Scavenging of dead invertebrates along an urbanisation gradient in Singapore

Abstract

Abstract.  1. Little is known about the animals that scavenge invertebrates in the tropics and the impact of human activities on such organisms.

2. We studied the scavenging process using baits representing five dead invertebrate types in six habitats along an urbanisation gradient in equatorial Singapore: primary forest, old secondary forest, young secondary forest, recreational park, mown grassland and impervious surfaces.

3. Ants were the dominant scavengers, except at night in grassland when an earwig (Labiduridae) dominated. In general, the forest sites had more scavenger species and shorter bait survival times than the non-forest habitats.

4. Bait survival time increased monotonically along the urbanisation gradient, suggesting that this parameter could be used as an indicator of habitat quality.

Introduction

Many animals die from causes other than predation, including parasitism, disease, ageing, dehydration, starvation, or partial consumption by a predator. Consequently, a substantial amount of dead animal biomass is available for scavengers (De Vault et al., 2003). Dead prey do not actively defend themselves, so many vertebrate and invertebrate predators regularly scavenge, but difficulties in quantifying scavenged material in animal diets have limited the number of studies. As a result, scavenging is often viewed as an interesting behaviour rather than an important process in the ecosystem.

Despite the greater amount of biomass produced by dead invertebrates (Seastedt et al., 1981; Fellers & Fellers, 1982), more studies have been done on the scavenging of vertebrate carcasses. The species composition of scavengers and the decomposition rate of the vertebrate carrion are well known from these studies (e.g. Payne & King, 1968; Kostecke et al., 2001; Travis et al., 2004). In contrast, the relatively few studies on the scavenging of invertebrate carcasses have dealt mainly with the ability of various predators to feed on dead prey, or with laboratory-based choice experiments between living and dead prey (e.g. Lang & Gsödl, 2001; Mayntz et al., 2005). Very few studies have considered dead prey as a nutrient source or investigated the exploitation of invertebrate carrion in the field (Seastedt et al., 1981; Fellers & Fellers, 1982; Retana et al., 1991; Bestelmeyer & Wiens, 2003; Yee & Juliano, 2006) and none of these have been in the tropics (Corlett, 2009). Moreover, few of these studies have looked at the impact of anthropogenic disturbances on the scavenging process (Retana et al., 1991; Bestelmeyer & Wiens, 2003).

The major objectives of this study were therefore to (i) identify the major animal taxa responsible for scavenging invertebrates in equatorial Singapore, (ii) compare the abundance and species composition of the scavenging guild between baits of different body types and in habitats with different levels of human disturbance, (iii) characterise the behaviour of the scavengers and (iv) compare the rate at which different baits are removed and/or consumed in the various habitats. Finally, we aimed to determine whether the vital characteristics of the scavenging process in different habitats can be used as bioindicators of habitat quality.

Methods

The study area

Singapore (103°50′E, 1°20′N) is a 700-km² island separated by <1 km from the southern tip of Peninsular Malaysia (Tan et al., 2010). It has a typical equatorial climate with a mean daily temperature of 27.5 °C, a mean annual rainfall of 2346 mm, and no month with a mean rainfall <100 mm (Meteorological Services Division, National Environmental Agency). Sampling was conducted between July 2008 and February 2009.

Progressive deforestation in the nineteenth century reduced forest cover to isolated patches distributed throughout the island (Corlett, 1997). The area under cultivation declined during the twentieth century, allowing the growth of secondary forest. Primary forest remnants remained only in those areas that have received continuous protection, all of which are within the current nature reserves located in the central part of the island. Today, more than half of Singapore’s land area is urbanised, with public parks and recreational grasslands scattered within the urban matrix. Six habitats were chosen along a gradient from little-modified remnants of the original forest cover of Singapore to full urbanisation: (i) primary lowland dipterocarp forest, (ii) old secondary forest (>50 years old), (iii) young secondary forest (≤50 years old), (iv) recreational park (mown grassland with trees and shrubs), (v) grassland (mown grassland without trees) and (vi) artificial impervious surfaces. The primary forest and old secondary forest are near MacRitchie Reservoir in the 2000-ha Central Catchment Nature Reserve, while the young secondary forest is on Kent Ridge, within the National University of Singapore. The recreational park is the 50-ha West Coast Park in southwest Singapore and the open grassland and impervious surfaces are within the University.

Methodology

Five taxa for which a reliable and uniform supply was locally available were used to represent the variety of invertebrate body types: house cricket (Acheta domestica, Gryllidae, Orthoptera), superworm (larvae of Zophobas morio, Tenebrionidae, Coleoptera), tree nymph (Idea leuconoe, Nymphalidae, Lepidoptera) and great eggfly (Hypolimnas bolina, Nymphalidae, Lepidoptera), and earthworm (Pontoscolex corethrurus, Glossoscolecidae, Oligochaeta). The butterflies were imperfect specimens from a butterfly farm. All taxa were killed by freezing and kept frozen until needed. They were then allowed to warm to ambient temperature and weighed just before use.

For each habitat, four replicate sites were established, at least 100 m apart. Twelve bait stations were installed at each site, separated by a minimum distance of 10 m. Each station consisted of the five different bait types placed directly on the ground at least 1 m apart. These separations were chosen to minimise the likelihood that a single colony of ants would discover more than one bait station or a single scavenger would discover more than one bait. The bait was observed every 40 min for 200 min after placement and then after 1 day. The bait was considered as successfully scavenged when it was moved >10 cm from its initial location or >80% of it was consumed (determined by visual observations by the same observer). Preliminary studies of 30 samples in the young secondary forest showed that the scavengers would bring the bait back to the nest after 10 cm displacement or consume the whole bait after 80% consumption (C. K. W. Tan & R. T. Corlett, unpubl. data). The risk of baits being displaced by wind was minimised by conducting the study on days with low wind speed as well as confirming that a scavenger was seen moving the bait. A species was considered a scavenger if it attempted to move the bait, moved the bait and/or fed on or fragmented the bait. Ants were identified to genus or species level in the field whenever possible. However, when determination was unclear, ant scavengers were photographed and voucher specimens were collected from the surrounding area using a pair of forceps, so as to minimise disturbance to the scavenging process. Classification to genus level was done using identification guides (Hölldobler & Wilson, 1990; Bolton, 1994). Several genera of ants were identified to species level using taxonomic journals or with the assistance of taxonomists. Daytime observations were made between 11:00 and 15:00 hours in sunny weather conditions. In addition, night observations were made using the night-shot mode of a video camera (Sony Handycam HDR-SR 10E, Sony Corporation, Tokyo, Japan) between 21:00 and 01:00 hours, with two bait stations per habitat type. These selected sites were at different positions from those chosen for the day observations. A total of 50 observations were made for each type of bait for each habitat type.

Statistical analyses

We analysed species abundance at different habitat and bait types, using R (version 2.10.0, R Development Core team, 2009), through separate generalised linear mixed models (GLMM) with Poisson error distribution and log link function, with the number of scavenger species (ant or other) as the response variable, bait type and habitat type as fixed factors, and bait stations nested within sites and sites nested within habitat as random effects.

The scavengers were assigned to four categories according to whether they scavenged alone (solitary foraging) or in groups and whether they ate the bait on-site or moved it away. The classification of each taxon was based on its most frequently adopted behaviour. To investigate how recruitment decisions of ant scavengers were related to bait size, the body lengths of ten individuals of each scavenging ant species were measured and the mass of the bait was divided by the mean length of the ants. Because solitary foragers exhibited no recruitment behaviour, ant species that foraged alone were omitted from this analysis. A two-sample t-test, assuming equal variance, was used to compare the bait mass:ant length ratio when there was recruitment with the ratio when there was no recruitment.

First detection time was transformed by x^0.15 to approximate assumptions of normality. The Box–Cox transformation was used to determine the optimal power transformation for purposes of normalisation of data (Box & Cox, 1964). We entered in a GLMM with linear link function, first detection time as the response variable, bait type and habitat type as fixed factors, and bait stations nested within sites and sites nested within habitat as random effects.

To examine the rate of scavenging in the various habitats, survival analysis (SPSS Statistics 17.0, SPSS Inc., Chicago, IL, USA) was conducted using Cox regression (Cox, 1972) with the factors habitat type and bait type. To verify that the data fulfils the proportional hazards assumption of Cox regression, the cumulative hazards functions (Kaplan–Meier model) for the covariates were plotted and the lines were examined to ensure that they did not cross one another. The survival functions were compared with one another using the Wald value test in the Cox regression model.

Results

The majority of scavenging events were by ants (Formicidae) (86%), followed by flies (Diptera) (6%), cockroaches (Blattodea) (2%) and others (6%) (see Table 1 for list of scavengers), with each additional taxon contributing <2%. The only situation where ants were not the dominant scavengers was at night in grassland, when an earwig (Dermaptera, Labiduridae) was the dominant scavenger. At other sites, differences between night and day were minor and the results have been pooled for all sites.

Table 1.   List of scavenger taxa with the number of scavenging events for each bait type in each habitat.

Ants (Formicidae)	PF					OSF					YSF					RP					GL					IS
	T	G	C	E	S	T	G	C	E	S	T	G	C	E	S	T	G	C	E	S	T	G	C	E	S	T	G	C	E	S
Anoplolepis gracilipes Smith	7	8	9	8	10	5	7	8	5	8	11	7	13	13	7											4	4	7	7	8
Aphaenogaster sp.	1	3	3	1	2
Camponotus sp1	1	2	1	1	5	3	3	3	2	1
Camponotus sp2				1							3	6	1	2	2
Camponotus sp3																						1	1	1	1
Camponotus gigas Latreille	1	1			1	1			1	1
Crematogaster sp.											1			1	2
Diacamma sp1											3	16	10	12	25								1	1	1
Diacamma rugosum Le Guillou	1	2	2	2					1
Dolichoderus sp.	2	1	1	1		1	1	2	2	4	18	6	7	17	9						2		1	2	2	1		2	1	2
Euprenolepis sp.	1	1		1	2
Iridomyrmex sp.																			1	2		1		1
Meranoplus bicolor Guérin-Méneville																1	2	6	7	9	3		1	1	5
Monomorium pharaonis Linnaeus			2		1																					1		2	2	2
Monomorium monomorium Bolton											1	2					1	5	3	3	6	5	10	6	11
Odontomachus rixosus Smith	2	5	2	3	1	4	3	2	3	5		1	1	1	2
Odontomachus simillimus Smith																3	1	2	3	1
Odontoponera transversa Smith	8	4	8	4	3	13	11	13	12	8	6	3	2	1							1	1		1	1
Oecophylla smaragdina Fabricius																6	6	10	1		16	13	8	8	7
Paratrechina sp1							1	1	3
Paratrechina longicornis Latreille																2	3	7	2	3						5	3	5	3	1
Pheidole sp1							2	3	3	3
Pheidole aglae Forel	2	1	2	3	1
Pheidole aristotelis Forel			1			1	1		1	1
Pheidole cariniceps Eguchi						2	2	2
Pheidole clypeocornis Eguchi	2	1		2	1
Pheidole longipes Smith	12	11	9	7	10			1	1	1
Pheidole megacephala Fabricius																										1	1	2	1	1
Pheidole parva Mayr											5	7	10	9	12	3	6	6	5	7						4	2	3	2	3
Pheidole plagiaria Smith	5	6	5	7	6	8	10	10	8	7	3	3	4									1			1
Pheidologeton diversus Jerdon											3	2		1	5	6	7	2	4	6
Philidris sp.	1		1	1	1
Polyrhachis armata Le Guillou								2			2	1	1	2	1
Polyrhachis illaudata Walker											1	5	4	11	5
Proatta brutteli Forel			1	2	1	2	1	1	1	3
Pseudolasius sp.	1	1	1	2	1
Tapinoma sp1		1	2	3	2	2	3
Tapinoma melanocephalum Fabricius												5				5	9	11	11	14	6	3	5	5	6		1	3	2	2
Technomyrmex kraepelini Forel			1	1
Tetramorium sp1																	1	1	2
Tetramorium sp2																1	2	3	5	3						1	2	1		1
Tetramorium smithi Mayr												1	2	3	3

Non-ant	PF					OSF					YSF					RP					GL					IS
	T	G	C	E	S	T	G	C	E	S	T	G	C	E	S	T	G	C	E	S	T	G	C	E	S	T	G	C	E	S
Habitats: PF, primary forest; OSF, old secondary forest; YSF, young secondary forest; RP, recreational park; GL, grassland; IS, impervious surface. Baits: T, tree nymph; G, great eggfly; C, cricket; E, earthworm; S, superworm.
Cockroaches (Blattodea)
Nymph			1			1	1	1			2	4	4	2								1	1	1
Sp. 1					1	1		1							1
Sp. 2	1	1	1	1	1					1
Sp. 3					1
Flies (Diptera)
Sarcophagidae	1				1			1								3	1		2				1
Muscidae				1			1				4	1		3		1	1		2		2	3	3	3	2
Sepsidae						3	2		5							2		1	3	3	1	7	2	2		1
Tephritidae				1			1							6	1				3	2
Calliphoridae				1	1
Others
Nasutitermes sp. (Isoptera)									1
Gryllidae									4
Laniatores, Opiliones								1
Silphidae	1			1	3	1		1	2	1	2	3	4	2	5
Scarabaeidae		1			1
Diplopoda									1
Oniscidea													1	2	1								1		5
Labiduridae sp. 1																					2	3	4	2	4
Labiduridae sp. 2													1
Cicadoidea							2											1	1			3	1	4	3
Lepidoptera											1
Pulmonata														2
Acridotheres tristis (Passeriforms)																										1	1			3
Macaca fascicularis (Primates)	4	4

Species richness of ants and other scavengers differed significantly among habitats and baits (Table 2). Post-hoc tests reported significantly more scavenger ant species in the forests and recreational park than in the grassland and impervious surfaces (mean ± SE, primary forest: 0.904 ± 0.027; old secondary forest: 0.768 ± 0.033; young secondary forest: 1.176 ± 0.045; recreational park: 0.860 ± 0.038; grassland: 0.432 ± 0.043; impervious surfaces: 0.288 ± 0.030). The cricket attracted significantly more ant species than did the great eggfly and tree nymph. In addition, the superworm and earthworm attracted significantly more ant species than the tree nymph (cricket: 0.867 ± 0.033; superworm: 0.807 ± 0.036; earthworm: 0.850 ± 0.041; great eggfly: 0.660 ± 0.036; tree nymph: 0.507 ± 0.036). As for other scavengers, there were significantly less non-ant species in imprevious surfaces than the other habitats (primary forest: 0.116 ± 0.022; old secondary forest: 0.128 ± 0.011; young secondary forest: 0.236 ± 0.031; recreational park: 0.128 ± 0.023; grassland: 0.212 ± 0.026; impervious surfaces: 0.032 ± 0.011). The differences in non-ant species abundance among baits were mostly the result of the earthworm attracting significantly more non-ant species than the cricket (cricket: 0.100 ± 0.017; superworm: 0.123 ± 0.019; earthworm: 0.200 ± 0.029; great eggfly: 0.137 ± 0.021; tree nymph: 0.150 ± 0.022).

Table 2.   Results of GLMM procedure for ant species abundance, other species abundance and first detection time. Habitat type and bait type were considered as fixed factors.

Response variable	Factors	χ2	d.f.	P
Ant species abundance	Habitat	182.06	5	<0.0001
	Bait	40.12	4	<0.0001
	Interaction	30.74	20	0.059
Other species abundance	Habitat	39.58	5	<0.0001
	Bait	11.42	4	0.022
	Interaction	33.87	20	0.027
First detection time	Habitat	59.40	5	<0.0001
	Bait	26.99	4	<0.0001
	Interaction	27.91	20	0.11

Most scavenger species moved the whole bait or fragmented it and moved the fragments (Table 3). Non-ant scavengers mostly scavenged alone, but most ant taxa recruited nestmates to scavenge as a group. Whether or not an ant recruited nestmates depended in part on the bait mass: ant length ratio, with high ratios encouraging recruitment (two-Sample t-test, d.f. =1, t = −20.64, P < 0.001).

Table 3.   Categories of scavenger behaviour.

	Feed on-site	Remove bait
Scavengers were categorised into the four groups based on field observations of their most frequently adopted behaviour.
Solitary	Odontomachus simillimus	Camponotus sp3
	Polyrhachis illaudata	Camponotus gigas
	All cockroaches	Diacamma sp1
	All flies	Diacamma rugosum
	All other non-ant taxa	Odontoponera transversa
		Polyrhachis armata
Group	Camponotus sp1	All other ant taxa
	Camponotus sp2	Termite
	Meranoplus bicolor
	Tetramorium smithi
	Woodlouse

First detection time differed significantly among habitats and baits, and there was no significant interaction between bait and habitat (Table 2). Post-hoc Bonferroni tests identified a significantly faster first detection time in the young secondary forest as compared to other habitats (Fig. 1). Also, first detection times in the grassland and impervious surfaces were significantly longer than that in the old secondary forest, primary forest and recreational park (Fig. 1).

Figure 1.  Mean first detection time in relation to habitat type and bait type. Error bars are standard errors. Habitats that are significantly different (Bonferroni test; P < 0.05) are denoted by different letters.

The survival times of the baits increased in the order primary forest < old secondary forest < young secondary forest < recreational park < grassland < impervious surfaces (Fig. 2a, Table 4). The survival time of different baits increased in the order cricket < superworm = earthworm = great eggfly < tree nymph (Fig. 2b).

Figure 2.  Cox regression model: cumulative survival functions for (a) habitat types across all baits; and (b) bait type across all habitats. The ranking of survival functions as depicted in Table 4 is shown below each graph.

Table 4.   Results of contrasts between overall habitat and bait survival functions.

Comparisons		Wald Statistic	d.f.	P
Comparisons between the survival functions of each pair of habitats or baits were conducted. P-values < 0.05 indicate a significant difference in the survival functions of the pairs.
Habitats		362.744	5	0.000
Primary forest	Old secondary forest	7.069	1	0.008
Old secondary forest	Young secondary forest	5.136	1	0.023
Young secondary forest	Recreational park	25.097	1	0.000
Recreational park	Grassland	13.538	1	0.000
Grassland	Impervious surfaces	14.769	1	0.000
Baits		84.027	4	0.000
Tree nymph	Great eggfly	27.164	1	0.000
Great eggfly	Earthworm	0.304	1	0.582
Earthworm	Superworm	0.909	1	0.340
Superworm	Cricket	6.454	1	0.011

Discussion

Scavenging in the lowland tropics is a race against microbial decay, putting a premium on rapid detection of carcasses. Carrion-feeding flies in the families Calliphoridae and Sarcophagidae arrive at vertebrate carcasses within minutes of death and are usually dominant on the soft tissues, although beetles are also important and ants sometimes exclude other animals (Corlett, 2009). Southeast Asia lacks specialist vertebrate scavengers, but opportunistic consumption of dead vertebrates has been reported for most mammalian carnivores in the region, a variety of other mammals, and many birds. This study has shown that the situation is very different with invertebrate carcasses, where ants are almost always dominant. The high percentage of scavenging events by ants suggests that they were more successful at finding dead invertebrates than scavengers that normally feed on the much larger carcasses of vertebrates (e.g. flies and beetles), as well as other generalist scavengers (e.g. cockroaches and vertebrates). Ants adopt complex foraging patterns (e.g. Moffett, 1988a) that are adapted for locating small food items efficiently and many species have a large number of workers, resulting in a high density of foraging activity on the ground (Beckers et al., 1989). Moreover, most ant species work in groups and the aggression they display may deter even much larger animals.

Most ant species in this study were found to move or fragment the bait, or recruit other nestmates to feed on it, while the non-ant scavengers were almost all solitary. Two ant species, Odontomachus simillimus and Polyrhachis illaudata, did not recruit nestmates, although recruitment by P. illaudata has been shown in another study (Liefke et al., 2001). In most ant species, recruitment was apparently from the nest, but in Pheidologeton diversus ants were recruited from a nearby trail system. This species uses a group hunting system, with foragers leaving the nest and foraging collectively in a swarm along a well-defined trail system that is constructed as the swarm progresses (Moffett, 1988a,b). Recruitment was more likely when baits were large relative to the size of the ants. If a food item is so large that a scout could carry only a small portion of it, then a mass recruitment system will result in many recruits quickly reaching, defending, and retrieving the food source (Pasteels et al., 1987; de Biseau et al., 1997). In this case, the increased food retrieval rate might compensate for a decreased discovery rate and give a selective advantage to the use of recruitment over individual foraging.

Yamamoto et al. (2009) found that workers of most arboreal ants fragment large prey at the site of prey capture and individual workers retrieve the smaller pieces to the nest, while in most ground-living species a group of workers retrieve large prey cooperatively without fragmentation. Our study used somewhat larger baits and fragmentation was common, even with ant taxa (such as Anoplolepis gracilipes, Oecophylla smaragdina, Odontomachus rixosus and Paratrechina longicornis) which usually retrieved baits intact in the Yamamoto et al. study.

The higher scavenger species richness in the less modified habitats (forests and recreational park) compared to the highly modified habitats (grassland and impervious surface) agrees with many other studies that have shown lower species richness of ants and other insects in disturbed habitats (Thompson & McLachlan, 2006; Brühl & Eltz, 2009). Moreover, in Singapore the scavenging ant community in non-forest habitats, which are entirely anthropogenic, included several non-native tramp ants, such as Anoplolepis gracilipes, Monomorium pharaonis, Paratrechina longicornis, Pheidole megacephala and Tapinoma melanocephalum, of which only A. gracilipes occurred also in tall forests.

The average first detection time for a habitat is a good indicator of scavenger abundance as well as the number of scout ants in an area (Sainte-Marie & Hargrave, 1987; Adler & Gordon, 1992). Scavengers in young secondary forest were faster in locating the baits relative to scavengers in the grassland and impervious surfaces, suggesting scavengers were relatively abundant in the young forest.

The more modified the habitat, the longer the survival times of the baits (Fig. 2a). This is expected, as the higher number of scavenging species and the earlier detection time in the forests relative to the artificial habitats would allow for a faster scavenging rate. The fact that all the dead invertebrates in the primary forest were located quickly and scavenged rapidly implies that scavengers were abundant and there was a high competition for such food items. It also demonstrates the dominance of scavenging over microbial decomposition in the recycling of dead invertebrates. The increasing survival times with decreasing habitat quality further suggests that scavenging is less efficient in the poorer sites.

Among the baits, crickets were detected earliest and removed fastest. The tree nymph was removed slowest. This apparent difference in first detection time and survival times may be explained by the nutrient composition, with a much higher protein contents reported for the house cricket (205 g kg⁻¹) and superworm (197 g kg⁻¹) than a representative earthworm Lumbricus terrestris (105 g kg⁻¹) (Finke, 2002). Butterflies contain a large amount of unpalatable material (wings and exoskeleton) which was left uneaten. Another factor may be bait size and shape. The average mass decreased in the order tree nymph (0.530 ± 0.013 g) (mean ± SE), superworm (0.451 ± 0.009 g), earthworm (0.385 ± 0.010 g), cricket (0.266 ± 0.007 g), and great eggfly (0.245 ± 0.008 g), while the butterflies are also very awkwardly shaped.

Although this study has shown that scavenging rates differ among the habitats, the amount of invertebrates that naturally die in the various habitats may also be dissimilar (Saint-Germain et al., 2007; Lakka & Kouki, 2009). In such a case, the delayed scavenging of a lower biomass of dead invertebrates in the artificial habitats might be as effective as the rapid scavenging of a high biomass of dead invertebrates in the natural forest. Future studies need to investigate the biomass of dead invertebrates in each habitat in order to analyse the scavenging rate relative to the abundance of dead invertebrates.

Overall, forests had more scavenger species and shorter bait survival times than the non-forest habitats. Of the parameters measured, bait survival time increased monotonically along the urbanisation gradient. Given that scavenging of invertebrates is an important ecological process, these results suggest that this parameter could be used as a simple bioindicator of habitat quality. The other parameters showed the same general pattern, but the most easily measured one, first detection time, is probably too sensitive to ant abundance to be used as a measure of overall quality.

Conclusions

Although there is no data for the tropics, the total biomass of invertebrate carrion produced per unit area per year probably greatly exceeds that of vertebrate carrion. Compared with dead vertebrates, however, dead invertebrates are small, short-lived food items for which the efficient foraging systems and numerical abundance of ants are particularly well suited. It appears that, at least in Singapore, this has not allowed a guild of specialist scavengers on invertebrates to evolve. Fast and efficient foraging by ants also ensures that microbial decomposers lose out. This scavenging process is remarkably resilient to massive anthropogenic habitat modification. Although dead invertebrates are detected less rapidly and removed more slowly in highly modified habitats, ants eventually found and removed most baits, even on artificial surfaces.

Acknowledgements

Mohamad Faizal, Nghiem Thi Phuong Le, Choo Yuan Ting, Ben Wong, Hamsa d/o Suppayan Sodimani, Tan Yingxin, Justin Loh and Jimmy Chan helped in the field. Professor Seiki Yamane, Dr Katsuyuki Eguchi, Dr John Fellowes, Dr David Lohman, Tan Junhao and Ms Lua Hui Kheng assisted with ant identification. Dr Peter Todd helped with the statistics. The Singapore Zoological Garden provided the butterflies used as baits in this study. All experiments comply with the current laws of Singapore.