicels. The third variety, C. desvauxii var. 1 (Fig. 1c), has not been described and has only been found in one additional cerrado reserve 10 km from CCPIU (Fazenda Água Limpa, 19°5′ S, 48°21′ W), despite the fact that it is locally abundant. In addition, no specimens resembling this variety were found in herbarium material from GH, MO, NY, UB or US [herbarium codes as listed in Thiers (continuously, updated)]. Therefore, this variety appears to be an extremely localized taxon; it has a prostrate habit, glabrous leaflets that resemble C. desvauxii var. brevipes and small flowers supported by a long pedicel. The three varieties are primarily annuals that germinate at the beginning of the wet season (November–December) and finish fruiting early in the dry season (May), but some individuals are able to survive the dry season and flower again during the following wet season, and scattered individuals of var. 1 can be found flowering throughout the year.

Figure 1.

 Leaves and extrafloral nectaries (inserts) of the varieties included in the study. (a) Chamaecrista desvauxii var. brevipes; (b) C. desvauxii var. modesta; (c) C. desvauxii var. 1. Circles highlight extrafloral nectaries. Scale bars: 5 mm (leaf pictures) and 0.5 mm (inserts).

Extrafloral nectar production

In March 2007, five leaves were collected from each of six individuals per variety. Length and width of the EFN on each leaf were measured using a dissecting scope. Because the shape of the top of the EFN in all varieties is either elliptical or round, the formula [area = (length × width × π)/4] was used to estimate its area. Measurements were averaged per individual, and differences among varieties were tested using an anova.

Extrafloral nectar production was measured in February 2005 on seven individuals of C. desvauxii var. brevipes, seven of var. modesta and five of var. 1. Three hours prior to measurement, one branch of each individual was bagged to prevent visitation to the EFNs and evaporation of the nectar. After this period, the bag was removed and the number of leaves with visible nectar secretion was recorded, along with the total number of bagged leaves. The amount of nectar secreted was measured by touching a piece of chromatography paper to each nectary. The area covered by the nectar is proportional to the volume produced, using the formula of Baker (1979). This method adequately measures low volumes of nectar. Measurements were taken simultaneously for the three varieties in early afternoon (from 12 to 16 h) and at the beginning of the night (from 19 to 23 h).

Extrafloral nectary removal and seed predator exclusion

A factorial design was used to test the role of EFNs as one factor and the impact of predispersal seed predators as the other on the reproductive success of each variety. Each treatment was randomly assigned in January 2006 to half of the individuals in each variety; more individuals were added to the experiment in March 2006, totalling 24 and four individuals of C. desvauxii var. brevipes marked in January and March, respectively, 24 and 12 of var. modesta and 20 and 12 of var. 1. The disparity in the number of treated individuals was a consequence of the differences in size and abundance of each taxon. Marked individuals were scattered throughout the field site. Each variety had a different distribution within the field site, which prevented us from marking individuals of all varieties in the same area. Individuals that were syntropic were included whenever possible. Owing to differences in distribution, relative size, phenology and sample size, responses of each variety were analysed separately, except for the ant attendance data.

Extrafloral nectary removal was achieved by cutting extrafloral nectaries from all leaves of the treated individuals using a hypodermic needle. This method significantly reduced the presence on the plant of all arthropods attracted by the extrafloral nectaries, while allowing crawling herbivores to reach the plant (Heil 2008). After the initial treatment was applied (between 31 January and 5 February), plants were visited weekly to remove EFNs from newly produced leaves, while nectaries on leaves in the control plants were pricked briefly with the needle. Seed predators were excluded by bagging each developing fruit from treated individuals with mesh bags. Bags were added as fruits matured, and plants were visited weekly to ensure that no developing fruits were unprotected. Each fruit of control individuals was bagged briefly at the beginning of fruit development; in addition, as the fruits approached ripeness, they were bagged to allow seed collection. By applying exclusion treatments at the whole-plant level, we avoided inflating our estimates of ant and seed predator effects, which tend to be higher when paired branches on the same plant are compared (Chamberlain & Holland 2009).

At the end of February, individuals of var. brevipes began to show an abnormal flower bud development. Initially, the ovary of these buds elongated, resulting in the protrusion of the style through the top of the corolla. After a couple of weeks, the corolla and stamens of those buds fell, revealing an elongated, nonfertilized ovary narrower than developing fruits. These flower buds never opened and did not produce pollen. In some cases, after several weeks, the ovary split open longitudinally and generated a new vegetative branch. In these cases, the first leaf would often have six or more leaflets (as opposed to the four leaflets that are the norm for the species), while subsequent leaves were indistinguishable from the regular leaves of var. brevipes. After the onset of this abnormal development, all subsequent flower buds produced by a given branch would present the same condition, which eventually spread to the whole individual and effectively terminated reproduction until the individual died. This abnormal floral bud development was also observed for var. 1 in Fazenda Água Limpa (19º5′ S, 48º21′ W) in 2008 and for C. desvauxii var. mollissima in the Panga Ecological Reserve (19º11′ S, 48º24′ W) in 2007, both in Uberlândia, Brazil. A similar pattern was observed in the following herbarium collections, all from Brazil: in C. desvauxii var. mollissima: Santos 1636 (MO), 10-III-1969 and Irwin 16236 (MO, UB, GH), 29-V-1966, both collected in Rio Turvo, c. 200 km N of Xavantina, Mato Grosso; Irwin 16668 (UB), 6-VI-1966, 60 km N of Xavantina, Mato Grosso; in Chamaecrista diphylla: Rizzo 7 (UB), 5-III-1966, 4 km from Aparecida de Goiás, Goiás; in Chamaecrista ramosa var. parvifoliola: Eiten 10906 (UB), 10-III-1969, Serra do Cipó, Minas Gerais. The cause of this abnormal development was not determined, but its occurrence was not affected by the exclusion treatments (Fisher’s exact test, P = 0.551). In total, 13 of 28 marked individuals were affected, but only in four of them were more than 10% of the flower buds abnormal. For the statistical analyses, we excluded reproductive structures that had this abnormal development and excluded the four individuals with more than 10% of their buds affected (two controls, one EFN removed, one ant and seed predator excluded).

Ant attendance

Ant attendance to EFNs was measured by counting the number of ants on all marked individuals, both during the day and at night. Ant attendance was measured in March and April 2006 at four different periods of the day: early morning, between 6 and 7 h; midday, between 12 and 13 h; afternoon, between 18 and 19 h; and night, between 0 and 1 h. All varieties were sampled during the same time period. In addition, ants were collected on each variety at various times of day throughout the field season for identification. Because of differences in the identity and size of ants among individuals, only presence or absence data were used to establish the success of the EFN removal treatment and differences in ant attendance among varieties. Data from the four time periods were pooled and, owing to the presence–absence nature of these data, differences among varieties and between ant treatments were tested using a generalized linear model with a binomial family and logit link, followed by contrasts among varieties.

Vegetative growth and damage to leaves

We counted the initial number of leaves and leaf scars of each individual as a measure of size. Ten weeks later, the number of leaves and leaf scars was recounted, and vegetative growth was calculated as the ratio between the size after 10 weeks and the initial size. A one-tailed Welch’s t-test, used because of differences in variance between groups, was used to test for differences in growth between EFN-removed and control plants.

In the second week of February 2006, five new leaves were marked on five or six individuals per treatment per variety. Thirty days later, damage to each of the four leaflets per leaf was ranked separately for missing leaf area and area attacked by pathogens. Rank values were 0 (no damage), 1 (up to 25% damage), 2 (between 25% and 50% damage), 3 (between 50% and 75% damage) and 4 (more than 75% damage). The values for each leaflet were summed to obtain a rank value for the whole leaf, and leaf ranks were averaged per individual for statistical analyses. Leaves on fallen or dead branches and dead individuals were excluded from the analyses. Differences in rank between EFN-removed and control individuals were tested using a nonparametric Wilcoxon rank sum test. The Holm variation of the Bonferroni procedure (Holm 1979) was used to adjust P-values for folivory and herbivory levels for each variety, to account for the use of separate analyses for related responses of the same experiment. In the light of the ongoing disagreements surrounding the use of procedures to reduce type I error in the literature (e.g. Nakagawa 2004), we present the nominal (nonadjusted) P-values and include the adjusted values in brackets whenever the nominal P-value is below 0.05.

Reproductive success

The fate of each flower bud produced by all individuals was followed by mapping those buds through their positions on marked branches. Although more than one flower bud might be found at the same position, the buds almost never developed simultaneously. Approximately 10 days were necessary for the buds to develop into flowers, which opened early morning and wilted by midday. Unpollinated flowers were aborted after 2 days, but fruits took between a few days and several weeks to begin development. Fruit and seed maturation took 1 month to complete. Individuals were censused at least once a week throughout the reproductive season, and reproductive structures were classified as bud, flower, ovary (the period between anthesis and the beginning of fruit development), new fruit (enlarged ovary with soft valves) or developing fruit (after valves harden and seeds are filled). In addition, mature fruits were collected to count the number of filled, consumed and aborted seeds per pod. Unenlarged ovules were considered nonfertilized, and the number of enlarged ovules was counted to assess treatment effects on pollination. The presence of small punctures on the fruit valves was used as an indicator of attack by sucking insects, while external damage to a fruit valve, presence of a large exit hole on the valve or presence of frass or chewed seeds indicated attack by chewers.

Negative binomial regressions were used to test for differences in investment in reproductive structures among treatments, using total number of flower buds per individual as a response and total number of leaves and leaf scars as the offset. In addition, differences among treatments in the average number of ovules and the average number of fertilized (enlarged) ovules per fruit per individual were assessed using anovas. The effect of each treatment on the number of open flowers and fruits was estimated using Poisson or negative binomial regressions with either number of flower buds or number of developing fruits as offsets, followed by likelihood ratio tests. These models were chosen because of the count nature of data and heteroscedasticity. A negative binomial was used when there was an indication of overdispersion in the Poisson regression. In cases where many individuals had no flowers, flower buds and/or fruits attacked by herbivores, hurdle regression models were used; the zero counts were modelled with a binomial logit and the positive counts with a negative binomial distribution, and the number of flower buds or developing fruits was used as an offset (Zuur et al. 2009). The number of fruits attacked by sucking and chewing insects was modelled separately. Differences in the total number of filled seeds per individual were tested using the maximum potential number of seeds (i.e. the total number of ovules in developing fruits, directly counted or estimated) as an offset in a negative binomial regression. Two individuals of var. 1 in the ant-exclusion treatment produced no developing fruits and were excluded from the analyses involving fruits. Probability values obtained from the regressions for sucking and chewing insects, for number of open flowers and for fruit and seed set were pooled for each variety and adjusted using the Holm–Bonferroni procedure.

Filled seeds were assumed to be viable, but a seed germination experiment with up to 30 seeds per individual, depending on availability, was also carried out. Seeds were weighed and scarified and placed in Petri dishes with filter paper under a 12:12-hour light cycle for 30 days. Seeds were observed daily for radicle emergence, and germinated seeds were removed from the dishes. Differences in seed weight among treatments were tested using anovas, while differences in the number of germinated seeds were tested using negative binomial regressions with the total number of tested seeds as an offset.

The effect sizes of ants on plant performance (number of flowers, fruits and seeds) in the presence or absence of seed predators were calculated for each variety using a ln-transformed response ratio calculated as L = ln(E(C)/E(E)), where E(C) are expected values for control (ant present) plants and E(E) are expected values for EFN-removed plants. This calculation follows Chamberlain & Holland (2009), so that positive values indicate positive ant effects. All statistical analyses were conducted in R 2.10.1 (R Development Core Team 2009), using packages stats, MASS (Venables & Ripley 2002), multcomp (Hothorn, Bretz & Westfall 2008) and pscl (Zeileis, Kleiber & Jackman 2008).


Extrafloral nectar production and ant attendance

Chamaecrista desvauxii var. brevipes had the largest extrafloral nectaries [3.56 ± 0.61 mm2 (mean ± SD), F2,15 = 75.45, P < 0.001], while var. modesta had nectaries of intermediate size (1.27 ± 0.43 mm2) and var. 1 had the smallest nectaries (0.54 ± 0.19 mm2). Accordingly, nectar production was higher for C. desvauxii var. brevipes (0.073 ± 0.077 μL h−1) than for the other varieties (var. modesta: 0.013 ± 0.017 μL h−1, var. 1: 0.005 ± 0.010 μL h−1; F33,2 = 8.34, P = 0.001). Average nectar secretion was also 23% higher during the day than at night (F33,1 = 4.69, P = 0.038), but there was no interaction between time period and variety (F33,2 = 0.17, P = 0.841).

During the course of this study, four morphospecies of the genera Camponotus, Pheidole, Pseudomyrmex and Linepithema were found on individuals of var. 1, although no ants were observed feeding at extrafloral nectaries of this variety. Var. modesta was visited by a total of eight morphospecies of the genera Camponotus (two morphospecies), Pheidole (two morphospecies, one of which was also found on var. 1), Crematogaster (two morphospecies), Cephalotes and Brachymyrmex. Var. brevipes was visited by all of the morphospecies found in the other two varieties, but also by larger-bodied ants, adding up to 28 morphospecies. These belonged to the genera Camponotus (10 morphospecies), Pheidole (three morphospecies), Pseudomyrmex, Crematogaster (four morphospecies), Ectatomma (two morphospecies), Gnamptogenys, Odontomachus, Solenopsis, Linepithema, Brachymyrmex (two morphospecies), Cephalotes and Paratrechina. Ants were routinely observed patrolling branches with flowers and, on rare occasions, walking inside open flowers, suggesting that the flowers had no ant-repellent qualities. Overall, ant attendance was lower on individuals with cut EFNs [pooled data for all time periods, likelihood ratio test (LRT) = 18.27, P < 0.001] and significantly higher on var. brevipes than on individuals of the other varieties (LRT = 23.03, P < 0.001; Fig. 2).

Figure 2.

 Proportion of individuals visited by ants in control and EFN-removed individuals of each variety of Chamaecrista desvauxii, at four periods of the day (dawn, noon, afternoon and night). EFN, extrafloral nectaries.

Vegetative growth and damage to leaves

In general, damage to leaves was relatively low, and more than 75% of the individuals had damage ranks below 1, corresponding to damage rates below 25%. Folivory was higher in EFN-removed individuals of var. brevipes [W = 14, P = 0.008 (adjusted P: 0.016)]. Median herbivory rank to leaves of plants with EFNs removed was 0.4, while median rank for controls was 0. For the other varieties, the exclusion treatment had no effects on the folivory scores (var. modesta: W = 46, P = 0.335, median = 0 for both treatments; var. 1: W = 43.5, P = 0.625, ant excluded median = 0.2, control = 0.5). Levels of pathogen attack were not affected by EFN removal in any variety (var. brevipes: W = 55.5, P = 0.915, EFN removed median = 0.8, control = 0.9; var. modesta: W = 51.5, P = 0.560, EFN removed = 0.9, control = 1.0; var. 1: W = 43, P = 0.605, median = 1.0 for both treatments). Despite the differences in folivory between individuals of var. brevipes with and without EFNs, vegetative growth over 2 months was not different among treatments for any of the varieties: (var. brevipes Welch’s t15 = 0.00, P = 0.501, EFN removed 27.1 ± 2.0%, control 28.6 ± 1.8 %; var. 1 Welch’s t13.3 = −0.15, P = 0.558, 31.1 ± 2.5 % vs. 29.8 ± 1.3%; var. modesta Welch’s t13 = 1.74, P = 0.053, 3.9 ± 0.1% vs. 9.7 ± 1.2%).

Reproductive success

There were no differences among treatments in the production of flower buds per individual (all P > 0.55). Likewise, the average number of ovules per fruit for each individual was not affected by the exclusion treatments in any of the varieties (all P > 0.220; 17.6 ± 1.6, 8.9 ± 0.8 and 8.3 ± 1.1 ovules per fruit for var. brevipes, var. modesta and var. 1, respectively). In addition, there was no indication of treatment effects on pollination rates, as assessed by the average number of fertilized ovules per fruit per individual (all P > 0.154; 14.2 ± 2.1, 8.2 ± 1.3 and 7.5 ± 1.1 enlarged ovules per fruit for var. brevipes, var. modesta and var. 1, respectively). There were no differences among treatments in rates of herbivore damage to flower and flower buds for any of the varieties (all P > 0.14; Fig. 3a). Fruits of EFN-removed individuals of var. brevipes suffered more damage from sucking insects, although this difference was not statistically significant after correcting for type I error [LRT = 4.90, P = 0.027 (0.161); Fig. 3b]. In addition, ants did not reduce damage from chewers in this variety (LRT = 0.10, P = 0.752; Fig. 3c). Seed predator exclusion significantly reduced attacks by both sucking and chewing insects to fruits of var. 1 [LRT = 9.62, P = 0.002 (0.011) and LRT = 7.07, P = 0.008 (0.039), respectively; Fig. 3b–c]. For var. modesta, the difference in attack by chewing insects between seed predator exclusion treatments was statistically significant before, but not after, adjustment [LRT = 5.59, P = 0.018 (0.108)].

Figure 3.

 Number of flower buds, flowers and fruits attacked by herbivores relative to the total number of each, for each variety of Chamaecrista desvauxii; each point represents one individual. (a) flowers and flower buds attacked by chewing insects; (b) fruits attacked by sucking insects; (c) fruits attacked by chewing insects. Lines connect predicted values for each treatment combination, based on negative binomial models. EFN, extrafloral nectaries.

The number of open flowers relative to the number of initiated flower buds was significantly affected by the EFN removal treatment in var. brevipes: EFN-removed var. brevipes individuals had a smaller percentage of flower buds that opened than controls [LRT 7.82, P = 0.005 (0.016), 60.8% ± 3.3% (mean ± largest SD) for EFN-removed vs. 75.5% ± 3.9% for controls]. In addition, more flower buds opened on EFN-removed var. modesta individuals, but this difference was not statistically significant after adjustment [LRT = 4.47, P = 0.034 (0.102), 49.4% ± 4.0% for EFN-removed vs. 38.9% ± 3.3% for controls]. EFN removal did not affect flower opening in var. 1 (LRT = 0.21, P = 0.647, 62.6% ± 4.4% for EFN-removed vs. 59.3% ± 4.6% for controls). None of the other treatment combinations affected flower opening in any of the varieties (P-values > 0.09).

Extrafloral nectary removal significantly reduced fruit set in var. brevipes, but only in the presence of seed predators [EFN LRT = 9.45, P = 0.002 (0.011); interaction LRT = 7.19, P = 0.007 (0.029), Fig. 4a]. For var. modesta, only seed predator exclusion had a significant effect on fruit set [LRT = 19.7, P < 0.001 (<0.001)], while for var. 1 none of the treatments affected fruit set (P > 0.21). Similarly, EFN removal reduced the total number of filled seeds in var. brevipes [LRT = 11.18, P < 0.001 (0.005)], but the effect was diminished when seed predators were excluded [interaction LRT = 6.10, P = 0.014 (0.041)]. Seed predator exclusion also increased the total number of seeds in var. modesta, although the difference was only marginally significant after adjustment [LRT = 6.03, P = 0.014 (0.070)], and had no significant effect on seed set of var. 1 (LRT = 1.42, P = 0.230).

Figure 4.

 Number of successful fruits and filled seeds for each variety of Chamaecrista desvauxii; each point represents one individual. (a) total number of fruits containing at least one filled seed relative to the number of successfully pollinated flowers; (b) total number of filled seeds relative to the potential number of seeds (i.e. total number of ovules in initiated fruits) per individual. Lines connect predicted values for each treatment combination, based on negative binomial models. EFN, extrafloral nectaries.

There were no differences in seed mass among treatments for any of the varieties (Table 1; F3,17 = 2.23, P = 0.122; F3,23 = 0.76, P = 0.528; F3,23 = 1.47, P = 0.249). Fewer filled seeds of var. brevipes from EFN-removed individuals germinated when compared with seeds from other treatments [Table 1; EFN LRT = 6.16, P = 0.013 (0.039), interaction LRT = 4.27, P = 0.039 (0.078)], but for the other varieties there were no significant differences among treatments (all probabilities > 0.224).

Table 1.   Seed mass [mean ± SD] and % germination [mean (range)] of each variety of Chamaecrista desvauxii and treatment combination; mean for % germination were calculated using arcsine-transformed values and back-transformed. [EFN, extrafloral nectary]
VarietyTreatmentSeed mass (mg)% germination
Var. brevipesControl7.29 ± 0.8581 (43–94)
EFN removed5.45 ± 1.6570 (28–100)
Seed predator excluded7.12 ± 1.5384 (42–97)
EFN and seed predator excluded7.59 ± 0.5392 (70–100)
Var. modestaControl6.12 ± 1.2279 (40–100)
EFN removed6.24 ± 1.8564 (0–100)
Seed predator excluded6.94 ± 0.4295 (74–100)
EFN and seed predator excluded6.62 ± 0.7692 (52–100)
Var. 1Control4.30 ± 0.6284 (14–100)
EFN removed4.28 ± 1.3966 (0–100)
Seed predator excluded4.95 ± 0.6195 (57–100)
EFN and seed predator excluded4.98 ± 0.3098 (81–100)


Our results show that the effectiveness of ants as a defence against herbivores in the C. desvauxii complex depends both on the variety considered and on the presence of seed predators, as determined by the experimental exclusion. The three varieties included in the present study are sympatric and are frequently found in the same areas within the study site. Generally speaking, they are exposed to the same abiotic and biotic conditions. Differences among varieties in the size of their extrafloral nectaries and amount of nectar produced, however, resulted in a gradient of ant attractiveness. Although nectar quality was not determined in the present study, populations of the congeneric C. fasciculata that produce the highest volume of extrafloral nectar also had nectar with the highest amount of sugar (Rios, Marquis & Flunker 2008). In addition, differences in ant attendance among the three C. desvauxii varieties were consistent with the differences in nectar quantity. Only the variety with the highest level of nectar production, var. brevipes, showed higher reproductive success in the presence of ants. However, this benefit was dependent on the presence of predispersal seed predators: When levels of seed predation were reduced through bagging, ants did not significantly increase seed set.

Our data indicate that the relationship between the varieties of C. desvauxii and ants is not specific, a result typical of facultative mutualisms (Bronstein, Alarcon & Geber 2006; Rico-Gray & Oliveira 2007; Chamberlain & Holland 2009). The most common visitors belonged to 10 morphospecies of ground-nesting ants in the genus Camponotus and six morphospecies of Crematogaster. Plants were visited opportunistically by ants in the vicinity, and the identity of the visiting ants varied over time for the same individual. This pattern is common for other cerrado plants that produce extrafloral nectar and likely stems from differences in humidity and temperature preferences among ant species (Oliveira & Pie 1998; Oliveira & Freitas 2004). Despite considerable overlap in the identities of ants found on the three varieties of C. desvauxii, the large nectaries of var. brevipes attracted a more diverse set of ants, including larger-bodied predatory ants in the genera Ectatomma, Gnamptogenys and Odontomachus that may be more aggressive towards large insect herbivores (Davidson & McKey 1993; Rico-Gray & Oliveira 2007). In addition, different ant species are effective against different types of herbivores (Miller 2007). Thus, attracting more than one ant species may further reduce overall herbivory rates, although this is not always the case, especially for obligatory ant–plant mutualisms (Rico-Gray & Oliveira 2007; Chamberlain & Holland 2009; Rosumek et al. 2009).

Excluding ants by cutting the EFN was generally effective (Fig. 2). Although ant presence was detected on some individuals of var. brevipes in the EFN removal treatment, this was largely because of the occurrence of several small-bodied Crematogaster individuals tending the nectaries of newly expanded leaves. Censuses were conducted before these nectaries were removed. In those cases, however, ant activity was restricted to the tips of the branches, and all activity ceased after the new nectaries were removed. In addition, no larger-bodied ants were observed in any EFN-removed individuals. The success of bagging fruits to prevent seed predation was lower for two reasons. First, the timing of bagging was constrained because young fruits are difficult to manipulate and are typically supported by a thin peduncle that can easily snap with the weight of a wet bag, and some fruits were attacked before the ovary was expanded enough to allow bagging. Second, sucking insects were often able to attack fruits through the holes in the material of the bag; thus, seed predator exclusion was mostly restricted to chewing insects. Importantly, early investment in reproductive structures (flower buds and ovules) and pollination rates were not affected by either treatment and therefore do not account for observed treatment differences in fruit and seed set.

As predicted based on EFN size, extrafloral nectar secretion and ant attendance, the effect of EFN removal was only observed in var. brevipes, for which the removal of nectaries resulted in higher folivory levels, fewer flowers, fruits and filled seeds and lower seed germination rates per individual. In contrast, EFN removal had only marginal effects on vegetative growth for var. modesta and no significant effects on var. 1. Despite the lack of significant differences in florivory levels, the percentage of flower buds that opened was negatively affected by EFN removal in var. brevipes. For var. modesta, EFN-excluded individuals tended to have more open flowers, but this increase was not statistically significant after applying the Holm–Bonferroni procedure. While a decrease in vegetative growth with EFN removal is expected if ants attack folivores, an increase in the number of flower buds that reached anthesis is unexpected, although the same (nonsignificant) trend was found in a meta-analysis of ant-exclusion experiments (Rosumek et al. 2009). For var. modesta, it is possible that, as a consequence of eliminating nectar secretion through the removal of extrafloral nectaries, resources that would be used in nectar production were reallocated to flower development, and flower buds that otherwise would be aborted were retained until anthesis. If this scenario is correct, this variety may be more resource-limited than var. brevipes. Moreover, florivory in var. brevipes is likely to have been underestimated because oviposition by moths in small flower buds of this variety may cause early abortion, before any damage is detected. In any case, the difference in flower opening between treatments did not translate into higher fruit or seed production in EFN-excluded plants of var. modesta.

Ants had a stronger effect on postpollination components of reproductive success of var. brevipes (fruit and seed set) than on flower opening (Table 2). However, the effect of ants on fruit and seed set was mitigated by the exclusion of seed predators: when fruits were bagged to reduce seed predation, the benefits from attracting ants largely disappeared (contrast values for var. brevipes between the two columns on Table 2). The fruits of C. desvauxii can be attacked by sucking insects, which insert their proboscis into developing seeds through the fruit valve, or by chewers, in this case either caterpillars that feed externally on the young, soft-valved fruits or caterpillars and weevil larvae that feed internally on the developing seeds. The protection provided by ants against herbivores depended on the type of attack: in var. brevipes, only fruit damage by sucking insects was reduced by increased ant presence (Fig. 4a), although this difference was not significant after using the Holm–Bonferroni procedure.

Table 2.   Mean effect sizes of ants in the presence or absence of seed predators for each variety of Chamaecrista desvauxii. Effect sizes were calculated as ln(E(C)/E(E)) following Chamberlain & Holland (2009), where E(C) and E(E) stand for the expected values of control [extrafloral nectary (EFN) present] and experimental (EFN removed) groups, respectively. Positive values reflect positive effects of ants on plants. Asterisks indicate significant effects of EFN removal (predator present column) or significant interactions (interaction column), with Holm–Bonferroni adjustments in brackets. ***P < 0.001; **P < 0.01; *P < 0.05; NS, not significant
ResponseVarietyAnt effect sizeInteraction
(Pred. present)(Pred. excluded)
Number of flowersVar. brevipes0.201** [*]0.071NS
Var. modesta−0.180* [NS]0.012NS
Var. 1−0.044 NS0.160NS
Fruit setVar. brevipes1.119*** [*]−0.161*** [*]
Var. modesta−0.440 NS0.143NS
Var. 10.124 NS0.002NS
Seed setVar. brevipes1.542*** [***]0.030* [*]
Var. modesta0.262 NS0.352NS
Var. 10.241 NS0.026NS

In facultative ant–plant associations, ants are usually more abundant where the food resources are more abundant (O’Dowd 1979). Concordantly, investment in extrafloral nectar secretion can be adjusted according to the reproductive state of the individual (Koptur & Lawton 1988; Miller 2007), increasing ant activity around more valuable plant parts (i.e. fruits, Holland, Chamberlain & Horn 2009). Indeed, floral nectaries may stay active after the corolla falls, attracting ants that reduce seed predation (Keeler 1981; Rico-Gray & Oliveira 2007). Accordingly, reproductive individuals of var. brevipes had higher nectar production than nonreproductive individuals of similar sizes [0.090 μL h−1 ± 0.065 (mean ± SD) vs. 0.024 μL h−1 ± 0.051, respectively; unpubl. data]. As an extension, it is possible that extrafloral nectar production is variable among other ontogenetic stages in all varieties and that ant defence increases survivorship and/or early growth in seedlings of vars. modesta and var. 1. This possibility should be addressed in future studies.

While ants may provide an effective defence against herbivores for some varieties of C. desvauxii, for others the benefit of attracting these animals is likely negligible. All varieties of C. desvauxii have one or two extrafloral nectaries on the leaf petiole, but the size of this structure is extremely variable among varieties and throughout this species’ geographic range (pers. obs.). Extrafloral nectar production is likely to be costly (O’Dowd 1979; Rutter & Rausher 2004); thus, the reduction or loss in extrafloral nectaries is likely to occur in areas where ant and/or herbivore populations are reduced (Bentley 1977; Rios, Marquis & Flunker 2008). Although the varieties included in the present study are sympatric, populations of var. 1 are often more dense in open, grassy vereda areas prone to seasonal flooding, where the other varieties are rarely encountered. Given that ants attracted by the extrafloral nectaries are mostly ground nesters, mutualistic ant populations in those sites are likely to be small, providing a possible scenario for nectary loss in those individuals. The increased relevance of ant protection for reproductive success in the presence of seed predators in var. brevipes illustrates an alternative mechanism for nectary loss: in areas or years with low herbivory rates, decrease in nectary size or loss altogether would have a positive effect on fitness if nectar production is costly for the individual (Bronstein, Alarcon & Geber 2006). For instance, the number of active extrafloral nectaries in Acacia drepanolobium trees declines when large herbivores are excluded over a long period of time, altering both the identity and the behaviour of the ants that occupy individual plants (Palmer et al. 2008). In C. fasciculata, benefits from ant–plant associations are likewise absent in sites in Florida where either ants or herbivores are scarce (Barton 1986). In Missouri populations of this species, both the total volume and the sugar content of extrafloral nectar are higher in sites where herbivore damage is higher, while density of hairs on the rachis is lower, suggesting a trade-off between those kinds of defences (Rios, Marquis & Flunker 2008).

Our results support the hypothesis that the outcomes of ant–plant mutualisms in C. desvauxii are variable among taxa of the C. desvauxii complex and depend on the biotic context. If extrafloral nectar production is costly, locally isolated populations of C. desvauxii that receive low benefits from ant attraction may be selected for reduced extrafloral nectar production and increased investment in alternative defences. Hybrids between closely related taxa with differing defence strategies can have intermediate defence levels that lead to higher herbivory rates and lower reproductive success than either of the parents (Léotard et al. 2008), reinforcing reproductive isolation. Extrafloral nectaries have a unique origin within the genus Chamaecrista but were secondarily lost in two clades (Conceição et al. 2009), in one case being replaced by glandular setae, suggesting that this scenario for nectary loss may already have occurred within this genus.


The authors thank S. Powell, E.A. Kellogg, R.E. Ricklefs, A.R. Zangerl, L. Abdala-Roberts, N.A. Barber, K.L. Barnett, K. Boege, J.M. Jeffries, H.P. Dutra, J.C. Flunker, R. E. Forkner, J. Landowski, A. Masís, D. Salazar, K. Schultz, G. Wang, the J. Bronstein lab, the Handling Editor and two anonymous reviewers for discussions and advice that helped guide this study, K. Del Claro for logistic support and helpful discussions, M. Ranal for advice on seed germination methods, R. Pacheco for ant identification and S. Powell for help in data collection. Financial support was provided by the Whitney R. Harris World Ecology Center (to B.B.-M.).