An Adaptation Or a Sensory Trap?

ANIMAL BEHAVIOUR, 2007, 74, 377e385 doi:10.1016/j.anbehav.2006.07.022

Protection in an anteplant mutualism: an adaptation or a sensory trap?

DAVID P. EDWARDS*,ROXANAARAUCO†, MARK HASSALL‡, WILLIAM J. SUTHERLAND*, KEITH CHAMBERLAIN§,LESTERJ.WADHAMS§ &DOUGLASW.YU* *School of Biological Sciences, University of East Anglia yFacultad de Ciencias Naturales y Matematicas, Universidad Nacional Federico Villarreal zSchool of Environmental Sciences, University of East Anglia xBiological Chemistry Division, Rothamsted Research

(Received 16 March 2006; initial acceptance 16 May 2006; ﬁnal acceptance 7 July 2006; published online 15 August 2007; MS. number: 8884R)

Many traits of ant plants and their ant symbionts are thought to be coevolved, but there is little evidence for adaptation in these symbioses. We investigated the ant trait of worker attraction to, and consequent patrolling of, new plant shoots, and we tested two hypotheses to explain the maintenance of this trait. (1) New shoots chemically mimic ant brood or alarm pheromones (a ‘sensory trap’) and thereby elicit worker patrolling of vulnerable plant parts. (2) Worker attraction to new shoots is the result of selection on the ant to direct patrolling to the plant parts that maximize the capture of plant-provided rewards. As our model system, we used the ant plant Cordia nodosa and its protecting ant symbiont Allomerus octoarticulatus var. demerarae. Gas chromatography analyses suggested that compounds were shared between new leaves and Allomerus brood, and Allomerus workers were attracted to brood extracts of nonself colonies, findings that are consistent with the sensory trap hypothesis. However, patrolling Allomerus workers were attracted only to new leaves, whereas brood-tending workers collected from inside plant domatia (‘nurses’) were attracted to Allomerus brood rather than to new leaves. Only patrollers were attracted to new leaves significantly more than to mature leaves, and nurse workers were larger than patroller workers, which suggests that the behavioural differences reflect caste differentiation. Therefore, we reject the sensory trap hypothesis. Our results are consistent with the idea that worker attraction to new shoots is the result of selection.

Keywords: Allomerus octoarticulatus; ant symbiont; cheating; coevolution; mutualism; sensory exploitation; signal

Symbioses between ants and plants have long been to house ant colonies (domatia; Bailey 1924; Brouat & considered to be classic examples of coevolution (Janzen McKey 2000; Heil & McKey 2003; Edwards et al. 2006a), 1966). On the plant side, adaptations for hosting ants partner selection mechanisms such as chemical camou- are thought to include hollow swellings and/or pouches ﬂage, wax surfaces and specialized domatia entrances (Fed- erle et al. 1997; Yu & Davidson 1997; Brouat et al. 2001; Yu 2001), food rewards, sometimes apparently specialized Correspondence and present address: D. P. Edwards, Institute of Inte- for certain ant species (Janzen 1966; Rickson 1976; Fiala & grative and Comparative Biology, University of Leeds, Leeds LS2 9JT, Maschwitz 1992; Fischer et al. 2002; Federle & Rheindt U.K. (email: [email protected]). R. Arauco is at the Facultad 2005; Heil et al. 2005), year-round leaf and sucker produc- de Ciencias Naturales y Matematicas, Universidad Nacional Federico tion, even in areas with distinct dry seasons (Janzen 1966), Villarreal, Pueblo Libre, Lima, Peru. M. Hassall is at the School of Environmental Sciences, University of East Anglia, Norwich, Norfolk and reductions in the levels of mechanical and chemical NR4 7TJ, U.K. K. Chamberlain and L. J. Wadhams are at the Biolog- defences (Rehr et al. 1973; Nomura et al. 2000; Eck et al. ical Chemistry Division, Rothamsted Research, Harpenden, Hertford- 2001; but see Heil et al. 2002). On the ant side, adapta- shire AL5 2JQ, U.K. D. W. Yu is at the School of Biological Sciences, tions for the planteant lifestyle are thought to include in- University of East Anglia, Norwich, Norfolk NR4 7TJ, U.K. creased colony size and worker aggressiveness (Janzen 377 0003e3472/07/$30.00/0 Ó 2007 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved. 378 ANIMAL BEHAVIOUR, 74,3

1966), permanent activity outside the nest (Janzen 1966; seeds, including methyl-6-methyl salicylate, which Seidel Fiala & Maschwitz 1990; Moog et al. 1998; Christianini et al. (1990) have hypothesized mimic brood recognition & Machado 2004), traits that allow ants to avoid partner pheromones, thereby eliciting collection by ants. Finally, selection mechanisms (Federle et al. 1997, 2001; Brouat the new leaves of the ant plant Leonardoxa africana et al. 2001), queen orientation to host plant chemicals africana produce higher quantities of methyl salicylate during dispersal and colonization (Inui et al. 2001; than do mature leaves and new leaves of the related non- Edwards et al. 2006b; Ju¨rgens et al. 2006), the pruning myrmecophyte L. africana gracilicaulis (Brouat et al. 2000). of host plant competitors (Janzen 1966, 1969, 1972; Thus, analogous to Seidel et al.’s (1990) suggestion that Davidson et al. 1988; Federle et al. 2002; Frederickson the seeds of ant garden epiphytes mimic ant brood, the et al. 2005), and floral castration (Yu & Pierce 1998; new leaves of ant plants might mimic ant brood and Stanton et al. 1999; Izzo & Vasconcelos 2002; Gaume thereby elicit attraction and patrolling. Alternatively, since et al. 2005). methyl-6-methyl salicylate has been identified as a possi- Seemingly the clearest example of ant adaptation to ble alarm pheromone in Ponerine ants (Duffield & Blum living on plants is the attraction of workers to particular 1975; Longhurst et al. 1980), a sensory trap might act by plant parts (typically, new shoots and/or sites of herbivore eliciting recruitment to a false alarm. However, a priori, damage), resulting in a nonrandom distribution of worker we consider alarm mimicry to be unlikely, since floral patrolling and a reduction in herbivory (Janzen 1966; Da- ant repellents are also thought to be alarm mimics (Will- vidson & McKey 1993; Agrawal 1998; Gaume & McKey mer & Stone 1997; Ghazoul 2001; Raine et al. 2002) and 1999; Heil & McKey 2003; Edwards et al. 2006a; but see cause workers to retreat (Wilson & Regnier 1971). Janzen 1975; Gaume & McKey 1999). However, there is In summary, we have two explanations for the attrac- little formal evidence that patrolling is an adaptation to tion of ant workers to particular plant parts. Either the planteant lifestyle, in the sense that patrolling is a trait attraction is an adaptation that directs patrolling to that arose because of a history of positive selection for pa- vulnerable plant parts, to maximize the ant’s plant-de- trolling per se (West-Eberhard 1992). An obstacle to show- rived rewards, or attraction is elicited by a sensory trap. We ing adaptation is that many anteplant associations appear aimed to distinguish between these two hypotheses by to have arisen by repeated de novo colonization or lineage testing whether leaf-patrolling workers are a distinct switching, rather than through co-cladogenesis of two in- worker caste. If a worker caste is attracted to the odours teracting lineages (Davidson & McKey 1993; Yu & David- of a vulnerable plant part (new leaves) but not to the son 1997; Quek et al. 2004), complicating a phylogenetic odours mimicked by a hypothesized sensory trap (ant approach. brood), then we can reject the sensory trap hypothesis, A plausible alternative explanation for worker attraction and the result would bolster the interpretation that ant to particular plant parts is that workers are attracted to patrolling is an adaptation. false signals emitted by the host plant, in the form of We used as our model system the ant plant Cordia no- a ‘sensory trap’. Sensory traps are signal mimics that dosa Lam. (Boraginaceae), an understorey tree found exploit the adaptive, neural responses of signal receivers across the Amazon basin (Wheeler 1942; Miller 1985). Cor- to elicit out-of-context behaviours. Traps have long been dia nodosa is inhabited by the specialist ants Allomerus cf. invoked in the sexual selection and predatoreprey litera- octoarticulatus var. demerarae Wheeler (Myrmicinae, A. de- ture (Christy 1995; Haynes & Yeargan 1999) to explain the merarae in previous papers) and four species of Azteca Forel evolution of signals that attract mates and prey, such as in (Dolichoderinae), one colony per plant (Yu et al. 2001, the classic flash mimicry by predatory Photuris fireflies that 2004). The plant provides its ant symbionts with domatia attracts male Photinus fireflies (Lloyd 1986). Sensory traps (Edwards et al. 2006a) and food bodies (Solano et al. have also been implicated in mutualisms, particularly in 2005), and workers of all the ant species intensively patrol pollination (Borgkarlson et al. 1985, 1987; Hughes et al. their host plants’ vulnerable new shoots and leaves, but 1994; Schiestl et al. 1999; Schiestl 2004; see also Schaefer not mature leaves, preventing herbivore damage (Yu & et al. 2004 for a review of communication in mutualisms). Pierce 1998; Edwards et al. 2006a). A likely example of a sensory trap is found in the We concentrated on Allomerus, which is the most abun- mutualism between Acacia trees and Pseudomyrmex ants, dant of the symbionts (Yu et al. 2001, 2004). We were un- where Pseudomyrmex workers might benefit by preying able to test Azteca workers because they responded poorly on pollinators, but at the cost of reducing the pollination to the hexane solvent used. However, we have observed success of its host plants; the Acacia flowers produce ant that both Azteca and Allomerus workers rapidly collect repellents, which appear to be chemical mimics of ant the brood of the other ant genus (D. Yu, unpublished alarm pheromones (Willmer & Stone 1997; Raine et al. data), possibly for food, leaving open the possibility that 2002; see also Ghazoul 2001). In another example, some a brood sensory trap could act across ant genera. South American ant genera construct arboreal carton nests To test for caste differentiation, we divided Allomerus (ant gardens), and workers retrieve and deposit the seeds workers into two operational groups, ‘nurses’ and ‘patrol- of specialized epiphyte species in the nest walls (Davidson lers’. Nurses were collected from within domatia, where 1988), which germinate and protect the nest against rain the brood are housed, and were presumed to be engaged (Yu 1994). Because the plants disperse separately from in brood care and feeding (Fig. 1a). Patrollers were col- the ant queens, ant workers seeding new gardens must lected directly from the surfaces of new shoots (Fig. 1b). identify the seeds of the correct epiphyte species. Workers By definition, the brood sensory trap explanation assumes appear to be attracted to chemical compounds on these the existence of just one caste, nurse ants, which are EDWARDS ET AL.: PROTECTION BY ANT SYMBIONTS 379

2100 mm), with large-scale river meandering. The study was conducted with permits from the Instituto de Recur- sos Naturales (INRENA), Peru. For all tests, ant workers were collected only from adult colonies as indexed by plant size, where the minimum number of domatia was 11. Colonies begin to reproduce at plant sizes of ﬁve domatia (Yu & Pierce 1998).

Behavioural Assays

We used a petri dish behavioural assay protocol, follow- ing Brouat et al. (2000). Two assays were conducted: the ﬁrst between wet hexane extracts of new leaves and ant brood, and the second between hexane elutions of new- leaf and mature-leaf volatiles, taken using a headspace technique. Both assays used pure hexane as a control.

New leaves versus ant brood The wet extracts of Allomerus brood were made from 1000þ eggs, larvae and pupae collected from the mature domatia of eight colonies. We used brood from several colonies to test whether Allomerus ants are attracted to gen- eral brood chemicals, because a sensory trap could not apply if ants respond only to brood from their own colony. Hexane extracts of new leaves were made from new shoots (immature leaves under 5 cm long) collected from six Allomerus-inhabited C. nodosa. For both extracts, brood or leaves were placed in a glass beaker cleaned with hexane and soaked in 50 ml of n-hexane (analytical grade) for 30 min. Wet extracts thus contained both volatile and surface chemicals, which do not control for the possi- Figure 1. (a) Nurse workers and brood in a mature domatium. (b) bility that ants are attracted to surface chemicals from mi- Patroller workers on a new shoot, including new domatium and flo- croscopic food bodies that are present on new leaves ral buds. Not to the same scale. (Solano et al. 2005). In contrast, headspace collection cap- tures only volatile chemicals but takes 24 h and is therefore not a suitable methodology for ant brood, which ‘deceived’ into committing the out-of-context behaviour desiccate after removal from the domatia. of tending new leaves. We first used behavioural assays to test for differential attraction of workers to plant and brood extracts, and New leaves versus mature leaves then we checked for size differentiation. The sensory trap Hexane elutions of new-leaf and mature-leaf volatiles hypothesis was therefore a null hypothesis, with three were obtained with a headspace entrainment methodol- subsidiary null hypotheses: (1) both nurses and patrollers ogy as follows. An air pump (Sensidyne AA120CNSN, will be positively and approximately equally attracted to Sensidyne, Clearwater, FL, U.S.A., with minimum free- the odours of brood and new leaves, relative to a neutral flow output of 2015 ml/min) powered by a 12VDC 7A SLA odour control; (2) both nurses and patrollers will be more battery pushed air into the system. Air was passed into attracted to the odours of new leaves than to a neutral a glass U-tube containing activated charcoal to remove odour control or to the odour of mature leaves, which are any volatiles and then through Teflon tubing into a bast- rarely patrolled (Yu & Pierce 1998; Edwards et al. 2006a); ing bag surrounding the test plant part. A shoot consists of (3) both nurses and patrollers are of the same size. one domatium and six attached leaves: four leaves from the domatium and two on the stem (Yu & Pierce 1998). Headspace was taken from the large central leaf of the METHODS four domatium leaves; care was taken to ensure that the leaf was not damaged during the positioning of the bag. We carried out the study at the Tambopata Jungle Another Teflon tube leaving the bag was connected to Lodge (Yu & Davidson 1997), in the buffer zone of the a volatile collection trap (VCT; ARS Inc., P/N number BahuajaeSonene National Park (formerly the Tambopatae VCT-1/4X3-SPQ, ARS Inc., Gainesville, FL, U.S.A.), which Candamo Reserve Zone), Madre de Dios, Peru (69W was fed via plastic tubing into a flowmeter (Fisher Scien- 12S; 200e500 m above sea level), between April 2002 tific FJC-625-035V, Fisher Scientific, Loughborough, U.K.) and May 2003. The characteristic habitat of the study set at 0.3 litres/min. The bag was sealed to the plant and area is moist to seasonal tropical forest (annual rainfall air tubes with plastic cable ties. Air was pulled through 380 ANIMAL BEHAVIOUR, 74,3

the VCT and flowmeter with a second air pump. Each en- we calculated from the relative areas of the filter papers trainment was run for 24 h, after which the filter was (7.1 cm2 each) and of the petri dish minus the area of eluted with a micropipette into a 2-ml Wheaton vial three filter papers (42.4 cm2), and with an ant population with 600 ml of analytical grade n-hexane. Hexane elutions of 30 individuals, that a mean of 3.3 ants would be found were stored in a freezer, and eluted VCTs were cleaned on each filter paper, leaving 20 ants on the remaining with a further 2 ml of n-hexane. open surface. We used these numbers as a baseline expec- For gas chromatography analysis (see Appendix for de- tation of ant distribution. tails), an additional wet extract of Allomerus brood from five colonies was made as described above, but with distilled ether. Using the headspace entrainment methodol- Morphology ogy described above, we also took three new-leaf, three We measured the head width between the linear mature-leaf and three empty-bag control VCTs and sealed margins of the compound eyes (intraocular width), and them in situ in glass tubes. The filters were then eluted the alitrunk length between the anterior margin of the with 600 ml of distilled ether in the laboratory. Distilled pronotum and the posterior margin of the propodeum, of ether was used in preference to hexane because ether gives 30 nurse and 30 patroller workers (one ant from each of 30 cleaner gas chromatography traces. adult colonies) using a 16 stereomicroscope with an Patrollers from a given colony were collected from eyepiece graticule. a new shoot (N ¼ 57 colonies) and nurses from within three mature domatia (N ¼ 94). All eggs were removed from the nurses. Ants from a particular colony were tested Statistical Analysis only once for each bioassay type and were not tested against extracts obtained from their own host plants. Because sequential worker counts in a single bioassay We conducted behavioural assays in petri dishes, 90 mm constitute nonindependent repeated measures, we con- in diameter, with Fluon liquid or PTFE spray applied to the ducted repeated measures ANOVAs from the fifth to the sides to prevent ants from escaping. For each test, 30 ants 10th minute using SPSS version 12.0 (SPSS Inc., Chicago, of a single group (patroller or nurse) were placed in a petri IL, U.S.A.). Repeated measures ANOVA assumes linearity dish. Then each of three filter paper circles, 30 mm in di- of the response variable; we therefore omitted counts from ameter and labelled with pencil, received 100 ml of the ap- the first 4 min, after which the time courses of ant re- propriate extract, held in the air with metal forceps for sponse were roughly linear. In all cases, residuals were nor- 15 s to allow some of the hexane to volatilize, and then mally distributed and homoscedastic. We used a contrast placed in the petri dish. One of the three extracts was al- analysis to compare the new-leaf treatment extract to ways a pure hexane control, and the other two extracts the other treatment extract (brood or mature-leaf) and to are described below. Observations began when the last fil- the hexane control. ter paper was entered, and at this point the temperature was recorded from a thermometer positioned next to the petri dish. Each minute for 10 min, we scored the number RESULTS of ants on, under or with their heads touching each filter paper. Each assay was conducted with a new petri dish and Gas Chromatography Analyses filter papers, and we controlled for temporal order and Gas chromatography analyses verified the presence of spatial orientation. For consecutive bioassays of the same chemical compounds in the samples that were absent in experiment, extracts were introduced in revolving order; the pure hexane control (see Appendix: Fig. A1). Also, the the first entered in one experiment was then entered sec- extracts of new leaves and of Allomerus brood contained ond, the second entered third, etc., with their positions some peaks with the same retention times, which suggests also being rotated consecutively around the dish. that some compounds might be shared. Furthermore, For the new-leaf versus brood bioassay, we conducted 30 those peaks were not present in the extract of mature patroller and 60 nurse bioassays, and for the new-leaf leaves. These results are consistent with the possibility of versus mature-leaf bioassay, we conducted 37 patroller and a brood sensory trap eliciting patrolling of new leaves. 66 nurse bioassays. Bioassays were conducted between temperatures of 23 and 34C for both the new-leaf versus brood experiment with patroller (X SE ¼ 29:5 0:2 C) New Leaves Versus Ant Brood and nurse (30.1 0.3C) ants (ManneWhitney U test: U ¼ 805.0, N1 ¼ 30, N2 ¼ 59, P ¼ 0.48), and for new-leaf Patrollers were significantly more attracted to new-leaf versus mature-leaf experiments with patroller (29.2 extract than to brood extract, and there was no significant 0.5C) and nurse (29.1 0.3C) ants (U ¼ 1185.5, difference between brood extract and the hexane control N1 ¼ 37, N2 ¼ 66, P ¼ 0.80). We also conducted control as- (Fig. 2a, Table 1). In contrast, nurses were significantly says, with 30 workers each, using three filter papers all more attracted to brood extract than to new-leaf extract, with pure hexane. These revealed no significant clustering and they were also significantly more attracted to brood of workers (repeated measures ANOVA: patrollers: extract than to the hexane control (Fig. 2b, Table 1). There F2,12 ¼ 0.0, P ¼ 0.97; nurses: F2,12 ¼ 0.1, P ¼ 0.88). was a significant interaction effect between worker group Assuming that ants do not respond to the samples but (nurse versus patroller) and extract type (brood versus instead distribute themselves randomly in the petri dish, new-leaf only; Table 1), confirming that the two worker EDWARDS ET AL.: PROTECTION BY ANT SYMBIONTS 381

8 6 (a) (a) 7 5 6 ∗

5 ∗∗ 4 4

3 3

2 12345678910 2 12345 678910 8 (b) 6 7 (b) Mean number of ants

6 Mean number of ants 5 5

4 ∗∗ 4

3 3 2 12345678910 Time (min) 2 Figure 2. Mean number of worker ants attracted to filter papers con- 1234 5 678910 taining new-leaf (,), brood (>) and pure hexane (6) extracts each Time (min) minute from the first to the 10th minute of the bioassay. (a) Patroller Figure 3. Mean number of worker ants attracted to filter papers con- bioassays and (b) nurse bioassays. Bars represent 1 SE. The dashed taining new-leaf (,), mature-leaf (B), and pure hexane (6) extracts line represents the null expected density of workers. Repeated mea- each minute from the first to the 10th minute of the bioassay. (a) Pa- sures ANOVAs were conducted on worker counts from the fifth to troller bioassays and (b) nurse bioassays. Bars represent 1SE.The the 10th minute. **P < 0.01, new leaf versus brood contrast analysis. dashed line represents the null expected density of workers. Repeated measures ANOVAs were conducted on worker counts from the fifth to types responded differently to the extracts. We therefore the 10th minute. *P < 0.05, new leaf versus mature leaf contrast have to reject the first prediction that nurses and patrollers analysis. should be positively and approximately equally attracted to the odours of brood and new leaves. the control (Fig. 3a, Table 2). Given that the mature-leaf response was below the baseline of 3.33 ants (Fig. 3a), one interpretation is that mature leaves might emit an ant repellent. However, the null number of ants expected New Leaves Versus Mature Leaves on the mature-leaf filter paper (3.33) is not significantly different from the observed mean of 2.83 0.27 (paired Patrollers were significantly more attracted to new-leaf t test: t 1.9, P ¼ 0.07). extract than to mature-leaf extract, and there was no 35 Nurse ants were positively attracted to all three extract significant difference between the new-leaf extract and types, with no significant difference between them (Fig. 3b, Table 2), despite a consistent trend towards Table 1. Responses of worker ants in new-leaf versus brood bioassays Table 2. Responses of worker ants in new-leaf versus mature-leaf Comparison NdfF P bioassays Comparison NdfF P Patroller 30 2 7.9 0.001 New-leaf versus brood 0.003 New-leaf versus control <0.001 Patroller 37 2 3.2 0.044 Nurse 60 2 11.0 <0.001 New-leaf versus mature-leaf 0.040 New-leaf versus brood 0.007 New-leaf versus control 0.82 New-leaf versus control 0.055 Nurse 66 2 0.4 0.68 Castesample 90 1 14.1 <0.001 Castesample 103 1 1.2 0.27 (new-leaf and brood only) (new-leaf and mature-leaf only) 382 ANIMAL BEHAVIOUR, 74,3

a slightly higher attraction to new-leaf extract than to the patrollers, which permanently patrol new leaves (Edwards mature-leaf extract and the control. et al. 2006a), the Crematogaster ‘defender’ caste is reported The interaction effect between worker type and extract, to recruit to new leaves only when the host plant and thus again without the hexane control, was not significant the colony is disturbed (Stapley 1999). (Table 2). In sum, patrollers were significantly more at- Caste differentiation is consistent with the alternative tracted to new-leaf extract than mature-leaf extract but hypothesis that the attraction of workers to new leaves is nurses were not. However, the interaction effect was not the result of selection on Allomerus to direct patrolling to significant because of high nurse attraction to all three fil- the plant parts that maximize the capture of rewards ter papers, nor are we able to reject the null expectation (Agrawal & Rutter 1998). Allomerus patrollers are unable that either of the leaf extracts is more attractive than the to capture live insects on leaves, instead constructing car- neutral odour control. Thus, we weakly reject the sensory ton traps on plant stems, which allows several workers to trap expectation that the two worker types should both be attack insect prey simultaneously (Dejean et al. 2005; attracted to new leaves over mature leaves. D. Edwards & D. Yu, personal observations), and the high density of patrollers on new leaves (Edwards et al. 2006a) deters oviposition by insects. Therefore, any re- Size Differences wards obtained are likely to be plant derived. Solano et al. (2005) reported the existence of microscopic food The intraocular width and alitrunk length of nurse bodies on this ant plant, and they suggested that the den- workers were significantly larger than those of patroller sity of these food bodies is higher on new leaves. Edwards workers (GLM: head: F ¼ 32.4, P < 0.001; alitrunk: 1,58 et al. (2006a) have shown that the modular structure of F ¼ 33.1, P < 0.001; Fig. 4). 1,58 C. nodosa, such that each new shoot combines leaves and a domatium (Fig. 1b), ties the provision of housing to the DISCUSSION successful growth of new leaves, and thus to patrolling. The existence of mutualistic behaviour at the vegetative We tested the hypothesis that a brood sensory trap pro- level is all the more striking given that Allomerus is a castra- duced by the new leaves of C. nodosa attracts workers of the tion parasite (Yu & Pierce 1998). Allomerus’s castration symbiotic ant A. octoarticulatus var. demerarae. Gas chroma- behaviour is thought to have evolved in response to com- tography analyses suggested shared compounds between petition with Azteca spp. ants for host plants (Yu et al. new leaves and Allomerus brood (Appendix), and nurse 2001, 2004), presumably after the evolution of the symbi- workers were attracted to brood extracts of nonself colo- osis and thus of leaf-patrolling behaviour. Castration by nies (Fig. 2b), suggesting that Allomerus brood emit a gen- Allomerus underlines the principle that hosts can coerce eral attractant, at least within species (Ho¨lldobler 1977). cooperative behaviour in their symbionts only when These results are consistent with a brood sensory trap. symbiont behaviour can be monitored (see Edwards et al. However, we rejected three null hypotheses predicted 2006a and references therein). by a brood sensory trap. Patrollers were attracted to new leaves but not to brood (and vice versa for nurses), only patrollers were attracted significantly more to new leaves Is There Another Sensory Trap? than to mature leaves, and nurses were significantly larger than patrollers. Thus, patrollers and nurses appeared to Methyl salicylate, which was found in large quantities in belong to different worker castes, and we must reject the the new leaves of Leonardoxa africana africana (Brouat et al. hypothesis that a brood sensory trap causes Allomerus 2000), is a herbivore-induced volatile that is released workers to be attracted to new leaves. almost immediately upon leaf damage (Pichersky & Caste differentiation in the context of an ant plant Gershenzon 2002), and attracts insect predators and para- mutualism has been identified in the ant Crematogaster sites (Dicke et al. 1990a, b; van Poecke et al. 2001; James nigriceps, which is a protection mutualist of Acacia drepano- 2003). Ants in Cecropia ant plants are known to respond lobium (Stapley 1999). However, in contrast to Allomerus to leaf damage with immediate recruitment (Agrawal & Dublin-Thaler 1999; Romero & Izzo 2004) and longer- 0.8 term patrolling (up to 24 h; Agrawal 1998). Therefore, the permanent release of methyl salicylate and/or other ∗∗∗ green leaf volatiles, even in the absence of leaf damage, 0.7 might be considered to be a form of ‘crying wolf’ that ∗∗∗ elicits constant patrolling on new leaves. However, such 0.6 damage mimicry cannot be considered to be a sensory trap because for such mimicry to select for patrolling, there must also be a plant-provided benefit, such as housing, Mean size (mm) 0.5 that is potentially threatened if the leaves are eaten (i.e. there must also be sheep if there is to be selection to defend 0.4 against wolves). In contrast, a sensory trap, by definition, Head Alitrunk does not rely on a benefit. Figure 4. Mean intraocular width and alitrunk length SE of nurses Alternatively, a sensory trap that mimics ant alarm (-) and patrollers (,). *P < 0.05. pheromone might have elicited new-leaf patrolling. EDWARDS ET AL.: PROTECTION BY ANT SYMBIONTS 383

However,weobservedthatAllomerusworkerspatrollingnew Agrawal, A. A. & Rutter, M. T. 1998. Dynamic anti-herbivore defense leaves assumed alarm stances (open mandibles and raised in ant-plants: the role of induced responses. Oikos, 83, 227e236. heads) only when approached up close by the investigator Bailey, I. W. 1924. Notes on neotropical ant-plants. III. Cordia but not when the investigator was far away and observing nodosa Lam. Botanical Gazette, 77,39e49. through binoculars (D. Edwards, personal observation). Benson, W. W. 1985. Amazon ant-plants. In: Amazonia (Ed. by G. T. e Combined with our initial reservations, this leads us also to Prance & T. E. Lovejoy), pp. 239 266. Oxford: Pergamon. reject the possibility of an alarm sensory trap for this system. Borgkarlson, A. K., Bergstrom, G. & Groth, I. 1985. Chemical basis for the relationship between Ophrys orchids and their pollinators. 1. Volatile compounds of Ophrys lutea and O. fusca as insect mi- e An Origin Role for a Sensory Trap? metic attractants/excitants. Chemical Scripta, 25, 283 294. Borgkarlson, A. K., Bergstrom, G. & Kullenberg, B. 1987. Chemical Notwithstanding our rejection of sensory traps as basis for the relationship between Ophrys orchids and their pollina- a current explanation for ant attraction, the observed tors. 2. Volatile compounds of O. insectifera and O. speculum as insect mimetic attractants/excitants. Chemical Scripta, 27, 303e311. weakly positive attraction of nurse ants to new leaves Brouat, C. & McKey, D. 2000. Origin of caulinary ant domatia and (Fig. 2b, Table 1) and the possibility of shared compounds timing of their onset in plant ontogeny: evolution of a key trait in Allomerus between new leaves and brood (Appendix) lead horizontally transmitted anteplant symbioses. Biological Journal of us to speculate that a brood sensory trap might have acted the Linnean Society, 71, 801e819. during the origin of this ant plant protection mutualism, Brouat, C., McKey, D., Bessiere, J. M., Pascal, L. & Hossaert- when there was no direct benefit of new-leaf protection McKey, M. 2000. Leaf volatile compounds and the distribution to the ants, subsequently declining to a remnant form of ant patrolling in an anteplant protection mutualism: prelimi- (see Macıás-Garcia & Ramirez 2005 for an example of nary results on Leonardoxa (Fabaceae: Caesalpinioideae) and Petal- a sensory trap evolving into a reliable signal of mate qual- omyrmex (Formicidae: Formicinae). Acta Oecologica, 21, 349e357. ity). First, ant brood mimicry by new leaves could have Brouat, C., Garcia, N., Andary, C. & McKey, D. 2001. Plant lock initiated a recurring association of nonsymbiotic Allome- and ant key: pairwise coevolution of an exclusion filter in an e rus workers with a premyrmecophytic C. nodosa, and the ant plant mutualism. Proceedings of the Royal Society of London, e regular patrolling, which would have initially benefited Series B, 268, 2131 2141. only the plant, could have provided a selection pressure Christianini, A. V. & Machado, G. 2004. Induced biotic responses to herbivory and associated cues in the Amazonian anteplant Maieta to evolve domatia and thus the symbiosis (see also Benson poeppigii. Entomologia Experimentalis et Applicata, 112,81e88. 1985). Second, sensory traps could have helped to reduce Christy, J. H. 1995. Mimicry, mate choice, and the sensory trap error in the application of host sanctions (Edwards et al. hypothesis. American Naturalist, 146, 171e181. e 2006a). The leaves of a new shoot grow for 2 3 weeks be- Davidson, D. W. 1988. Ecological-studies of neotropical ant gar- fore the associated domatium can be used to house ant dens. Ecology, 69, 1138e1152. brood (D. Edwards, unpublished data). Therefore, at least Davidson, D. W. & McKey, D. 1993. Anteplant symbioses: stalking early in the evolution of patrolling, before separate castes the Chuyachaqui. Trends in Ecology and Evolution, 8, 326e332. evolved, mimicry of ant brood could have helped to en- Davidson, D. W., Longino, J. T. & Snelling, R. R. 1988. Pruning of sure the regular protection of new leaves during the pro- host plant neighbors by ants: an experimental approach. Ecology, longed maturation of domatia, avoiding the incorrect 69, 801e808. application of a host sanction. Dejean, A., Solano, P. J., Ayroles, J., Corbara, B. & Orivel, J. 2005. Insect behaviour: arboreal ants build traps to capture prey. Nature, 434, 973. Acknowledgments Dicke, M., Sabelis, M. W., Takabayashi, J., Bruin, J. & Posthumus, M. A. 1990a. Plant strategies of manipulating predatoreprey Special thanks to our assistants Pedro Cauper-Vallez and interactions though allelochemicals: prospects for application in Arturo Balareso for their tireless help during data collec- pest control. Journal of Chemical Ecology, 16, 3091e3118. tion. We also thank Niclas Kolm and two anonymous Dicke, M., van Beek, T. A., Posthumus, M. A., Ben Dom, N., van referees for insightful comments, Paul Dennis for assis- Bokhoven, H. & de Groot, A. E. 1990b. Isolation and identifica- tance with gas chromatography analysis, Thomas tion of volatile kairomone that affects acarine predatoreprey inter- Hendrickson and the Tambopata Jungle Lodge for logistical actions. Journal of Chemical Ecology, 16, 381e394. assistance and permission to conduct the study on their Duffield, R. M. & Blum, M. S. 1975. Methyl 6-methyl salicylate: land, Peru’s Instituto de Recursos Naturales (INRENA) for identification and function in a ponerine ant. Experientia, 31, 466. research permits and the National Environmental Research Eck, G., Fiala, B., Linsenmair, K. E., Bin Hashim, R. & Proksch, P. Council (NERC) for a studentship awarded to D.P.E. 2001. Trade-off between chemical and biotic antiherbivore defense in the south-east Asian plant genus Macaranga. Journal of Chemical Ecology, 27, 1979e1996. References Edwards, D. P., Hassall, M., Sutherland, W. J. & Yu, D. W. 2006a. Selection for protection in anteplant mutualism: host sanctions, Agrawal, A. A. 1998. Leaf damage and associated cues induce host modularity, and the principal-agent game. Proceedings of aggressive ant recruitment in a neotropical ant-plant. Ecology, the Royal Society of London, Series B, 273, 595e602. 79, 2100e2112. Edwards, D. P., Hassall, M., Sutherland, W. J. & Yu, D. W. 2006b. Agrawal, A. A. & Dublin-Thaler, B. J. 1999. Induced responses to Assembling a mutualism: ant symbionts locate their host plants by herbivory in the neotropical anteplant association between Azteca detecting volatile chemicals. Insectes Sociaux, 53, 172e176. ants and Cecropia trees: response of ants to potential inducing Federle, W. & Rheindt, F. E. 2005. Macaranga ant-plants hide food cues. Behavioral Ecology and Sociobiology, 45,47e54. from intruders: correlation of food presentation and presence of 384 ANIMAL BEHAVIOUR, 74,3

wax barriers analysed using phylogenetically independent con- Janzen, D. H. 1969. Allelopathy by myrmecophytes: ant Azteca as an trasts. Botanical Journal of the Linnean Society, 84, 177e193. allelopathic agent of Cecropia. Ecology, 50, 147e153. Federle, W., Maschwitz, U., Fiala, B., Riederer, M. & Ho¨lldobler, Janzen, D. H. 1972. Protection of Barteria (Passifloraceae) by Pachy- B. 1997. Slippery ant-plants and skilful climbers: selection and pro- sima ants (Pseudomyrmecinae) in a Nigerian rain-forest. Ecology, tection of specific ant partners by epicuticular wax blooms in Mac- 53, 885e892. aranga (Euphorbiaceae). Oecologia, 112, 217e224. Janzen, D. H. 1975. Pseudomyrmex nigropilosa: parasite of a mutual- Federle, W., Fiala, B., Zizka, G. & Maschwitz, U. 2001. Incident ism. Science, 188, 936e937. daylight as orientation cue for hole-boring ants: prostomata in Ju¨rgens, A., Feldhaar, H., Feldmeyer, B. & Fiala, B. 2006. Chemical Macaranga ant-plants. Insectes Sociaux, 48, 165e177. composition of leaf volatiles in Macaranga species (Euphorbiaceae) Federle, W., Maschwitz, U. & Ho¨lldobler, B. 2002. Pruning of host and their potential role as olfactory cues in host-localization of plant neighbours as defence against enemy ant invasions: Crema- foundress queens of specific ant partners. Biochemical Systematics togaster ant partners of Macaranga protected by ‘wax barriers’ and Ecology, 34,97e113. prune less than their congeners. Oecologia, 132, 264e270. Lloyd, J. E. 1986. Firefly communication and deception: ‘oh, what Fiala, B. & Maschwitz, U. 1990. Studies on the South-east Asian a tangled web’. In: Deception: Perspectives on Human and Nonhu- anteplant association Crematogaster borneensis-Macaranga: adap- man Deceit (Ed. by R. W. Mitchell & N. S. Thompson), pp. 113e tations of the ant partner. Insectes Sociaux, 37, 212e231. 128. Albany: State University of New York Press. Fiala, B. & Maschwitz, U. 1992. Food bodies and their significance Longhurst, C., Baker, R. & Howse, P. E. 1980. A multicomponent for obligate ant-association in the tree genus Macaranga (Euphor- mandibular gland secretion in the ponerine ant Bothroponera soror biaceae). Botanical Journal of the Linnean Society, 110,61e75. (Emery). Journal of Insect Physiology, 26, 551e555. Fischer, R. C., Richter, A., Wanek, W. & Mayer, V. 2002. Plants feed Macıás-Garcia, C. & Ramirez, E. 2005. Evidence that sensory traps ants: food bodies of myrmecophytic Piper and their significance for can evolve into honest signals. Nature, 434, 501e505. the interaction with Pheidole bicornis ants. Oecologia, 133, 186e192. Miller, J. S. 1985. Systematics of the genus Cordia (Boraginaceae) in Frederickson, M. E., Greene, M. J. & Gordon, D. M. 2005. ‘Devil’s Mexico and Central America. Ph.D. thesis, University of Michigan. gardens’ bedevilled by ants. Nature, 437, 495e496. Moog, J., Drude, T. & Maschwitz, U. 1998. Protective function of Gaume, L. & McKey, D. 1999. An anteplant mutualism and its host- the plant-ant Cladomyrma maschwitzi to its host, Crypteronia griffi- specific parasite: activity rhythms, young leaf patrolling, and thii, and the dissolution of the mutualism (Hymenoptera: Formici- effects on herbivores of two specialist plant-ants inhabiting the dae). Sociobiology, 31, 105e129. same myrmecophyte. Oikos, 84, 130e144. Nomura, M., Itioka, T. & Itino, T. 2000. Variations in abiotic defense Gaume, L., Zacharias, M. & Borges, R. M. 2005. Anteplant con- within myrmecophytic and non-myrmecophytic species of Macar- flicts and a novel case of castration parasitism in a myrmecophyte. anga in a Bornean dipterocarp forest. Ecological Research, 15,1e11. Evolutionary Ecology Research, 7, 435e452. Pichersky, E. & Gershenzon, J. 2002. The formation and function of Ghazoul, J. 2001. Can floral repellents pre-empt potential anteplant plant volatiles: perfumes for pollinator attraction and defense. Cur- conflicts? Ecology Letters, 4, 295e299. rent Opinion in Plant Biology, 5, 237e243. Haynes, K. F. & Yeargan, K. V. 1999. Exploitation of intraspecific van Poecke, R. M. P., Posthumus, M. A. & Dicke, M. 2001. Herbi- communication systems: illicit signalers and receivers. Annals of vore-induced volatile production by Arabidopsis thaliana leads to the Entomological Society of America, 92, 960e970. attraction of the parasitoid Cotesia rubecula: chemical, behaviou- Heil, M. & McKey, D. 2003. Protective anteplant interactions as ral, and gene-expression analysis. Journal of Chemical Ecology, 27, model systems in ecological and evolutionary research. Annual 1911e1928. Review of Ecological Evolution and Systematics, 34, 425e453. Quek, S. P., Davies, S. J., Itino, T. & Pierce, N. E. 2004. Codiversi- Heil, M., Delsinne, T., Hilpert, A., Schurkens, S., Andary, C., fication in an ant-plant mutualism: stem texture and the evolution Linsenmair, K. E., Sousa, M. S. & McKey, D. 2002. Reduced of host use in Crematogaster (Formicidae: Myrmicinae) inhabitants chemical defence in ant-plants? A critical re-evaluation of a widely of Macaranga (Euphorbiaceae). Evolution, 58, 554e570. accepted hypothesis. Oikos, 99, 457e468. Raine, N. E., Willmer, P. & Stone, G. N. 2002. Spatial structuring Heil, M., Rattke, J. & Boland, W. 2005. Postsecretory hydrolysis of and floral avoidance behavior prevent ant-pollinator conflict in nectar sucrose and specialization in ant/plant mutualism. Science, a Mexican ant-acacia. Ecology, 83, 3086e3096. 308, 560e563. Rehr, S. S., Feeny, P. P. & Janzen, D. H. 1973. Chemical defence in Ho¨lldobler, B. 1977. Communication in social Hymenoptera. In: Central American non-ant-acacias. Journal of Animal Ecology, 42, How Animals Communicate (Ed. by T. A. Sebeok), pp. 418e471. 405e416. Bloomington, Indiana: Indiana University Press. Rickson, F. R. 1976. Anatomical development of leaf trichilium and Hughes, L., Westoby, M. & Jurado, E. 1994. Convergence of elaio- mullerian bodies of Cecropia peltata L. American Journal of Botany, somes and insect prey: evidence from ant foraging behaviour and 63, 1266e1271. fatty acid composition. Functional Ecology, 8, 358e365. Romero, G. Q. & Izzo, T. J. 2004. Leaf damage induces ant recruit- Inui, Y., Itioka, T., Murase, K., Yamaoka, R. & Itino, T. 2001. ment in the Amazonian ant-plant Hirtella myrmecophila. Journal of Chemical recognition of partner plant species by foundress ant Tropical Ecology, 20, 675e682. queens in Macaranga-Crematogaster myrmecophytism. Journal of Schaefer, H. M., Schaefer, V. & Levey, D. J. 2004. How plante Chemical Ecology, 27, 2029e2040. animal interactions signal new insights in communication. Trends Izzo, T. J. & Vasconcelos, H. L. 2002. Cheating the cheater: doma- in Ecology and Evolution, 19, 577e584. tia loss minimizes the effects of ant castration in an Amazonian Schiestl, F. P. 2004. Floral evolution and pollinator mate choice in a ant-plant. Oecologia, 133, 200e205. sexually deceptive orchid. Journal of Evolutionay Biology, 17,67e75. James, D. G. 2003. Field evaluation of herbivore-induced plant vol- Schiestl, F. P., Ayasse, M., Paulus, H. F., Lofstedt, C., Hansson, atiles as attractants for beneficial insects: methyl salicylate and the B. S., Ibarra, F. & Francke, W. 1999. Orchid pollination by sexual green lacewing Chrysopa nigricornis. Journal of Chemical Ecology, swindle. Nature, 399, 421e422. 29, 1601e1609. Seidel, J. L., Epstein, W. W. & Davidson, D. W. 1990. Neotropical Janzen, D. H. 1966. Coevolution of mutualism between ants and ant gardens I. Chemical constituents. Journal of Chemical Ecology, acacias in Central America. Evolution, 20, 249e279. 16, 1791e1816. EDWARDS ET AL.: PROTECTION BY ANT SYMBIONTS 385

Solano, P. J., Belin-Depoux, M. & Dejean, A. 2005. Formation and Yu, D. W. & Pierce, N. E. 1998. A castration parasite of an ant-plant structure of food bodies in Cordia nodosa (Boraginaceae). Comptes mutualism. Proceedings of the Royal Society of London, Series B, 265, Rendus Biologies, 328, 642e647. 375e382. Stanton, M. L., Palmer, T. M., Young, T. P., Evans, A. & Turner, Yu, D. W., Wilson, H. B. & Pierce, N. E. 2001. An empirical model M. L. 1999. Sterilization and canopy modification of a swollen of species coexistence in a spatially structured environment. Ecol- thorn acacia tree by a plant-ant. Nature, 401, 578e581. ogy, 82, 1761e1771. Stapley, L. 1999. Physical worker castes in colonies of an acacia-ant Yu, D. W., Wilson, H. B., Frederickson, M. E., Palomino, W., De la (Crematogaster nigriceps) correlated with an intra-colonial division Colina, R., Edwards, D. P. & Balareso, A. A. 2004. Experimental of defensive behaviour. Insectes Sociaux, 46, 146e149. demonstration of species coexistence enabled by dispersal limita- West-Eberhard, M. J. 1992. Adaptation: current usages. In: Key- tion. Journal of Animal Ecology, 73, 1102e1114. words in Evolutionary Biology (Ed. by E. F. Keller & E. A. Lloyd), pp. 13e18. Cambridge, Massachusetts: Harvard University Press. Wheeler, W. M. 1942. Studies of Neotropical Ant-plants and Their Appendix Ants. Cambridge, Massachusetts: Museum of Comparative Zool- Gas chromatography (GC) analysis was conducted in ogy, Harvard College. a 5890GC Hewlett Packard Series II fitted with a cool-on- Willmer, P. G. & Stone, G. N. 1997. How aggressive ant-guards column injector and a flame ionization detector. Start assist seed-set in Acacia flowers. Nature, 388, 165e167. temperature was 40C for an initial time of 1 min. The Wilson, E. O. & Regnier, F. E. 1971. Evolution of alarm defense system in Formicine ants. American Naturalist, 105, 279e289. temperature was then increased to 240 Cat10C/min, Yu, D. W. 1994. The structural role of epiphytes in ant gardens. staying at 240 C for 30 min. A HP1 50-m column with Biotropica, 26, 222e226. 0.32-mm internal diameter and film thickness 0.52 mm Yu, D. W. 2001. Parasites of mutualisms. Biological Journal of the was used, with hydrogen as the carrier gas at 3 ml/min. Linnean Society, 72, 529e546. A 10-ml glass syringe, cleaned 20 times with redistilled Yu, D. W. & Davidson, D. W. 1997. Experimental studies of species- hexane, was used to inject 1 ml of test sample into the specificity in Cecropia-ant relationships. Ecological Monographs, 67, GC apparatus. Upon entry of the syringe contents, the 273e294. GC run was started immediately.

12000 (a) 10000 8000 6000 4000 2000 0 9 1011121314151617

4000 (b) 3000 2000 1000 0

Counts 9 1011121314151617

4000 (c) 3000 2000 1000 0 9 1011121314151617

4000 (d) 3000 2000 1000 0 9 1011121314151617 Time (min) Figure A1. Gas chromatography trace of (a) Allomerus brood (wet extract) and of (b) new-leaf, (c) mature-leaf and (d) control headspace entrainments. Arrows highlight peaks that are at the same retention time and, therefore, might represent shared compounds.