Abstract We studied the relationship between Hirtella myrmecophila (Chrysobalanaceae), a common but little- studied Amazonian ant-plant that produces leaf-pouches as domatia, and its obligate ant partner, Allomerus octoart- iculatus. Field observations revealed that H. myrmecophi- la drops domatia from older leaves, a characteristic that is unique among myrmecophytes. The physiological mecha- nism for abortion of domatia is currently unknown, but this characteristic allows for the existence, within the same plant, of branches with and without ants. Older branches generally bear only old leaves with no domatia and therefore have no ants, whereas younger branches have leaves of various ages. Ants forage mainly on new leaves, and experimental removal of ants showed that A. octoarticulatus is crucial for defense of these leaves against insect herbivores. However, A. octoarticulatus also acts as a castration parasite, severing the plant?s inflo- rescences. Mature flowers and fruits were only found on older branches with no ants, and flower production was 8 times greater on plants whose ants were experimentally removed than on control plants. Given the reproductive costs inflicted by its mutualistic partner, we suggest that abortion of domatia is a strategy developed by H. myrme- cophila to minimize the effects of cheating by A. octoart- iculatus. These results support the view that evolutionary conflicts of interest between mutualistic species often im- pose selection for cheating on the partner, as well as for mechanisms to retaliate or to prevent super-exploitation. Opposing selection pressures, operating independently on the two partners, probably help to maintain the evolution- ary stability of this mutualistic relationship. Keywords Allomerus ? Myrmecophytes ? Mutualism ? Hirtella ? Herbivory Introduction Myrmecophytes, also known as ant-plants, are plants that have evolved obligate, mutualistic relationships with ants (Janzen 1966; Beattie 1985; Benson 1985; Davidson and McKey 1993). To house ants, these plants have evolved special hollow structures, know as domatia, in which ants nest (Janzen 1966; Beattie 1985; Benson 1985). Many ant-plants also provide food to their associ- ated ants, in the form of nectar or food bodies (Janzen 1966; Baudoin 1975; Janzen 1975; O?Dowd 1980; Beattie 1985; Vasconcelos 1991). In exchange, ants often protect plants against herbivores (Janzen 1966; McKey 1984; Benson 1985; Vasconcelos 1991; Davidson and McKey 1993; Fonseca 1994; Federle et al. 1998), against encroaching vines and competing plants (Janzen 1966; Benson 1985; Davidson and McKey 1993; Federle et al. 1998), or provide nutrients essential for plant growth (Janzen 1966; Treseder et al. 1995). However, not all as- sociated ants are mutualistic. Some ant species act like parasites by utilizing domatia and food rewards without providing benefits (Janzen 1975; McKey 1984; Gaume and McKey 1999), while others prune the reproductive or vegetative structures of their host-plants, thus nega- tively affecting plant growth and reproduction (Yu and Pierce 1998; Stanton et al. 1999; Yu 2001). Among ant species protecting myrmecophytes from herbivores (Vasconcelos 1991; Davidson and McKey 1993; Fonseca 1994), many do not forage or nest off their hosts, and for these species colony growth and re- productive success depend strongly on host-plant growth (Fonseca 1993, 1999). However, ant-plants can theoreti- cally control the production of domatia, maintaining the available space for the colony at an optimum size that better reflects the plants? interest in terms of defense al- location (Fonseca 1993). If the plant?s allocation to domatia is not the optimal solution for the ants, the inter- T.J. Izzo ? H.L. Vasconcelos (?) Coordena??o de Pesquisas em Ecologia, Instituto Nacional de Pesquisas da Amaz?nia (INPA), C.P. 478, 69011-970 Manaus, AM, Brazil H.L. Vasconcelos Present address: Instituto de Biologia, Universidade Federal de Uberl?ndia, C.P. 593, 38400-902 Uberl?ndia, MG, Brazil, e-mail: heraldo@umuarama.ufu.br, Tel.: +55-34-32182243, Fax: +55-34-32182243 Oecologia (2002) 133:200?205 DOI 10.1007/s00442-002-1027-0 P L A N T A N I M A L I N T E R A C T I O N S Thiago J. Izzo ? Heraldo L. Vasconcelos Cheating the cheater: domatia loss minimizes the effects of ant castration in an Amazonian ant-plant Received: 7 January 2002 / Accepted: 11 July 2002 / Published online: 20 August 2002 ? Springer-Verlag 2002 201 ests of the two partners are in conflict (Fonseca 1999), and cheating ? the use of mutualistic resources or servic- es without providing any benefits in return ? can evolve. This study focuses on the relationship between Hirtel- la myrmecophila (Chrysobalanaceae), a common but lit- tle-studied Amazonian ant-plant that produces leaf- pouches as domatia, and its obligate ant partner, All- omerus octoarticulatus. We begin by showing that H. myrmecophila loses its domatia from older leaves. We then analyzed the effects of A. octoarticulatus against plant herbivores and on plant reproduction. Given the re- productive costs incurred by its mutualistic partner, we suggest that abscission of domatia is a strategy devel- oped by H. myrmecophila to minimize the effects of cheating by A. octoarticulatus. Materials and methods Study area The study was conducted at an 800-ha forest preserve, run by the Biological Dynamics of Forest Fragments Project (a collaborative project between INPA ? the Brazilian National Research Institute for the Amazon ? and the Smithsonian Institution). This preserve is situated about 70 km north of Manaus (2?25?S, 59?48?W), with- in an area of approximately 500,000 ha of relatively undisturbed, upland (terra-firme) Amazonian rain forest that is being devel- oped by the Manaus Free Trade Zone Authority (SUFRAMA). The preserve is on moderately rugged terrain, dissected by small creeks, and lies at an elevation of 50?100 m. Canopy height of forest trees is about 35 m, with some emergent trees reaching up to 50 m. The understory is relatively open and characterized by an abundance of stemless palms. Precipitation in Manaus averages 2,100 mm annually and varies seasonally, with a rainy period be- tween November and May and a dry period between June and Oc- tober (Ribeiro 1976). Study species The genus Hirtella (Chrysobalanaceae) has 98 species, of which only 7 are mymercophytic (Prance 1972). H. myrmecophila is a small (<10 m) understory tree commonly found in non-flooded forests of the central Amazon. It produces ant cavities (domatia) at the base of the leaves, but these domatia are subsequently lost as the leaf ages (Fig. 1). The physiological mechanism that promotes the abortion of domatia is currently unknown. The domatia of H. myrmecophila are highly vascularized, and are irrigated by sec- ondary veins independently from the remaining leaf lamina. Therefore, disruption of sap flux to those secondary veins could cause the necrosis of the domatia without affecting the leaf lami- na. H. myrmecophila reproduces year-round. It has fasciculate in- florescences, 1?3.5 cm long, which are produced usually at the tip of the branches (Fig. 1). In mature forests near Manaus (Brazil), this myrmecophyte has only one obligate associate, the ant A. octoarticulatus. Of 600 plants inspected (Izzo 2001), 583 (97.16%) hosted colonies of A. octoarticulatus, 12 plants were un- inhabited by ants, and the remaining 5 plants housed either Cre- matogaster limata parabiotica (n=4) or Azteca sp. (n=1). The last two species, however, were never found nesting in the plant, but only tending scale insects (Izzo 2001). A. octoarticulatus is a tiny ant (<2 mm), relative of the fire- ants, which lives exclusively in myrmecophytes, including also Duroia saccifera (Rubiaceae) and H. physophora in central Ama- zonia (Fonseca 1999; Ribeiro et al. 1999). A taxonomic revision of the genus Allomerus is urgently needed, as in many cases the species delimitations are unresolved. For instance, in our study site about 3% of the plants were occupied by A. octoarticulatus var. septemarticulatus, which is distinguished from A. octoarticu- latus by the number of antennal segments (seven in septemarticu- latus and eight in octoarticulatus sensu stricto). Apart from this single difference in worker morphology, we have not found any other morphological, behavioral or ecological difference between the two species varieties, so we assume these to be a single spe- cies. In spite of that, all observations and experiments described below were performed with plants inhabited by A. octoarticulatus sensu stricto. A. octoarticulatus does not tend scale insects and does not for- age off its host-plant. Its major food source appears to be insects that venture onto the plant foliage. During field observations, we saw A. octoarticulatus attacking and carrying to the interior of the domatia several kinds of insects, including caterpillars, beetles, and Homoptera, as well as termites that we experimentally placed onto the leaf lamina. The predatory behavior of A. octoarticulatus is very similar to that of A. decemarticulatus (Dejean et al. 2001). Fig. 1 a Young branch of the ant-plant, Hirtella myrmecophila, containing new and mature leaves with domatia (leaf pouches lo- cated at the base of the leaf, where associated Allomerus octoart- iculatus ants nest). b On the same plant, older branch with leaves where domatia were aborted (right), and an inflorescence (left) Distribution of foraging ants according to leaf age We counted the number of ants foraging on the leaf surface in 4 leaves per plant, for a total of 57 randomly selected plants (1?2.5 m tall). These were instantaneous counts, performed during the day, once for each plant. In each plant, we selected one new expanding leaf, one mature leaf, and two old leaves (one from a branch containing only old leaves, and one from a branch with new and mature leaves). Old leaves were defined as mature leaves that had lost their domatia (Fig. 1b), or that were about to lose the domatia (in the latter case, the domatia was already in the process of necrosis and was not utilized by ants). In addition, old leaves, in contrast to mature leaves, were frequently covered by epiphylls. The term ?old? was used here to emphasize that these leaves were older than mature leaves with domatia. Old leaves were generally not senescing leaves, since many of the old leaves we marked sur- vived for over 1 year. Of 84 marked old leaves during the ant-re- moval experiment (see below), only 10 (11.9%) died after 18 months. To remove the effects of leaf size, ant activity was expressed per unit of leaf area (workers per 10 cm2 of leaf) rather than per leaf. Leaf area was measured using a transparent plastic grid with a precision of 0.25 cm2. Ant-removal experiment We selected 40 H. myrmecophila trees 1?2.5 m high. This size was chosen to facilitate access to all branches. The plants were as- signed randomly into two treatments: ant removal by application of organophosphate insecticide Malathion (Indol do Brasil) inside all domatia, or control (plants whose ants were maintained). In to- tal, four plants were lost during the experiment to natural mortali- ty, e.g. that resulting from treefalls. There were no initial differ- ences in total leaf numbers of control and treatment plants (mean?SE: control=84.5?6.7; treatment=85.2?8.0). Production of new leaves and flowers was recorded monthly during a 9-month period. At the end of that period, we recorded the total number of leaves per plant. These data were used to calculate the percent leaf increment per plant as: [(Nf/Ni)?1]*100, where Ni is the initial number of leaves and Nf is the number of leaves 9 months after the beginning of the experiment. Rates of leaf herbivory (percent leaf area damaged per month) were recorded for new, expanding leaves for a period of 1 month, and for mature and old leaves for a 2-month period. In each plant, we randomly selected three leaves per age category. New leaves were those produced after the beginning of the experiment. For each leaf we recorded, monthly, the total leaf area, and the area damaged. Differences in herbivory rates between control and treatment plants were assessed using the Mann-Whitney U-test. A separate test was done for each leaf category. Since the same hy- pothesis was being tested several times, probabilities were correct- ed using the Bonferroni procedure. All analyses were done using SYSTAT (Wilkinson 1996). Effects on plant reproduction To further assess the effects of A. octoarticulatus on plant repro- duction, we conducted two experiments. First, we selected 16 trees that possessed undamaged young inflorescences on branches with ants. From one branch we removed, using insecticide, all ants from the most distal three leaves, which were then isolated from the remaining leaves with Tanglefoot (The Tanglefoot Company, Mich.). Another branch from the same plant was marked as a con- trol, and remained unmanipulated. The second experiment took advantage of the fact that in H. myrmecophila two types of branches are found. Younger branches (as those used in the above described experiment) have ants, whereas older branches bearing only old leaves do not (see Re- sults). In each of 11 randomly selected trees, we tied 1 old branch with at least 1 young inflorescence to a nearby young branch with 202 no flowers. As a control, we tied together two old branches, one with a young inflorescence and one without any flowers. In both experiments, we followed the fate of each marked in- florescence for 1 month, recording if the inflorescence died or be- came a fully formed (mature) inflorescence. Results Distribution of foraging ants according to leaf age and branch type Ant activity was highly concentrated on new, expanding leaves; the average number of foraging ants per unit of leaf area was an order of magnitude greater on newly ex- panding leaves than on mature (Wilcoxon signed ranks test Z=6.6; P<0.001) or old leaves (Z=6.6; P<0.001; Ta- ble 1). We never found foraging ants on old leaves in branches containing only old leaves, whereas a few ants were found on old leaves from branches containing also new and mature leaves (Table 1). Therefore, within the same plant, there were branches with and without ants. The proportion of branches with and without ants was found to be variable among trees, probably due to tree age and environmental factors; however, on average 65% (SE=2.2; n=42 plants) of the branches from a given tree housed ants in all or most of their leaves, whereas 35% of the branches bore only old leaves lacking ants. Ant-removal experiment Experimental removal of ants from entire individuals of H. myrmecophila resulted in significantly higher herbi- vory rates, particularly for newly expanding leaves (Ta- ble 2). New leaves not protected by ants were heavily at- tacked by leaf-eating insects, especially caterpillars, bee- tles, and grasshoppers. Levels of herbivore damage on mature and old leaves were very low and did not differ between control and treatment plants (Table 2). Table 1 Effects of leaf type and branch type on the distribution of foraging ants (Allomerus octoarticulatus) in the ant-plant Hirtella myrmecophila. Younger branches were those that presented leaves of different ages, most of which had domatia. Older branches were those that contained only old leaves without domatia. No ants were found in old leaves from older branches, whereas in younger branches ant activity was highly concentrated on new, expanding leaves. Values represent means?SE (n=57 plants) Branch type Leaf type Abundance of foraging ants (number per 10 cm2 of leaf surface) Young New (expanding) ? 5.48?0.79 with domatia Mature ? with domatia 0.11?0.02 Old ? without domatia 0.01?0.01 Old Old ? without domatia 0 203 Treatment plants produced threefold fewer new leaves than did control plants within a period of 9 months (Ta- ble 2). Moreover, 57% of the few leaves produced by treatment plants were completely lost to herbivores 1 month after emergence, compared to just 1.4% of those from control plants (Table 2). Due to both lower rates of leaf production and elevated rates of herbivore damage, ant-removal plants exhibited negative growth over the course of the experiment, whereas numbers of leaves in- creased by 12% on average in control plants (Table 2). Removal of ants also dramatically affected flower pro- duction. Despite suffering more herbivory, treatment plants produced an average of 8 times more flowers (in- florescences) than did control plants (Table 2). Effects on plant reproduction The presence of inflorescences on branches with ants al- ways elicited strong recruitment of ants. Recently formed inflorescences were always covered with ants, and within 2 or 3 days became dry and died. Due to the small size of A. octoarticulatus and its sensibility to proximity of a human observer, we were not able to di- rectly observe and describe the castration behavior in de- tail. However, results from our field observations and ex- periments provide strong indirect evidence for the role of A. octoarticulatus as a castration parasite of H. myrme- cophila. First, we never found fully formed flowers or fruits in young branches of ant-inhabited plants, whereas 49.1% (?4.9 SE, n=42) of the old branches of these same plants bore mature flowers or fruits. Second, we fre- quently found young, dead flowers on young branches (with ants) but not on old ones (without ants) (10.6?2.0% of young branches with at least 1 dead inflo- rescence vs 1.8?0.9% of old branches; Wilcoxon signed ranks test: Z=4.2; P<0.001). Third, when young inflores- cences were protected from ants by removing ants from the three most distal leaves, all but one inflorescence survived, whereas on the branches with ants all inflores- cences died (?2=24.6; P<0.001). Finally, when we tied one branch without ants and with at least one young in- florescence to a branch with ants, the ants destroyed all inflorescences. In contrast, on control branches, all inflo- rescences survived (?2=18.2; P<0.001). Discussion Our results suggest that while A. octoarticulatus protects H. myrmecophila from insect herbivory, it also lowers its fitness by castrating flowers. In response, H. myrmecop- hila appears to abort domatia in order to force ant aban- donment from older branches and therefore flowering on these same branches. We now discuss each of these con- clusions and their implications for mutualism theories. The role of Allomerus in plant defense Experimental removal of A. octoarticulatus resulted in strong herbivore damage to H. myrmecophila young leaves, indicating that young leaves rely heavily on ants for defense. Mature leaves and old leaves are not defend- ed by ants, since virtually no damage was recorded after ant removal (Table 2). The number of patrolling ants on young, expanding leaves was an order of magnitude greater than on mature leaves (Table 1). This difference cannot be attributed to differences in nest location, since A. octoarticulatus occupied domatia of young and ma- ture leaves equally. A recent study with Leonardoxa africana shows that associated Petalomyrmex ants are at- tracted to young leaves by leaf volatiles (Brouat et al. 2000), but if a similar phenomenon occurs in H. myrme- cophila remains to be seen. In general, our results support earlier findings with an African (Leonardoxa africana) (McKey 1984) and an Asian (Crypteronia griffithii) (Moog et al. 1998) my- rmecophyte. In these plants, contrasting with many other ant-plants studied so far (e.g., Janzen 1966; Vasconcelos 1991; Fonseca 1994), chemical defenses have not been completely replaced by biotic (ant) defenses (McKey 1984; Moog et al. 1998). Rather, biotic and chemi- cal/mechanical defenses coexist, each restricted to leaves of a different age class. Why do some myrmecophytes rely completely on biotic defenses whereas others do Variable Control (with ants) Treatment (ants removed) U-test statistic P % leaf area damaged in 1 month ? new leaves 1.17?0.77 54.22?6.58 3.0 <0.001* % leaf area damaged in 1 month ? mature leaves 0.06?0.02 1.20?0.60 117.0 0.18* % leaf area damaged in 1 month ? old leaves 0.40?0.13 0.25?0.09 189.5 1.00* % of new leaves that did not reach maturitya 1.39?0.89 56.96?5.86 10.0 <0.001 Number of new leaves produced in 9 months 17.06?1.96 6.79?1.06 285.0 <0.001 % leaf increment in 9 months 11.89?3.76 ?5.16?2.88 263.5 0.001 Number of inflorescences produced in 9 months 0.88?0.39 6.74?2.31 87.0 0.013 *Bonferroni-adjusted probability aAborted or completely eaten by herbivores 1 month after emergence Table 2 Effects of ant removal on leaf and flower production, and on herbivory rates (% total leaf area damaged in 1 month) for new, mature, and old leaves of Hirtella myrmecophila. Ant removal significantly increased herbivory in new leaves, negatively affect- ed plant growth, but increased flower production. Values represent means?SE. For treatment plants n=19 in all cases, and for control plants (ant-maintained) n=19 for data on herbivory rates and n=17 for the remaining variables not? Possibly, for ant-plants whose leaves are very long- lived, the cost of maintaining a large worker force of ants throughout the life of the leaf is too high relative to the cost of providing leaves with permanent chemical or mechanical protection. In this case, a trade-off from biot- ic to chemical and mechanical defenses as the leaf ages is expected (McKey 1984, 1988). H. myrmecophila, like L. africana and Crypteronia griffithii, also has long-lived leaves (>2 years; T.J. Izzo, unpublished results). Howev- er, leaf longevity may not be the only factor involved. Studies with Tachigali myrmecophila, an Amazonian ant-plant whose leaves live more than 6 years, do not support the trade-off model of chemical and ant defenses (Fonseca 1994). The extremely high rates of herbivory observed on new H. myrmecophila leaves whose ants were removed, coupled with the fact that ant-removed plants were not able to replace the leaves lost to herbivores, suggest that these plants are likely to die if not recolonized by ants. Allomerus as a castration parasite of Hirtella myrmecophila Mature flowers and fruits were only found in older branches without ants, suggesting that either: (1) A. octo- articulatus destroys the reproductive structures of its host-plant, or (2) there is a physiological difference be- tween branches with and without ants and, for some rea- son, those with ants discontinue flower development. Our experiments indicate that that there are no physio- logical differences between branches with and without ants, but rather that A. octoarticulatus only destroy flow- ers from the branches they inhabit. Therefore, A. octoart- iculatus castrates its host-plant. This behavior is also seen in its congener A. demerarae which lives in Cordia nodosa (Boraginaceae) (Yu and Pierce 1998; Yu 2001). Destruction of H. myrmecophila flowers by A. octo- articulatus always occurred very early in flower devel- opment. The ants attacked floral buds immediately after they were produced. In that sense, the energetic loss to the plant is less than if attacks occurred only after flow- ers had matured or after fruit set. This castration behav- ior is probably beneficial for the ant colony, because it increases nesting space (Yu and Pierce 1998; Yu 2001), given the likely trade-off between plant growth and re- production (Baudoin 1975; Clay 1990; Yu and Pierce 1998; Stanton et al. 1999; Kover 2000; Yu 2001). Be- cause both leaves and flowers of H. myrmecophila are produced at the branch tips (Fig. 1), a given branch may not be able to produce the two structures concomitantly. In the Cordia-Allomerus system, as in other ant-plant systems in which ant parasitism is recorded, plant popu- lations persist because other obligate plant-ant species that do not affect plant reproduction inhabit some plants (Janzen 1975; McKey 1984; Yu and Pierce 1998; Gaume and McKey 1999; Yu 2001). However, in the Hirtella- Allomerus system, only one ant species inhabits the plant. Although A. octoarticulatus is crucial for the de- fense of young H. myrmecophila leaves, it also interferes with plant reproduction. Therefore, the abortion of domatia from old leaves ? which do not rely on ants for defense ? appears to be a strategy evolved by H. myrme- cophila in order to ?cheat the cheater?. With this strate- gy, the plant reduces the reproductive losses caused by its mutualistic partner while maintaining the benefits and stability of the association. Alternatively, domatia abor- tion could be simply a response to damage caused by pathogens, herbivores, or an ant-predator (e.g., Federle et al. 1999). However, this hypothesis appears unlikely. During our field observations, we never detected any or- ganism causing this type of damage. Moreover, experi- mental removal of ants did not cause subsequent necrosis and abortion of domatia. We have not observed abortion of domatia in D. sac- cifera or in H. physophora, even though these plants are also associated with A. octoarticulatus. We do not have any observations on flowering and fruiting in D. saccif- era, whose height at maturity (>10 m) makes it difficult to monitor its reproductive phenology. However, prelim- inary observations with H. physophora reveal that inflo- rescences are most frequently produced on the tree trunk, where Allomerus ants do not forage. In addition, the peti- oles of the inflorescences of H. physophora are thicker and more densely covered with trichomes than those of H. myrmecophila, which may render them less suscepti- ble to ant damage. These data suggest that flower mor- phology and flower position within the plant may be im- portant determinants of whether or not plants are suscep- tible to ant castration. Although phylogenetic studies with myrmecophytic Hirtella are clearly needed, our pre- liminary observations with H. physophora indicate that domatia loss is not a common trait among all the species in the genus, and thus probably originated secondarily in H. myrmecophila. Implications for the evolution of mutualisms To our knowledge, this is the first study to demonstrate empirically that by modifying their morphology, plants can avoid castration by associated ants. Prior studies have suggested this could be the case, though it was ob- served in only a few individuals and intra-individual ex- pression of the trait was highly variable (Yu and Pierce 1998). In contrast, the abortion of domatia in H. myrme- cophila as a leaf ages appears to be not only consistent within an individual?s lifetime, but also characteristic of the species as whole. The Hirtella-Allomerus system resembles other spe- cies-specific mutualisms where mechanisms have evolved to enhance benefits of association and reduce costs due to cheating, and where such mechanisms pro- mote the stability of the relationship (Tyre and Addicott 1993; Pellmyr and Huth 1994; West and Herre 1994; Pellmyr et al. 1996). In mutualistic systems involving Ficus and Yucca and their respective pollinators, the plants ?pay? their specific pollinators with potentially vi- 204 Gaume L, McKey D (1999) An ant-plant mutualism and its host- specific parasite: activity rhythms, young leaf patrolling, and effects on herbivores of two specialist plant-ants inhabiting the same myrmecophyte. Oikos 83:130?140 Izzo T (2001) Influ?ncia de Allomerus octoarticulatus sobre a herbivoria e reprodu??o de Hirtella mymercophila. MSc The- sis, INPA/Universidade do Amazonas, Manaus Janzen DH (1966) Coevolution of mutualism between ants and acacias in Central America. Evolution 20:249?275 Janzen DH (1975) Pseudomyrmex nigropilosa: a parasite of a mutualism. Science 188:936?937 Kover PX (2000) Effects of parasite castration on plant resource allocation. Oecologia 123:48?56 McKey D (1984) Interaction of the ant-plant Leonardoxa africana (Caesalpinaceae) with its obligate inhabitants in a rain forest in Cameroon. Biotropica 16:81?99 McKey D (1988) Promising new directions in the study of ant- plant mutualisms. In: Zimmer B (ed) Proceedings of the XIV International Botanical Congress. Koeltz, K?nigstein/Taunus, pp 335?355 Moog J, Drude T, Maschwitz U (1998) Protective function of the plant-ant Cladomyrma maschwitzi to its host, Crypteronia griffithii, and the dissolution of the mutualism (Hymenoptera; Formicidae). Sociobiology 31:105?129 O?Dowd DJ (1980) Pearl bodies of a neotropical tree Ochroma pyramidale: ecological implications. Am J Bot 67:543?549 Pellmyr O, Huth CJ (1994) Evolutionary stability of mutualism between yuccas and yucca moths. Nature 372:257?260 Pellmyr O, Leebens-Mark J, Huth CJ (1996) Non-mutualistic yucca moths and their evolutionary consequences. Nature 380: 155?156 Prance GT (1972) Crysobalanaceae. Flora Neotropica. M 9. Hafner, New York Ribeiro JEL, Hopkins MJG, Vicentini A, Sothers CA, Costa MA, Brito JM, de Souza MAD, Martins LHP, Lohmann LG, Assun??o PACL, Pereira EC, da Silva CF, Mesquita MR, Proc- opio LC (1999) Flora da Reserva Ducke: guia de identifica??o das plantas vasculares de uma floresta de terra-firme na Ama- z?nia. INPA-DFID, Manaus Ribeiro MNG (1976) Aspectos climatol?gicos de Manaus. Acta Amazonica 6:229?233 Stanton ML, Palmar TM, Evans A, Turner ML (1999) Steriliza- tion and canopy modification of a swollen thorn acacia tree by a plant-ant. Nature 401:378?381 Treseder KK, Davidson DW, Ehleringer JR (1995) Absorption of ant-provided carbon dioxide and nitrogen by a tropical epi- phyte. Nature 375:137?139 Tyre AJ, Addicott JF (1993) Facultative non-mutualistic behaviour by an ?obligate? mutualist: ?cheating? by yucca moths. Oeco- logia 94:173?175 Vasconcelos HL (1991) Mutualism between Maieta guianensis Aubl. a myrmecophytic meslatome, and one of its ant inhabit- ants: ant protection against insect herbivores. Oecologia 87:295?298 West SA, Herre EA (1994) The ecology of the New Word fig-pa- rasiting wasps Idarnes and implications for the evolution of the fig-pollinator mutualism. Proc R Soc Lond B 258:67?72 Wilkinson L (1996) SYSTAT 6.0 for Windows: Statistics. SPSS, Chicago, Ill Yu DW (2001) Parasites of mutualisms. Biol J Linn Soc 72: 529?546 Yu DW, Pierce NE (1998) Castration parasite of an ant-plant mu- tualism. Proc R Soc Lond B 265:375?382 205 able seeds for the growing larvae, in exchange for the pollination services provided by adult female insects. Therefore, the decrease in plant reproductive success is counter-balanced immediately by the pollination service (Tyre and Addicott 1993; Pellmyr and Huth 1994; West and Herre 1994; Pellmyr et al. 1996). In constrast, H. myrmecophila suffers a direct loss in reproductive suc- cess in exchange for an indirect increase in fitness ob- tained through protection against herbivores. Our study shows the importance of herbivory as a key factor behind the evolution of myrmecophytism in H. myrmecophila, and probably also in many other ant- plants. It also reinforces the view that, in mutualistic sys- tems, there is often a conflict of interests between part- ners, and that a mechanism to prevent cheating is neces- sary to stabilize the system. Acknowledgements We thank Allen Herre, Diane Davidson, Emilio Bruna, Phil Ward, Roger Hutchings, William Laurance, and two anonymous referees for commenting on drafts of this manuscript. The Brazilian Council of Research and Scientific De- velopment (CNPq) and the Biological Dynamics of Forest Frag- ments Project (a joint project between INPA and the Smithsonian Institution) provided research support. References Baudoin M (1975) Host castration as a parasitic strategy. Evolution 29:335?352 Beattie AJ (1985) The evolutionary ecology of ant-plant mutual- isms. Cambridge University Press, Cambridge Benson WW (1985) Amazon ant-plants. In: Lovejoy TE (ed) Amazonia. Pergamon Press, Oxford, pp 239?266 Brouat C, McKey D, Bessiere J-M, Pascal L, Hossaert-McKey M (2000) Leaf volatile compounds and the distribution of ant patrolling in an ant-plant protection mutualism: preliminary results on Leonardoxa (Fabaceae: Caesalpinioideae) and Pet- alomyrmex (Formicidae: Formicinae). Acta Oecol 21:349?357 Clay K (1990) Parasitic castration of plants by fungi. TREE 6:162?166 Davidson DW, McKey D (1993) The evolutionary ecology of ant- plant relationships. J Hymenopt Res 2:13?83 Dejean A, Solano PS, Belin-Depoux M, Cerdan P, Corbara B (2001) Predatory behaviour of patrolling Allomerus decemart- iculatus workers (Formicidae; Myrmecinae) on their host- plant. Sociobiology 37:571?578 Federle W, Maschwitz U, Fiala B (1998) The two partner system of Camponotus (Colobopsis) sp.1 and Maracanga puncticul- ata (Euphorbiaceae): natural history of the exceptional ant partner. Insectes Soc 45:1?16 Federle W, Leo A, Moog J, Azarae HI, Maschwitz U (1999) Myrmecophagy undermines ant-plant mutualisms: ant-eating Calloscirus spp. squirrels (Rodentia; Sciuridae) damage ant- plants in SE Asia. Ecotropica 5:35?43 Fonseca CRS (1993) Nesting space limits colony size of the plant- ant Pseudomyrmex concolor. Oikos 67:473?482 Fonseca CRS (1994) Herbivory and the long-lived leaves of an Amazonian ant-tree. J Ecol 82:833?842 Fonseca CRS (1999) Amazonian ant-plant interactions and the nesting space limitation hypothesis. J Trop Ecol 15:807?825