Effects of natural and synthetic alarm pheromone and individual pheromone components on foraging behavior of the giant Asian honey bee, Apis dorsata

Alarm pheromones play an important role in social insects and enhance collective fitness by providing information about dangers such as predators (Billen and Morgan, 1998). For example, honey bees can use alarm pheromones to co-ordinate colony defence (Free, 1987; Pirk et al., 2011). However, the role of alarm pheromones in communicating danger at food sources is less well understood (Nieh, 2010; Goodale and Nieh, 2012). Honey bees (Apis mellifera, A. dorsata and A. florea) avoided flowers upon which a conspecific had been attacked by predators and therefore potentially released alarm-based odors (Llandres and Rodriguez-Gironés, 2011; Llandres et al., 2013). Although it is highly likely that these bees released alarm pheromone compounds that elicit conspecific avoidance of these inflorescences, it was not clear whether they were avoiding alarm pheromone or other olfactory cues. Therefore identifying and testing different chemical components of alarm pheromone in honey bees is important for understanding what elicits avoidance following attacks at natural food. In fact, with the exception of the European honey bee, A. mellifera, little is known about which alarm pheromone components elicit alarm. Understanding the role of different alarm pheromone components is relevant because this tells us which components elicit alarm, provides information on shared and thus potentially ancestral components in different honey bee species, and opens the possibility of heterospecific responses to shared pheromone components.

Most social insects use alarm pheromones, which have multiple evolutionary origins, in some cases evolving from compounds with toxic properties (Blum, 1969). In ants, alarm pheromones occur in nearly all tested species and are typically produced by glands associated with defence or offence, such as sting poison glands (Holldobler and Wilson, 1990). Alarm pheromones are similarly widespread in social wasps and are found in the Vespa, Vespula, Polistes, Ropalidia, Polybia, Dolichovespula, Provespa and Polybioides, where they are generally released from sting venom glands (Fortunato et al., 2004). However, one species, Vespula squamosa, evidently releases alarm pheromone from a gland in the head (cephalic gland) (Landolt et al., 1999). This strategy is shared by all studied species of stingless bees (Hymenoptera, Apidae, Meliponini), which lack stings and therefore attack by biting (Roubik, 1989). Stingless bees use cephalic secretions as alarm pheromones, in all examined cases from mandibular glands (Nieh, 2004; Schorkopf et al., 2009). Surprisingly, bumble bees (Hymenoptera, Apidae, Bombini) generally do not seem to use alarm pheromones (Maschwitz, 1964), although Llandres et al. (Llandres et al., 2013) recently demonstrated that Bombus terrestris would avoid flowers upon which a conspecific had been attacked and potentially deposited an alarming odor. The source of this odor remains unknown.

Honey bees (Hymenoptera: Apidae, Apini) produce alarm pheromone in the sting gland and a pheromone that has sometimes been described as an alarm pheromone in their mandibular glands, but which does not galvanize colony aggression (Vallet et al., 1991; Couvillon et al., 2010). In fact, recent evidence suggests that A. mellifera mandibular gland pheromone is a defensive compound for paralysing hive intruders (Papachristoforou et al., 2012). Suwannapong et al. (Suwannapong et al., 2011) tested the response of A. andreniformis and A. florea foragers to synthetic compounds found in bee mandibular glands. They applied these compounds to filter papers placed beneath flowers and found that some compounds, depending upon their concentration, repelled or attracted foragers of A. andreniformis. The tested compounds only repelled A. florea. It would therefore be useful to determine the responses of honey bees to alarm compounds, not only from mandibular glands, but from the entire bee.

Upon attack, honey bees produce an alarm pheromone that elicits different responses depending upon the setting. For example, exposure to A. mellifera sting alarm pheromone results in attacks from defenders near the nest (Wager and Breed, 2000) but causes aversion at food (Goodale and Nieh, 2012) and reduces recruitment to this food location (Nieh, 2010). Such context dependency is also exhibited by honey bee queen pheromone, which elicits a wide variety of behaviors (the queen retinue, division of labor, swarm clustering, drone attraction, etc.) depending upon the receiver and the environment (Free, 1987).

In the highly social bees (honey bees and stingless bees), alarm pheromones can increase aggression and flight activity at the nest entrance, a defensive response (Boch and Shearer, 1971; Roubik, 1989). However, at a food source, alarm pheromones ultimately cause avoidance. Stingless bee foragers (Melipona panamica) hover around a feeder with mandibular gland alarm pheromone extract, but then preferentially land on a control feeder without the pheromone (Nieh, 1998). Other stingless bee species, Scaptotrigona depilis and Trigona spinipes, avoid a nectar feeder with conspecific mandibular gland extracts (Schorkopf et al., 2009). Similarly, honey bee (A. mellifera) foragers avoid sting gland alarm pheromone placed on a sucrose feeder (Goodale and Nieh, 2012). A major component of A. mellifera sting gland pheromone, isopentyl acetate, will repel bees from visiting when added to a sugar solution feeder (Ferguson and Free, 1979) or to oil-seed rape, field beans and sunflowers (Free et al., 1985). However, it is not clear whether other compounds in honey bee alarm pheromone can also elicit such avoidance.

Predation at food sources can elicit alarm, and our study was inspired by predation observations. Hornets (Vespa tropica) caught A. dorsata foragers on flowers of Syzygium jambos L. near our field site in Menglun, China. We also observed weaver ants (Oecophylla smaragdina) attacking A. dorsata foraging on inflorescences (Calliandra haematocephala). Ants attacked 3% of visiting bees (980 bee visits observed) and 31% of these attacks successfully killed the bee forager (Li et al., 2014). Ant presence alone significantly reduced bee floral visitation by 40% (Li et al., 2014). During these attacks, we smelled A. dorsata alarm phermone released by attacked bees, and we wished to determine whether this alarm pheromone would repel foragers. Apis dorsata is a good species to study because little is known about its behavioral responses to alarm pheromone. Comparative information about alarm pheromone responses from multiple bee species is necessary to understand alarm pheromone evolution. In addition, A. dorsata is widespread throughout Asia (Oldroyd and Wongsiri, 2006; Hepburn and Radloff, 2011), where it is an important pollinator of crops (Wongsiri et al., 2001; Partap, 2011) and native plants (Corlett, 2011). Finally, A. dorsata is sympatric with two other native Apis species with which it shares alarm pheromone components (Boch et al., 1962; Morse et al., 1967; Koeniger et al., 1979) and, potentially, alarm information.

We therefore tested the effects of natural alarm pheromone, individual pheromone components that we chemically identified, and a synthetic mixture of these components on A. dorsata foraging at C. haematocephala inflorescences. Our goal was to understand how A. dorsata alarm pheromone would affect conspecific foraging behaviour, and to determine which chemical components have the strongest effect on conspecific foragers.

We used solid phase micro-extraction (SPME) fibers to collect alarm pheromones from the airspace around bees captured in vials. The alarm pheromone components that we identified (Fig. 1, Table 1) largely match those reported by previous investigators (Table 2). The two most abundant compounds are gamma-octanoic lactone and isopentyl acetate. Gamma-octanoic lactone has not been previously identified in any honey bee alarm pheromone, although it is found in the cephalic extract of A. laboriosa (Table 2). In addition, we provide the first identification of the following minor A. dorsata alarm pheromone compounds: 1-octanol (found in A. mellifera sting gland pheromone), and decanal (not previously reported in Apis alarm pheromone; Tables 1, 2).

We gave foragers a choice between two inflorescences, one with the treatment odor and the other with a control. We assayed the choices of 1561 bees, counting each choice only once. For compounds that elicited a strong aversion, such as natural alarm pheromone (Table 3, Fig. 2), foragers exhibited a classic behavior. They slowly hovered over the inflorescence with the alarm odor, and then landed on the control inflorescence or flew away from the array. Foragers significantly avoided landing on inflorescences with natural alarm pheromone, the synthetic mixture, and all alarm pheromone components (χ² tests, P≤0.006), with the exception of isopentyl propionate, octyl acetate and 3-methyl-1-butanol (P≥0.42; Table 3).

The overall repellency of our floral array matches these landing choices: 33–49% of approaching foragers left without making a choice when the array presented natural alarm pheromone, the synthetic mixture, gamma-octanoic lactone, isopentyl acetate or (E)-2-decenyl acetate. However, only 20–24% of approaching bees left without landing when the array presented isopentyl propionate, octyl acetate or 3-methyl-1-butanol (Table 3).

There is a significant effect of compound type on the number of bees that landed (ANOVA, F_9,288=21.68, P<0.0001; Fig. 2). There is no effect of trial date (F_1,288=0.01, P=0.91). The numbers of bees that landed on the inflorescence-only control (no added odor) and on the hexane-only treatment are, respectively, 5.6±1.3 and 6.0±1.9 bees per 10 min trial [no significant difference, Tukey's honestly significant difference (HSD) test, P>0.05]. Releasing natural alarm pheromone at an inflorescence significantly decreased the number of landing bees to 1.7±1.5 bees (significantly lower than either control treatment, Tukey's HSD test, P<0.05; Fig. 2).

There is no significant difference between aversion to natural alarm pheromone and the synthetic mixture (Tukey's HSD test, P<0.05). The two control treatments and three of the test treatments (isopentyl propionate, octyl acetate and 3-methyl-1-butanol) elicited the least aversion and were visited by approximately equal numbers of bees (no significant differences, Tukey's HSD test, P<0.05). All other tested compounds elicited intermediate aversion levels (Tukey's HSD test, P<0.05; Fig. 2).

Apis dorsata foragers avoided their alarm pheromone when this was presented at a food source, a situation that occurs when foraging bees are attacked by predators. This avoidance lasted for at least 10 min, the duration of our trials. Using gas chromatography–mass spectrometry (GC–MS) analysis, we identified eight chemical components in the alarm pheromone, of which three components (1-octanol, decanal and gamma-octanoic lactone) have not previously been identified in this species. GC–MS analyses of hexane extracts of sting glands would probably have yielded more compounds, but we used SPME to focus on airborne volatiles. We then tested six major compounds [3-methyl-1-butanol, isopentyl acetate, isopentyl propionate, octyl acetate, gamma-octanoic lactone and (E)-2-decen-1-yl acetate] and found that a synthetic mixture of these compounds (in the average ratio identified in our GC–MS analyses) elicited behaviors statistically indistinguishable from forager responses to an alarmed conspecific.

We show that gamma-octanoic lactone, isopentyl acetate and (E)-2-decen-1-yl acetate are active compounds and act like natural alarm pheromone (Table 3). It is possible that foragers also respond to other alarm pheromone compounds, but these three compounds are sufficient to elicit forager aversion. Of all the compounds tested, gamma-octanoic lactone appears to be the most effective. It elicited a landing aversion response that was not significantly different from the response to natural alarm pheromone (Fig. 2). Isopentyl propionate, octyl acetate and 3-methyl-1-butanol did not repel bees from landing (Table 3). Responses to these compounds were not significantly different from responses to the hexane-only control (Fig. 2). These reactions were probably not nestmate-specific because bees visiting our inflorescences probably came from multiple nearby colonies. Moreover, in the context of defence, honey bee alarm pheromone can elicit a general defensive response from nearby colonies, not just from the colony of the attacked bee (Graham, 1992).

We identified three new compounds not previously reported in A. dorsata alarm pheromone (Table 2). It is possible that some of these compounds, particularly those with low relative abundance, were not part of the alarm pheromone but were volatiles produced by other bee body parts. However, gamma-octanoic lactone was produced in high relative abundance (Table 1) and therefore may be part of the alarm pheromone. With respect to the two A. dorsata alarm compounds, 1-octanol is found in the sting alarm pheromone of A. mellifera (Collins and Blum, 1982; Wager and Breed, 2000), and decanal has been detected in adult A. mellifera workers (Torto et al., 2005).

Changes in inflorescence visual appearance are unlikely to account for our results. When we applied the synthetic alarm pheromone, we were careful to not bruise the inflorescence and did not touch it with the tip of our pipette. When we applied natural alarm pheromone, we took care to prevent the struggling bee from bruising the inflorescence during alarm pheromone deposition pheromone. It is possible that the appearance or natural odor of the test inflorescence was slightly altered by contact with the struggling bee, not by the bee's alarm pheromone. Nonetheless, bees exhibited the same statistically indistinguishable level of landing avoidance to natural and synthetic alarm pheromone (Fig. 2), even though no bee made physical contact with the inflorescence in the synthetic alarm pheromone tests.

Other bee-deposited odor marks probably did not affect our overall results. Bees can leave odor marks that are repellant or attractive (Saleh et al., 2007; Yokoi and Fujisaki, 2009; Llandres et al., 2013). To exclude prior odor-marking, we enclosed all inflorescences in bags while they were still buds, before they produced nectar and would have attracted bees. We only removed the bags right before a test. During a trial, we immediately removed bees as soon as they landed, but it possible that they released odors when they landed. Although landing bees or the process of capturing bees may have produced odors that influenced the choices of subsequent bees, there are nonetheless strong, significant differences between the number of bees choosing the different treatments, even though all trials were conducted under similar, standardized conditions.

Finally, bee avoidance of our test odors probably does not arise from avoidance of inflorescences that smelled differently because of our treatments. Bees rejected specific olfactory compounds that we added, but did not reject isopentyl propionate, octyl acetate or 3-methyl-1-butanol (Table 3, Fig. 2). Honey bees do not necessarily reject food sources when novel odors are subtracted or added. Heiling et al. (Heiling et al., 2003) showed that honey bees continued to visit flowers at the same rate even when floral odors were removed. Suwannapong et al. (Suwannapong et al., 2011) added different compounds (identified from honey bee mandibular glands) to flowers and reported attraction, repulsion or no significant behavioral changes depending upon the compound and its concentration. Likewise, Goodale and Nieh (Goodale and Nieh, 2012) showed that honey bees could smell and avoid the odor of conspecific haemolymph but did not avoid the odor of bumble bee hemolymph added to a food source.

We will now consider each of the major alarm compounds in order of response strength. Gamma-octanoic lactone elicited the greatest response of any single compound (Fig. 2). The landing aversion response elicited by this compound was indistinguishable from forager responses to natural alarm pheromone (Fig. 2). It is also one of the most abundant compounds that we identified in A. dorsata alarm pheromone (Table 1). This compound was first reported in the cephalic extracts of A. laboriosa, but was not previously found in the alarm pheromone of A. dorsata (Blum et al., 2000). It has not been reported in any other honey bee species.

It is possible that Blum et al. (Blum et al., 2000) did not previously identify gamma-octanoic lactone in A. dorsata alarm pheromone because of methodological differences. We used SPME analysis of the total volatile odors produced by alarmed A. dorsata foragers. The advantage of this approach is that it captures all of the volatile odors that bees release when they are alarmed, not just the components of a single gland. Our analysis could therefore include other components produced during alarm such as compounds produced by cephalic glands. Blum et al. (Blum et al., 2000) dissected out sting shafts of A. dorsata workers into a solvent and then analysed the resulting mixture. Interestingly, Blum et al. (Blum et al., 2000) found gamma-octanoic lactone in A. laboriosa cephalic extracts, although they did not find this compound in A. dorsata cephalic extracts. There may also be differences in the alarm pheromone components of different A. dorsata populations. We used A. dorsata foragers collected in Xishuangbanna, China, and Blum et al. (Blum et al., 2000) used A. dorsata workers from Nepal.

This compound has now been identified in the alarm pheromone of every Apis species whose alarm pheromone has been analysed, with the exception of A. laboriosa, and is therefore the most commonly found sting pheromone compound in Apis. Isopentyl acetate also occurs in the sting gland pheromone of A. mellifera, A. cerana, A. dorsata and A. florea (Boch et al., 1962; Morse et al., 1967). Of these four species, A. dorsata produces the highest amount of isopentyl acetate per bee (Koeniger et al., 1979). In A. mellifera, A. florea and A. cerana, this compound can elicit attacks when provided within 60 cm of the nest (Morse et al., 1967). Koeniger et al. (Koeniger et al., 1979) reported that Apis dorsata foragers will attack a moving ball placed near the nest and baited with isopentyl acetate. At a floral resource, isopentyl acetate strongly repulsed A. dorsata foragers from landing (Table 3, Fig. 2).

Blum et al. (Blum et al., 2000) reported that the major component of sting shaft extracts of A. dorsata workers was 1-acetox-2-decene [synonymous with the 2-decen-1-yl acetate reported by Koeniger et al. (Koeniger et al., 1979) and Veith et al. (Veith et al., 1978)]. The correct name for this compound is (E)-2-decen-1-yl acetate (J. Millar, personal communication). Veith et al. (Veith et al., 1978) first discovered this compound in A. dorsata and A. florea workers, and in both species it elicits aggression when placed on a bait near the nest (Koeniger et al., 1979). Our study tested forager responses, and (E)-2-decen-1-yl acetate repelled foragers from landing on an inflorescence (Table 3, Fig. 2).

Finally, our study and Blum et al. (Blum et al., 2000) also identified a minor component, isopentyl propionate, in A. dorsata sting alarm pheromone. However, this compound did not elicit aversion. In our paired choice test, 47% of foragers landed on the inflorescence with this compound, a pattern not significantly different from random (P=0.42; Table 3).

We suggest that alarm pheromones evolved primarily in the context of colony defence, not for warning foragers at a food source, because the colony is the unit of selection and the likelihood of encountering a nestmate, even in species that communicate food location, is lower at an inflorescence than near a nest. However, even if the primary benefit of alarm pheromone signal production is colony defence, foragers from the same colony, other colonies of the same species, and even different species can benefit from sensing and using this information. Thus the simple, reflexive and reliable response of an attacked bee producing alarm pheromone, regardless of context, has broader implications for information flow. For example, A. cerana, A. dorsata and A. florea have overlapping distributions and share a common alarm pheromone component, isopentyl acetate (Table 2). Predation upon one species could therefore alert other species of danger and amplify the effects of a predator on pollination, a key ecosystem service provided by bees (Brittain et al., 2013). It would be interesting to test for such alarm pheromone eavesdropping. In the future, it would also be valuable to determine the alarm pheromone components of all Apis species for a detailed evolutionary analysis that considers biosynthetic pathways, behavior and phylogeny. Such knowledge can enhance our understanding of the ecology of information use (Schmidt et al., 2010) and how these alarm signals evolved.

This research was conducted in full compliance with the laws of the People's Republic of China. No permits were required for our field studies. We used A. dorsata, an abundant, non-protected species.

The giant Asian honey bee A. dorsata ranges from western India throughout continental and oceanic Asia, including Sulawesi, Indonesia and the Philippines (Oldroyd and Wongsiri, 2006; Hepburn and Radloff, 2011). We conducted our study (January to April 2013 and December 2013 to January 2014) at the Xishuangbanna Tropical Botanical Garden in Menglun, China (21.921°N, 101.231°E), during the dry season, when C. haematocephala blooms. At this site, there was a 300 m² patch of C. haematocephala (Nevling and Elias, 1971) trees in bloom. We observed wild colonies of A. dorsata foraging at this patch. The foragers probably came from ~40 colonies of A. dorsata located about 1 km away from our study site. Because we used naturally foraging bees, we could not determine how many different colonies came to our inflorescences. However, we conducted our studies for more than 1 year and used 20 different trees, leading to a high likelihood that we used bees from multiple colonies. We also captured all bees that approached our test array, and therefore recorded each bee's choice only once. Captured bees were subsequently frozen to eliminate potential pseudoreplication.

We separately analysed the alarm pheromone volatiles of 18 different bees. We used a small, clean glass bottle to capture each bee while it was foraging on an inflorescence. To elicit alarm, we clipped the bee's legs together with a clean metal binder clip. The bees soon began producing alarm pheromone. The advantage of this approach is that it captures all of the volatile odors that bees release when they are alarmed, not just the components of a single gland. For controls, we analysed the volatiles produced by 18 different non-alarmed bees. We also obtained these bees by using a small glass bottle to gently capture each bee as it foraged on an inflorescence. However, we did not clip the bee's legs together with a binder clip and took care to avoid agitating or disturbing the bee.

To obtain the odor, we waited 1 min and then quickly removed the bee from the vial to minimize odor loss. We adsorbed odors released into the glass bottle with a SPME fiber (75 μm CAR/PDMS, Supelco, Bellefonte, PA, USA) inserted into the glass bottle for 30 min. For analysis, this fiber was then desorbed in the gas chromatograph inlet at 220°C for 5 min. We used an HP 7890A gas chromatograph (Agilent Technologies, Santa Clara, CA, USA), equipped with a HP-5MS column (30 m×0.25 mm, 0.25 μm film thickness), and linked to an HP 5975C mass spectrometer (Agilent Technologies). Helium was used as a carrier gas at a flow of 1 ml min⁻¹, and the injector temperature was set to 260°C. Column temperature was 40°C and held for 1 min, increased to 80°C at a rate of 3°C min⁻¹, and then finally increased to 220°C at a rate of 5°C min⁻¹. We used the NIST08 MS library to identify possible compounds and then confirmed these compounds by comparing the GC retention times and MS of the alarm pheromone samples with those of authentic compounds (typical chromatogram shown in Fig. 1, average data shown in Table 1).

To estimate the average amount of each compound per forager, we used an average of published values of isopentyl acetate (20 μg) per A. dorsata forager (Morse et al., 1967; Koeniger et al., 1979). For all other tested compounds, we then calculated the estimated amount per bee based upon the relative amount (RA) ratios. For example, RA_{gamma-octanoic lactone}/RA_{isopentyl acetate}=0.323 (Table 1). Thus we estimated the amount of gamma-octanoic lactone to be 0.323×20 μg=6.46 μg per bee. We used these estimates to determine the amount of each compound to use in our floral choice tests.

From 20 trees, we randomly chose two inflorescences per tree. To obtain inflorescences that contained nectar and no odor marks, we placed mesh bags around C. haematocephala inflorescences while they were still buds and waited ~3 days until they became mature inflorescences. To ensure that the trees were ant free, we carefully inspected them before and after we applied ant-excluding rings of sticky Tanglefoot™ resin around the trunks and branches. We used paired choice tests between a treatment and a control inflorescence, following standard designs for testing bee foraging choices (Dukas, 2001; Abbott, 2006). There are two types of control treatments: inflorescence-only control (paired with the natural alarm pheromone treatment) and hexane control (paired with the synthetic compound treatments).

When the inflorescences were fully mature, we used them in a paired choice test by cutting two off and placing one per tripod on two 1.5-m high tripods separated by 40 cm and placed ~1 m away from the tree. We chose this 40 cm distance based upon other studies, which have used paired food sources separated by 6–30 cm and found strong effects on bee orientation when one food source had signals or cues of predation (Dukas, 2001; Abbott, 2006; Goodale and Nieh, 2012; Tan et al., 2013).

Bees then chose between the inflorescences during a test that lasted for 10 min and was conducted between 13:00 and 16:00 h on sunny days with abundant and consistent foraging activity. We chose 10 min because it corresponded to a period in which a fairly constant number of A. dorsata visited the C. haematocephala inflorescences. We tested eight different treatments: natural alarm pheromone, synthetic alarm pheromone mixture and six synthetic alarm pheromone components. We conducted 20 trials per treatment. Each treatment was replicated with inflorescences from 20 different trees. Inflorescences from the same tree were used for all eight treatments. Each pair of inflorescences was used for only one 10 min trial period and was then discarded.

After setting out the test stimulus, we waited 2 min to allow the alarm pheromone to volatilize and the hexane solvent (in tests of chemical compounds) to fully evaporate. Only a few bees approached the inflorescences during the 2 min pre-trial interval. During this pre-trial interval, we used insect nets to capture any bees that approached and did not count them as part of the trial. We then began the 10 min trial period. To exclude the possibility of bees being influenced by the choices of other bees, we only counted choices made in the absence of other bees foraging near the array. There were abundant natural floral resources nearby. Thus it was rare for one bee to approach the array while another bee was choosing. In all experiments, the treatments consisted of an odor applied to the test inflorescence, and therefore no other bees were present when foragers made choices. The experimenters were blind to which chemical treatment they used during these behavioral tests. During these tests, ants were excluded from the trees, and no other predators or bee species approached the inflorescences.

We counted each bee choice only once. Bees that landed on inflorescences were immediately captured with an aspirator. Bees that investigated the array (approached within ~5 cm of one of the inflorescence) but did not land were captured with insect nets as they began to fly away. Such investigating bees generally flew slowly, and we could therefore capture them with nets. By capturing these approaching bees, we ensured that each approaching bee was a different bee and could therefore calculate the percentage of approaching bees that landed during the trial.

We first tested the effect of natural alarm pheromone on an inflorescence, simulating a predator catching a bee by its legs. We caught an A. dorsata forager with a net and used a clean metal binder clip to hold its legs. We then carefully positioned the bee on top of the test inflorescence. The bee immediately began producing alarm pheromone, and we allowed it to mark the inflorescence with alarm pheromone for 1 min. We then removed the bee, sealing it in a glass vial to exclude further alarm pheromone production. We took care to not alter the visual appearance of the inflorescence or to bruise it.

All subsequent experiments focused on forager responses to synthetic compounds. We analysed the chemical components of alarm pheromone (see above) and identified six major compounds that we separately tested: 3-methyl-1-butanol, isopentyl acetate, isopentyl propionate, octyl acetate, gamma-octanoic lactone and (E)-2-decen-1-yl acetate (Fig. 1, Table 1). We also tested a synthetic mixture of these compounds, using the estimated mean quantities per bee (Table 1). To the test inflorescence, we applied one bee-equivalent of the alarm pheromone mixture or one bee-equivalent of the tested individual compound. We obtained all compounds from Aladdin Reagent Database Inc. (Shanghai, China). The (E)-2-decen-1-yl acetate was kindly synthesized for us by J. Millar (University of California Riverside, Riverside, CA, USA).

Using a micropipette, we dispensed 10 μl of alarm pheromone compound (containing one bee-equivalent of the test compound, Table 1) onto the test inflorescence, and 10 μl hexane onto the control. We carefully applied just the fluid onto the inflorescence and did not bruise the inflorescence with the pipette tip. We waited 2 min to allow the hexane to evaporate and then began the 10 min trial. We used hexane because it is an excellent solvent commonly used in olfactory bioassays (Millar and Haynes, 1998). Hexane also rapidly evaporates. At the average air temperature of our trials (>30°C), hexane quickly evaporated because it has a vapor pressure >187.11 mmHg, nearly sixfold greater than the vapor pressure of water under the same conditions (Beyer, 1988). Moreover, honey bees (A. mellifera) are not perturbed by hexane and will choose a hexane control feeder over a test feeder with a sting gland extract (Goodale and Nieh, 2012).

We used χ² tests to determine whether bees were attracted to the different olfactory compounds. Our null hypothesis is that equal numbers of bees will choose the control and treatment inflorescences in our paired choice tests. We then used ANOVA to compare the effect of different olfactory treatments and trial date (day of the year) on the number of bees landing on the different inflorescences. We counted each bee only once. We performed Tukey's HSD test to perform pairwise comparisons of all treatments (Zar, 1984). All data met parametric assumptions as determined through residuals analysis (Zar, 1984). We report means ± 1 s.d.

We would like to thank Jocelyn Millar for generously synthesizing (E)-2-decen-1-yl acetate for our experiments, and the Central Laboratory at Xishuangbanna Tropical Botanical Garden for technical support with the GC–MS analysis. We would also like to thank the editor and anonymous reviewers whose comments helped improve our manuscript.