In horizontally transmitted mutualisms, mutualists disperse separately and reassemble in each generation with partners genetically unrelated to those in the previous generation. Because of this, there should be no selection on either partner to enhance the other's reproductive output directly. In symbiotic ant–plant mutualisms, myrmecophytic plants host defensive ant colonies, and ants defend the plants from herbivores. Plants and ants disperse separately, and, although ant defence can indirectly increase plant reproduction by reducing folivory, it is unclear whether ants can also directly increase plant reproduction by defending seeds. The neotropical tree Cordia alliodora hosts colonies of Azteca pittieri ants. The trees produce domatia where ants nest at stem nodes and also at the node between the peduncle and the rachides of the infloresence. Unlike the stem domatia, these reproductive domatia senesce after the tree fruits each year. In this study, I show that the tree's resident ant colony moves into these ephemeral reproductive domatia, where they tend honeydew-producing scale insects and patrol the nearby developing fruits. The presence of ants significantly reduced pre-dispersal seed predation by Amblycerus bruchid beetles, thereby directly increasing plant reproductive output.
Mutualistic interactions are ubiquitous and have important effects on ecological communities, but a persistent question is how cooperation is initiated and sustained between species that are composed of evolutionarily selfish individuals. The most common framework for explaining the evolution and persistence of mutualisms is the iterated Prisoner's dilemma, where mutualism results from repeated interactions between partners, and from increased investment by one partner yielding increased rewards from the other [1–3]. These two conditions have been formalized as the mechanism of ‘partner fidelity feedback’ , and the idea that cooperation could actually yield higher net fitness benefits than exploitation  has recently gained theoretical and empirical support [6,7]. Yet when mutualists disperse separately between generations, successful growth of the partners can be coupled via partner fidelity feedback, but the successful reproduction per se of one partner does not depend on that of the other. Indeed, if investment in reproductive output reduces mutualistic investment, then there may be a conflict of interest between the partners and selection for reproductive exploitation.
Ant–plant mutualisms are classic systems for studying the costs and benefits of interspecific cooperation . Ants defend plants from their natural enemies, and plants provide ant colonies with food and, in the case of myrmecophytes, housing . Myrmecophytic plants host symbiotic ant colonies inside plant cavities called domatia. Because a given ant colony associates with a single plant throughout its lifetime, but partners disperse separately, symbiotic ant–plant mutualisms are predicted to be good examples of ‘short-term’ (within-generation) partner fidelity feedback . Consistent with this prediction, recent empirical studies of ant–plant symbioses have demonstrated positive feedback between plant rewards and ant defence [7,10,11]. However, ants may reduce herbivory or even increase plant growth without necessarily increasing plant fitness , and direct evidence that ants increase plant reproductive output is relatively scarce [but see 13–16]. In fact, ants often hinder plant reproduction by deterring pollinators and/or by sterilizing the plant through flower removal [17–22]. These behaviours do not cost the ants (except in those cases where the plant responds with its own sanctions ), because the offspring of current ant–plant partners are unlikely to reassociate in the next generation. Reducing plant reproduction can actually benefit the ants if the plant is diverting resources to reproduction that could otherwise be used for current ant rewards .
Despite these potential conflicts of interest, ants have been shown to increase plant reproductive output by defending flowers from herbivores and pathogens in some ant–plant symbioses [14,15]. Willmer & Stone  suggested that African Acacia zanzibarica plants could manipulate ant defensive behaviour by chemically repelling ants when flowers first dehisce to discourage ants from interfering with pollinators, but reducing these chemical repellants when fruits mature to encourage ant defence against seed predators. However, there was no direct evidence of fruit defence by ants in that study, and very few other studies have measured rates of pre-dispersal seed predation in myrmecophytic plants. Two different studies reported no relationship between the level of ant defence and seed predation by bruchid beetle larvae in the African myrmecophyte Acacia drepanolobium (a close relative of A. zanzibarica) [24,25]. Similarly, in Janzen's classic study of mutualistic coevolution in an ant–plant symbiosis, Pseudomyrmex ants reduced acacia plant mortality, but bruchid beetles still caused more than 99% pre-dispersal seed mortality [26,27]. In some non-symbiotic ant–plant mutualisms, ants attracted to extrafloral nectar have been shown to reduce pre-dispersal seed predation [28,29]. However, myrmecophytes are not commonly reported to produce ant-attracting rewards near developing fruits. Moreover, symbiotic ants, in contrast to free-living ones, may experience selective pressure to redirect plant reproductive energy to growth. To the best of my knowledge, there are still no examples of symbiotic ants that deter seed predators, despite the huge potential benefits to plants.
Here, I investigated whether symbiotic Azteca pittieri ants defend the seeds of their Cordia alliodora host trees in the neotropics. This mutualism is common in seasonally dry tropical forests, where the tree grows during the rainy season and reproduces during the dry season. As the tree grows in the rainy season, it produces new domatia for ants at stem nodes. Ants tend particularly high densities of honeydew-producing scale insects in new stem domatia during the rainy season, and these honeydew rewards seem to promote ant defence of new leaves . When the tree reproduces in the dry season, it produces new domatia on inflorescences (figure 1a). These ‘reproductive domatia’ are noted less frequently than the stem domatia in the literature, even in studies of the tree's reproductive biology [30,31]. In one of the earliest studies of the C. alliodora symbiosis, Wheeler  observed that the reproductive domatia, unlike the stem domatia, senesce each year, drying up after the tree fruits and eventually falling from the tree (figure 1b). Yet, Wheeler did not describe ant behaviour around reproductive domatia, perhaps because adult trees are frequently more than 10 m tall, which makes reproductive structures difficult to observe. More recent studies have noted that ants do not visit tree flowers without mentioning the reproductive domatia or ant behaviour at fruits [30,33].
To test whether reproductive domatia promote ant defence of tree seeds, I investigated ant colonization of the reproductive domatia, ant defensive behaviour on developing fruits and rates of bruchid seed predation between trees with and without active ant colonies. Cordia alliodora suffers from seed predation by Amblycerus spp. bruchid beetle larvae throughout its Middle American range [34–36]. Previous studies have shown that ants can effectively defend C. alliodora leaves [11,37,38]. To the best of my knowledge, this is the first study to investigate ant defence of reproductive structures in this system.
2. Material and methods
(a) Study site and system
This study was conducted at the Chamela-Cuixmala Biosphere Reserve, Jalisco, Mexico (19°30′ N, 105°02′ W), in seasonally dry tropical forest, where annual precipitation averages 748 mm and the dry season lasts six to eight months . The topography is hilly, and tree species that predominate on hillsides, including C. alliodora, are dry-season deciduous . The primary ant symbiont of C. alliodora at this site is a single species, A. pittieri, which occupies approximately 98% of all C. alliodora trees (EG Pringle 2007, unpublished data). Azteca pittieri tends several species of honeydew-producing scale insects, including soft scale insects (Coccidae) and mealybugs (Pseudococcidae), exclusively inside the domatia. Ants must chew into new C. alliodora domatia, which are formed without an opening; scale insects thus depend on ants for entry to domatia. Although scale insects may sometimes disperse independently, ants frequently carry scales among domatia within the host tree (EG Pringle 2007, personal observation).
The fruits of C. alliodora are wind-dispersed, and trees produce both flowers and fruits in the dry season when they are not growing or producing new leaves. In wetter habitats, where its reproductive biology has been most studied, C. alliodora can produce viable seed within 5 years of germination . Time to reproductive maturity is longer in Chamela where growth is slowed by the severe dry seasons. Mature trees at Chamela typically reach a maximum height of approximately 11 m and do not begin flowering until they are approximately 6–7 m tall (EG Pringle 2009, unpublished data). Amblycerus atkinsoni (Coleoptera: Bruchidae) beetles are the most common C. alliodora seed predators at the site .
On 11 October 2011, Hurricane Jova made landfall close to Chamela as a category 3 storm with maximum sustained winds of 195 km h−1 . Many of the large trees in the forest were knocked down; for some tree species, greater than or equal to 30% of mature individuals fell over (K Renton 2012, personal communication). In February 2012, I identified 14 leaning C. alliodora trees along approximately 3.5 km of the trail system that were flowering and fruiting according to their usual phenology (electronic supplementary material, figure S1). In nine of these trees, A. pittieri ant colonies were intact and exhibited typical behaviour. These unique circumstances made it possible to observe ant behaviour on and near reproductive structures by standing on the ground or on a 1.2 m ladder.
(b) Domatia dissections and stable isotopes
To investigate the extent to which ants colonize the reproductive domatia, I inspected 44 reproductive domatia on the 14 leaning trees identified in February 2012. On six of the nine trees where ants were active, I removed and dissected one reproductive domatium and its nearest stem domatium. I looked for ant brood and scale insects in each domatium and identified scale insects to family (Coccidae or Pseudococcidae). Azteca pittieri colonies are monogynous and reproduce only during the rainy season (EG Pringle 2008, unpublished data), so I did not find ant reproductives. I also opened several senesced reproductive domatia on these same trees in June 2012.
To test whether ants inhabiting reproductive domatia received different nutritional rewards than ants inhabiting stem domatia, ant worker bodies (heads and alitrunks) and gasters (posterior abdomens) were analysed for %C, %N, ∂13C and ∂15N. Nutrients in ant bodies have been incorporated into ant tissues, whereas ant gasters, which include the stomach and intestines, reflect the ants’ recent diet . Ants were dried at 60°C for greater than or equal to 48 h. The gaster of each worker was removed, and sets of approximately 20 bodies and of approximately 40 gasters were ground separately with a mortar and pestle. Analyses were performed at the UC Davis Stable Isotope Facility.
(c) Behavioural experiments
To test whether ants defend tree fruits, I conducted behavioural assays in February 2012 to investigate ant patrolling and defensive behaviour. Fruits were full-sized but soft and immature; this stage should correspond to the time of bruchid oviposition. I assessed ant response to tethered, similarly sized Heterotermes convexinotatus (Isoptera: Rhinoermitidae) or Nasutitermes nigriceps (Isoptera: Termitidae) termites. Termites were standardized and accessible baits for these dry-season assays, and aggression towards termites has become a standard behavioural assay for arboreal ants . Although ant response to tethered termites could be more aggressive than to ovipositing bruchids, these behavioural assays provided a replicated assessment of how quickly and effectively ants could thwart an intruder.
To compare ant defence of fruits to that of other tissues, I conducted matched tests by simultaneously pinning one termite to a fruit and another to an old leaf (new leaves are not present in the dry season when C. alliodora fruits). I pinned termites to fruits where the calyx met the marcescent corolla and to leaves that were located immediately posterior to reproductive domatia; the locations of paired termites were approximately equidistant from the reproductive domatium (figure 1a). I conducted paired trials to time how quickly ants found the termite on each structure. These ‘latency to find’ trials ended, and the time was recorded when an ant touched the pinned termite with her antennae or at 30 min. In total, 25 such paired trials were conducted on eight individual trees. When multiple trials were conducted on a single tree, distinct reproductive domatia were used for each trial, and trials were conducted on different days (with one exception, when trials were separated by 4 h). Trials within trees were not completely independent, because there is only one ant colony per tree, but ants never recruited nest-mates from the same domatia for any given trial. In a mixed-effect ANOVA, tree identity explained only 8% of the variance in a restricted maximum-likelihood analysis. Ants thus produced mostly independent foraging and recruitment responses in each trial. Differences between trees were accounted for by analysing the data using a matched-pair Wilcoxon signed-rank test. I also analysed the data using a stratified Kaplan–Meier survival analysis where the 30 min endpoint was treated as right-censored data, indicating that the true latency to find was greater than 30 min.
After a single ant found the termite in the paired trial, I measured ant colony recruitment. Beginning 1 min after the termite was found each minute thereafter for 15 min, I counted the number of ants on the fruit or leaf containing the termite. Because these trials were conducted only when the termite had been found within 30 min in the ‘latency to find’ trials described above, seven individual trees were used for n = 14 recruitment trials for ants on fruits and n = 9 recruitment trials for ants on leaves. These data were analysed in a repeated-measures mixed-effect ANOVA where the number of ants was predicted by the fixed effects of plant structure (fruit or leaf), minute within the 15 min (time), and the interaction between plant structure and time, and by the random effect of tree identity nested within plant structure.
During the 15 min recruitment trials, I also recorded the time at which the ants first began to tear the termite into pieces (‘time to tear apart’), which corresponds to when ants are biting sufficiently that most mobile herbivores would move away. If the ants did not begin to tear the termite into pieces within 15 min, the ‘time to tear apart’ was recorded as the maximum 15 min. Trials were compared between fruits and leaves by a t-test.
(d) Bruchid seed predation
To test whether ant defence of tree fruits leads to lower seed predation by bruchid beetles, I compared bruchid seed predation in trees with and without active ant colonies. Five trees were selected among the nine trees with active ants to compare with the five trees without active ants. Individual trees were separated by greater than or equal to 200 m of trail. Ant colony activity was assessed by vigorously tapping greater than or equal to three reproductive domatia. If ants left greater than or equal to one domatium to patrol, the tree was considered to have ants, whereas if ants did not respond to the disturbance, then the tree was considered to have no ants. Mature fruits were collected from each of these 10 trees in April 2012 just prior to large-scale wind dispersal. Fruits were collected from three infructescences per tree (50 ± 22.5 fruits per infructescence × 3 = 150 ± 40.6 fruits per tree). For each tree, the three infructescences were selected haphazardly and separated by vertical distances of greater than or equal to 0.5 m. Fruits were separated from stems and stored in paper envelopes until sorting in June 2012.
I sorted fruits into the following categories: (i) mature, intact fruits; (ii) mature fruits with bruchid exit holes (figure 3 inset) and (iii) aborted or inviable fruits. In nearly all cases, the seeds of fruits with bruchid exit holes had been entirely eaten, and thus would not germinate at all (EG Pringle 2009, unpublished data). It was not possible to assess what caused aborted or inviable fruits (defined as all fruits without bruchid exit holes that could be crushed), so the estimates of bruchid damage considered here are conservative. Adult bruchid beetles and hymenopteran parasitoids (presumed bruchid predators) present in the envelopes were also counted and collected. Data from trees with and without ants were compared with Wilcoxon tests.
(a) Do ants colonize reproductive domatia?
Ants had colonized 35 of the 44 reproductive domatia I examined and responded defensively to tapping in 29 of these. Worker ants and scale insects were present in the six reproductive and the six stem domatia I dissected from trees where ants were active. However, only 50% of the reproductive domatia contained ant brood, compared with 100% of the stem domatia. In addition, mealybugs (Pseudococcidae) were present in both reproductive and stem domatia, but soft scale insects (Coccidae) were present only in stem domatia. Ants had completely abandoned all dry, senesced reproductive domatia three months later.
(b) Do ants receive distinct nutritional rewards in reproductive domatia?
There were no significant differences in worker-ant carbon or nitrogen content between reproductive domatia and stem domatia (electronic supplementary material, figure S2). Despite the potential for ants to share food among domatia within the colony, ants from reproductive domatia had somewhat lower per cent carbon and nitrogen and ∂13C values than ants from stem domatia. Patterns evident in ant bodies were magnified in ant gasters, suggesting that the marginal nutritional differences could result from recent diet in the different types of domatia.
(c) Do ants defend fruits?
Ants defended fruits more effectively than they defended old leaves. Ants found termites pinned to fruits significantly faster than they found termites pinned to leaves (figure 2a and the electronic supplementary material, figure S3; matched-pair Wilcoxon, p < 0.04; survival analysis, χ2 = 3.66, d.f. = 1, p < 0.06). After finding the termite, ants recruited faster and in higher numbers to termites pinned to fruits than to leaves (figure 2b; slope of recruitment to the maximum number of ants: t-test, t = −2.19, p < 0.04; repeated-measure mixed-effect ANOVA, plant structure: F1,363 = 18.58, p < 0.0001). The pattern of recruitment over time was not different to fruits than to leaves (repeated-measure mixed-effect ANOVA, time: F1,362 = 20.33, p < 0.0001; time × plant structure: F1,362 = 0.06, p = 0.8). Ants were somewhat faster to begin tearing termites apart on fruits than on leaves, but this effect was not significant (‘time to tear apart’: 6.2 ± 0.7 min on fruits, 8.0 ± 1.9 min on leaves, t-test, p = 0.4).
The random effect of tree (and associated ant colony) identity represented only 0.6% of the total variance in the repeated-measure mixed-effect model. In the ANOVA, tree identity had a weak effect on recruitment to fruits (t = −2.54, p < 0.02) and a non-significant effect on recruitment to leaves (t = 0.56, p = 0.6).
(d) Do ants reduce bruchid seed predation?
Trees with ant colonies lost fewer seeds to predation than trees without ant colonies (figure 3; Wilcoxon test, Z = 2.51, p < 0.02). There were no differences in the total number of fruits per infructescence between trees with and without ants (t = −0.88, d.f. = 28, p = 0.4). The number of adult bruchid Amblycerus spp. beetles recovered in seed envelopes was also significantly lower from trees with ants than from trees without ants (Wilcoxon test, Z = 2.21, p < 0.03). Most bruchids were identified as A. atkinsoni, but at least one individual of Amblycerus scutellaris was also present. There were no significant differences between trees with and without ants in the number of parasitoid wasps (Braconidae, Eupelmidae and Pteromalidae) recovered in seed envelopes (Wilcoxon test, Z = 0.67, p = 0.5) or in the proportion of inviable seeds (Wilcoxon test, Z = 0.42, p = 0.7).
These results suggest that C. alliodora trees encourage ants to defend developing fruits, which reduces bruchid seed predation and directly increases plant reproductive success. Worker ants occupied the reproductive domatia with honeydew-producing scale insects when fruits were developing. Ants were faster to find and faster and more numerous to recruit to fruits than to leaves in matched trials. Trees without active ant colonies lost over 30% more of their seed crop to bruchid seed predation than trees that were defended by ants (from 27.4 ± 2.7 and 28.2 ± 2.5 viable seeds per infructescence for trees with and without ants, respectively). Ants moved back to the stem domatia when the reproductive domatia senesced post-reproduction.
Cordia alliodora trees appear to harness ant defence near developing seeds by producing reproductive domatia. Ants’ failure to defend plant seeds from bruchids in previous studies of ant–plant symbioses was associated with the absence of direct food rewards (i.e. nectaries or food bodies) near fruits in those systems [24–26]. Cordia alliodora provides food for ant colonies only indirectly via honeydew-producing scale insects, and the reproductive domatia provide a source of these rewards near the vulnerable seeds. The ants benefit from the reproductive domatia primarily as new sites for tending scale insects, not as additional space, because the ants abandon the reproductive domatia before the rainy season, when colonies grow and reproduce. Nutritional rewards in reproductive domatia may be particularly attractive to ants in the height of the dry season, when trees do not grow but instead pull resources towards reproductive structures . Pringle et al.  showed that young stem domatia host high densities of scale insects, which appeared to encourage ant defence of nearby young plant tissues; reproductive domatia appear to play a similar role by encouraging ant defence of nearby fruits.
What maintains ant defence of seeds in this horizontally transmitted mutualism where the ants do not benefit from increasing tree reproductive output? Ant defence of seeds in this system appears to be a ‘by-product’ [sensu 4,48] of typical ant behaviour around domatia. The tree's investment in reproductive domatia, which indirectly feed the ants, promotes ant defensive behaviours that reflect the colony's interests in foraging and self-protection. Moreover, in the rainy season, these same ant behaviours around stem domatia promote tree growth, and thus may have experienced positive selection from partner fidelity feedback [11,49]. By contrast, partner fidelity feedback seems unlikely to explain ant defence of tree seeds for at least two reasons. First, although the reproductive domatia indirectly provide the ants with nutritional rewards, there is apparently no feedback mechanism whereby ant defence of seeds accrues higher fitness benefits for ants. Moreover, it seems unlikely that these rewards produce fitness benefits for ants that approach those for trees of the more than 30% increase in viable seeds (discounted by the cost of the scale insects ). Although ants may benefit somewhat from the reproductive domatia, nutrient and isotope data indicated that ants did not obtain better diets in reproductive domatia than in stem domatia. Second, partner fidelity feedback occurs only when a mutualist incurs costs in order to benefit its partner. Engaging in defensive behaviour around reproductive domatia may incur small energetic costs to A. pittieri, but these costs are probably offset by the nutritional rewards provided by reproductive domatia. Such low costs for ants are more typical of by-product benefits, and such benefits may in fact commonly enforce mutualism stability .
A few caveats deserve noting. I did not observe ant behaviour when trees were flowering. Cordia alliodora flowers are pollinated by a wide variety of insects, particularly Lepidoptera and Hymenoptera , that could be deterred by ants. It is possible that the benefits ants provide by reducing seed predation are negated by the costs of pollinator deterrence. This possibility seems unlikely, however, for two reasons. First, there was no difference in the proportion of inviable fruits in trees with and without ants, as I would expect if ants had reduced pollination. Second, C. alliodora is a commonly studied agroforestry tree, and previous reports have consistently stated that ants do not visit flowers [30,33]. Indeed, Wheeler  described C. alliodora flowers as having an odour ‘like that of decayed urine’, which is consistent with the hypothesis that C. alliodora, like many other tropical trees, produces ant-repelling flower chemicals [16,51].
Although these results strongly suggest that ants reduce bruchid seed predation, the correlative experimental design does not completely rule out the alternative hypothesis that ants abandon trees that are prone to high levels of bruchid attack. Nevertheless, this alternative explanation seems unlikely for several reasons. First, trees with and without ants were similar sizes and had similar total numbers of fruits, suggesting that trees without ants were otherwise healthy. Second, ant colonies associate with trees for multiple years and cannot move among trees except between generations. Third, ants were completely absent in only one of the five trees designated as not having ants. In the other four trees, ants were present in low numbers, such that they had not colonized reproductive domatia and/or did not respond to disturbance. Small A. pittieri colonies do not provide effective tree defence [11,52], but variation in ant colony size among trees is common and usually reflects the colony's ontogenetic stage .
Finally, the results presented here may vary geographically, and it is worth considering to what extent the reproductive benefits provided by ants, as opposed to the benefits of reduced folivory, influenced the evolution of this mutualism. Cordia alliodora is found throughout the neotropics, from Mexico to Argentina, and the quality of ant defence varies throughout this range . Ants are particularly good defenders at the Chamela site studied here . Interestingly, A. atkinsoni, which was by far the most abundant bruchid seed predator in this study, is endemic to Jalisco, Mexico , and is replaced by different species of Amblycerus farther south . Rates of seed predation may thus also vary among locations.
A question for future research is what factors restrict the distributions of these specialist bruchid seed predators, given that the range of their C. alliodora host tree extends virtually uninterrupted through the neotropics . One important factor could be ambient climate, particularly precipitation in the tropical dry forests where the tree is common . The timing and amount of precipitation determine the tree's reproductive phenology, which could have important consequences for reproductive isolation among species that interact with C. alliodora flowers and fruits. Moreover, precipitation regimes in these forests are changing with the changing climate. Shifts in rainfall could have important consequences for phenology, seed predators, and the fitness benefits and future stability of this ant–plant mutualism.
Raw data: Dryad doi:10.5061/dryad.684mm.
This work was supported by the University of Michigan Society of Fellows.
I thank Felipe Campos-Cerda for his help in the field, Jesus Romero-Nápoles, Enrique Ramírez-García, Beatriz Rodríguez-Velez and Rudolf Scheffrahn for identifications, and Katherine Renton, Todd Palmer and Daniel Janzen for discussion. I am grateful to the Estación de Biología Chamela of the Universidad Nacional Autónoma de México for permission to conduct the work, and to the participants of the 2013 Gordon Research Conference in Plant–Herbivore Interactions for their thoughtful feedback. Noa Pinter-Wollman and Elizabeth Wason provided helpful comments on the manuscript.
- Received February 26, 2014.
- Accepted April 8, 2014.
- © 2014 The Author(s) Published by the Royal Society. All rights reserved.