Poison and alarm: the Asian hornet Vespa velutina uses sting venom volatiles as an alarm pheromone

Pheromones are important information agents and help to regulate colony behaviour in social insects, such as honey bees (Slessor et al., 2005), ants (Hölldobler, 1995), termites (Wen et al., 2014) and wasps (Bruschini et al., 2010; Turillazzi and Bruschini, 2010). Alarm pheromones can play dual roles, by activating nest defence and serving as a warning that allows foragers to avoid dangerous sites. For example, honeybee alarm pheromone can attract guards to the nest entrance for nest defence (Boch and Shearer, 1971; Roubik, 1989) and repel foragers from foraging sites with predators (Li et al., 2014; Wang et al., 2016). In social wasps and hornets, nest guards can also release alarm pheromones to recruit nest defenders (Bruschini et al., 2008). Hornet alarm pheromones can likewise be used to mark foraging sites (Ono et al., 2003). However, less is known about the pheromones involved in hornet defence compared with bee and ant alarm pheromones.

The diversity of alarm pheromone usage and glandular sources in social hornets and wasps provides key variation that can be used to understand the evolution of alarm pheromones in social insects. Determining the character states is therefore important. What species use alarm pheromones, what is the source, and what chemical components are involved? The ritualization hypothesis predicts that chemical weapons can become associated with attack or defence and thereby evolve into alarm pheromones. In fact, sting venom appears to be a primary source of alarm pheromone. Polistes dominulus (Bruschini et al., 2006; Landolt et al., 1998), Vespula squamosa (Heath and Landolt, 1988; Landolt et al., 1995), Vespa crabro (Veith et al., 1984), Vespa mandarinia (Ono et al., 2003) and Vespa simillima xanthoptera (Ono et al., 2003) provide examples. In these species, volatile alarm pheromones are released when sting venom is exuded by the stinger or via stinging (Ali and Morgan, 1990; Downing, 1991; Jeanne, 1981; Landolt and Akre, 1979). In Vespula spp., mandibular glands may provide alarm pheromones (Reed and Landolt, 2000), as they do in the stingless bees (Schorkopf et al., 2009). However, like some bumble bee species (Maschwitz, 1967), a few hornet and wasp species appear to lack alarm pheromone (Landolt et al., 1998).

In most social insects studied to date, alarm pheromones consist of multi-component blends (Bruschini et al., 2010; Turillazzi and Bruschini, 2010; Hölldobler, 1995; Slessor et al., 2005; Wen et al., 2014). Hornet alarm pheromones also contain multiple components, some of which are known to elicit alarm behaviour. It is useful to consider the functions of these different components. Some compounds may act synergistically with components that elicit an alarm response or serve different functions (Bruschini et al., 2010). Components that trigger alarm behaviour include N-3-methylbutylacetamide in Vespula squamosa (Heath and Landolt, 1988; Landolt et al., 1995), 2-methyl-3-butene-2-ol in Vespa crabro (Veith et al., 1984), 2-pentanol, 3-methyl-1-butanol, 1-methylbutyl 3-methylbutanoate in V. mandarinia and V. simillima xanthoptera (Ono et al., 2003), nonan-2-one in V. orientalis (Saslavasky et al., 1973), and amides and ketones in Dolichovespula maculata (Jimenez et al., 2016). However, the functions of other components remain unclear. For example, the roles of venom volatiles such as tridecane, pentadecane, pentadecene and undecane in seven Stenogastrinae species (Dani et al., 1998) are unknown, as are the functions of alkanes, monounsaturated alkenes and 2-alcohols in one Polistinae species (Sledge et al., 1999). Some compounds, particularly minor ones, may not have a clear adaptive value but could arise as metabolic byproducts.

To learn more about hornet alarm pheromones, we focused on Vespa velutina Lepeletier 1836, a common honey bee predator (Tan et al., 2007, 2016) and pest in Asia (Liu et al., 2016; de Haro et al., 2010) and a species that has recently invaded parts of Europe, to the detriment of the European honey bee Apis mellifera (Rortais et al., 2010; Villemant et al., 2006). Vespa velutina appears to be a fairly derived species within the genus Vespa (Perrard et al., 2013). Initial observations led us to believe that V. velutina has an alarm pheromone. When alerted near the nest, guards would exit to search for the disturbance. Once a guard found the intruder, it attempted to sting or exuded venom from its stinger. As a result, multiple hornets were immediately alerted and began attacking the target. Our goals were therefore to identify the source of the alarm pheromone, determine its behavioural effects on hornet workers, chemically analyse the pheromone, use electroantennography to identify active components, and test whether these components can elicit attacks.

We used 15 Vespa velutina colonies from two sites separated by over 100 km: Kunming Botanical Garden (KBG) in Kunming, China (25.44°N, 105.37°E) and a site in Wuding (WD), China (25.94°N, 104.27°E). Some experiments required dissecting hornets, but V. velutina is not an endangered species; in fact, it is invasive (Rortais et al., 2010; Villemant et al., 2006), and, in some areas, is considered a harmful pest (de Haro et al., 2010). All studies were carried out in compliance with relevant provincial and national guidelines.

In preliminary observations, we noted that V. velutina hornets disturbed at their nests would extend their stingers. A droplet of exuded venom was sometimes visible on these stingers. Subsequently, guards exited the nest in search of the disturbance. We therefore tested for the alarming effect of sting venom by observing hornet responses to different concentrations of sting gland extract applied to a target. We define an attack as a hornet landing on the filter paper and showing a sting posture by bending its abdomen.

We pipetted 0 (hexane-only control), 0.01, 0.1 and 1.0 gland equivalents onto a filter paper (5 mm×15 mm) placed behind a dry Quercus acutissima leaf hanging on a tree branch 30 cm from the nest entrance. Each wasp has one sting gland and thus one gland=one wasp equivalent (eq). Immediately after adding the treatment, we began a 3 min trial during which we counted the total number of hornets attacking the leaf. Only one quantity was used per trial. We conducted three trials per colony per sting gland concentration and ran one trial per colony per day. Each day, a different order of presentation was used, following a pseudo-random pattern that interspersed the different quantities but ensured that each was tested the same number of times. We used a new leaf for each trial and video recorded each trial.

Chromatography separated the venom compounds into two fractions – monoketones and diketones. In separate bioassays, we therefore tested whether monoketones or diketones would elicit alarm behaviour. We used microscale silica chromatography to separate these pheromone components. A glass capillary tube (1.2 mm inner diameter, 110 mm long) was filled with 55 mg of 400 mesh silica gel (Haiyang, Qingdao). The extracts were concentrated down to 20 µl (see above) and added to the silica column, which was then successively washed with 55 µl hexane, 260 µl ethyl acetate/hexane and 260 µl ethyl acetate/hexane (1:1, v/v). Fraction 1 (monoketones) eluted at the sixth tube (130 to 195 µl 4:1 acetate/hexane v/v) and fraction 2 (diketones) eluted at the ninth tube (1:1 acetate/hexane, v/v). The components in each fraction were confirmed with GC analysis. Fraction 1 contained monoketones and fraction 2 contained undeca-2,10-dinone (identified by comparison with an authentic standard) and an unknown undecene-2,10-dinone (Table 1).

We also video recorded hornet responses to these different fractions. We added 4 eq of the test fraction (or an equivalent volume of pure solvent in control trials) to a piece of clean filter paper placed <1 cm behind a leaf and presented the sample 30 cm from the nest entrance for 3 min, as in the whole venom bioassay. We measured aggression by counting the total number of hornets that tried to attack the sample. We conducted one trial per colony per day (detailed sample sizes in Table 2).

We tested hornet responses to four of the major or most volatile identified venom volatiles: heptan-2-one (most volatile), non-8-en-2-one (most volatile), undecan-2-one (major component) and nonan-2-one (major component). We used the same 3 min bioassay as above, but tested hornet aggression responses to 0, 10, 100, 1000 and 10,000 ng of pure synthetic standards. We conducted one trial per colony per day (detailed sample sizes in Table 2).

We extracted volatile pheromones with solid phase microextraction (SPME). After comparison of fibres, we selected a 65 µm PDMS/DVB blue fibre (Supelco, CA) because it rapidly adsorbed the most volatiles. Using clean glass 5 ml vials, we collected the headspace volatiles of attacked workers from three different nests or a dissected and crushed worker venom gland for 30 min (samples sizes given in Table 2). To collect alarm volatiles from a living worker, we gently caught it with a cotton sieve in front of its nest. It was briefly cold-anesthetized on ice for 2 min and transferred into a clean collection vial that was immediately capped. After it revived, the PTFE-lined cap was penetrated with a needle and this needle was used to disturb the hornet by lightly touching (without piercing) its thorax a total of 10 times over 30 s. During this process, the hornet exhibited alarm behaviour and began to exude venom from the tip of its stinger. The hornet was then confined to the bottom of the vial with the needle, and the SPME fibre was introduced in a sleeve through the cap. We thereby only obtained volatiles. The fibre had no direct contact with the needle, the hornet, cap or vial walls. For chemical analysis, each SPME fibre was desorbed in the GC injection port at 250°C.

In addition, pheromone was extracted from venom gland contents for use in some of the bioassays. Workers were anesthetized in a freezer and then their venom sacs were dissected out and extracted with hexane. We placed 10 glands in 100 µl hexane in a clean glass vial. After 2 h, the solvent and two washes of 50 µl of hexane were transferred to a 250 µl micro-vial insert tube. The extract was concentrated 10× to a final volume of 20 µl with a gentle nitrogen flow for all bioassays, compound identification and compound quantification. All extracts were kept at −20°C until use.

We purchased commercially available heptan-2-one (CAS 110-43-0, Sigma-Aldrich), nonan-2-one (CAS 821-55-6, J&K, Beijing, China), undecan-2-one (CAS 112-12-9) and other reagents (TCI, Tokyo, Japan). The non-8-en-2-one was synthesized by condensation of 6-bromo-hexene and ethyl acetoacetate in the presence of sodium ethoxide followed by hydrolyzation and decarboxylation. The heptan-2,6-dinone was synthesized via condensation of dibromomethane and ethyl acetoacetate in the presence of sodium ethoxide followed by hydrolyzation and decarboxylation (low yield but detectable for GC-MS identification). Undeca-2,10-dinone was synthesized in the same manner using 1,5-dibromopentane. All synthetic compounds were purified with silica gel chromatography.

SPME extracts and derivatives were analysed by GC-MS, using an HP 7890A-5975C (Agilent, US) with an HP-5ms capillary column (30 m×250 µm×0.25 µm, Agilent, US). The carrier gas was helium flowing at 37 cm s⁻¹. The oven ramp was set as 50°C for 2 min, followed by 5°C min⁻¹ and then 280°C for 10 min. For the quadrupole mass spectrometry, a 70 eV EI ion source was used at 230°C. The mass range scanned consisted of m/z ratios of 28.5–300 at a rate of 2×4 scans s⁻¹. The detection abundance threshold was set to 10. Data were analysed using Chemstation software (Agilent Technologies) and AMDIS (NIST).

We used micro-scale derivatization to narrow down the number of potential compounds by determining possible functional groups in the unknown gas chromatography-electroantennographic detection (GC-EAD) active compounds. To determine if unknown GC-MS peaks with a mass-to-charge ratio (m/z) of 43 were acetic esters, we hydrolyzed 20 µl of supernatant from a pooled extract (10 glands in hexane) by adding 20 µl of 0.1 mol l⁻¹ NaOH and stirring for 30 min. The organic layer was then chemically analysed.

We used NaBH₄ reduction to confirm the existence of ketone groups (Attygalle, 1998): 20 µl of 0.5 mol l⁻¹ NaBH₄/NaOH solution was added to the supernatant obtained from a different extract of 10 glands. This mixture was neutralized with 0.5 mol l⁻¹ HCl, stirred for 30 min, and then analysed with GC.

We used platinum-catalysed reduction to determine if there were rings or olefinic bonds in the compounds. Approximately 0.1 mg of Pt/C catalyst was added to a 10 gland extract in hexane. The extract was stirred under hydrogen for 30 min. After removal of the Pt/C particles by filtration, the solvent was subjected to chemical analysis (Attygalle, 1998).

SPME extracts, solvent extracts and derivatives were analysed using GC with a flame ionization detector (FID). We used an HP-7890B GC (Agilent, US) with FID and splitless injection at 250°C. For GC-FID analysis, an HP-5 column (30 m×320 µm×0.25 µm, Agilent, US) was used with nitrogen flowing at 37 cm s⁻¹ as carrier gas. The oven ramp was set to 50°C for 2 min, then 10°C min⁻¹ to 280°C for 5 min. We used GC-FID quantity-response standard curves to quantify each known compound in a venom gland against pure synthetic standards. For quantification, we used extracts from 15 foragers from three colonies (five foragers per colony). We calculated the linear retention index (LRI) using retention times of C8-C15 n-alkanes analysed under the same GC and GC-MS conditions.

GC-EAD was used to measure the electrophysiological olfactory responses of hornet antennae to volatile sting venom compounds collected by SPME. We used the same instruments and protocol as Wang et al. (2016). In brief, a custom EAD system was coupled to the HP7890B GC. GC conditions were the same as for the GC-FID analysis. An HP-34465A digital multimeter (Keysight) controlled by BenchVue software (Keysight) running on a PC was used to record antennal responses. For electroantennogram (EAG) analysis, the odour preparation was delivered to the antennal preparation with a custom stimulus controller (Wang et al., 2016).

For GC-EAD and EAG, we followed the same capture method used to analyse alarm volatiles (see above) and detached one antenna per hornet (left or right, randomly chosen) at its base with iris scissors. The distal end of the antennae was cut open with scissors to improve signal strength and both ends were mounted between two glass electrodes filled with insect Ringer's solution. The antennal preparation was positioned in a clean and wet air flow (40 cm s⁻¹, room temperature, relative humidity >95%) conducting the odours from the GC column outlet or an odour pipette, as appropriate. Sample sizes for the GC-EAD and the EAG experiments are given in Table 2. We tested the following compounds: heptan-2-one, undecan-2-one, nonan-2-one, and non-8-en-2-one.

To determine the effect of venom quantity and venom fractions on the number of hornets that attacked the target and to analyse the effect of identified GC-EAD active compounds on hornet attacks, we used a repeated-measures general linear model (GLM) with a Poisson distribution, log link, maximum likelihood estimation and an overdispersion parameter. Colony was the repeated measure. We used Dunnett's test to make comparisons corrected for Type I error between bee responses to the blank control and the different compound quantities.

To test for the independence of attacks (whether each attacker added additional alarm pheromone), we ran a univariate repeated-measures analysis of variance of the number of attackers with time (attacks per minute) and colony as factors. If alarm pheromone accumulated during attacks, there should be a significant increase in attacks over the 3 min trial to venom fractions that elicited attacks. The results (see below) suggested that attackers did not add alarm pheromone to our stationary target, perhaps because it did not fight back or struggle like a living target. Each attack appeared to be largely independent of prior attacks.

In all models, we included colony as an effect and used post hoc likelihood-ratio contrast tests. For multiple tests run on the same data, we used Bonferroni corrections, reporting results as significant only if P<α_Bonferroni. All analyses were conducted with JMP Pro v.12.0.1.

In chemical analyses, all volatiles from attacked workers were identical to those from dissected venom glands (Fig. 1A). We identified the same 16 major compound peaks (Fig. 1A, Table 1) in the volatiles of all nine attacked hornets. All 15 samples (each a separate GC-MS run) of hornet venom volatiles contained these identical peaks. The venom gland is therefore the source of the alarm pheromone.

Venom extract significantly attracted hornets (quantity effect: L-R χ²₃=92.19, P<0.0001) in all colonies, but some colonies had stronger responses (colony effect: L-R χ²₂=10.01, P=0.01). All quantities ≥0.01 venom gland equivalents (eq) attracted more hornets than the control (contrast tests, L-R χ²₁≥13.61, P≤0.0002<α_Bonferroni=0.017, Fig. 1B).

Chemical analysis of the venom of V. velutina workers revealed 16 major compounds, of which 13 elicited reproducible antennal (GC-EAD) responses (Fig. 3, Table 1). Known structures were confirmed with authentic chemical standards.

Unknown major GC-EAD active peaks 10 and 15 were further analysed via MS interpretation and derivatization (Fig. 2). After hydrolyzation, peaks 10 and 15 were unchanged, confirming that these compounds did not contain ester structures. Both peaks disappeared after NaBH₄- and Pt-catalysed reduction, indicating the presence of ketone and olefinic structures.

The compound corresponding to peak 10 had an m/z of 168. The ratio of its isotope peak at m/z 169 was 12.20%, indicating a formula of C₁₁H₂₂O, with a ring double bond (RDB) value of 2. A base peak with m/z 43 resulted from the loss of CH₃C=O⁺. The existence of 2-ketone groups with γ-H was suggested by m/z 58 resulting from a McLafferty rearrangement. Thus, the compound is probably an undecen-2-one.

The compound corresponding to peak 15 had m/z 182, indicating a formula of C₁₁H₂₂O₂ with a RDB value of 3. This compound had a base peak with m/z 43, and a characteristic ion with m/z 58 from a McLafferty rearrangement, indicating 2-ketone groups with γ-H. Because there were two oxygen atoms, we hypothesize that this compound has two 2-ketone groups and an olefinic double bond. Thus, peak 15 probably corresponds to an undecene-2,10-dinone. The peak areas of undecan-2-one (peak 13) and undecane-2,10-dinone (peak 16) increased after Pt-catalysed reduction, suggesting the presence of ketones.

The venom ketones can be separated into two fractions consisting primarily of monoketones (fraction 1) and diketones (fraction 2, Fig. 4A). There was an overall effect of treatment type on the number of attacks that a target received (L-R χ²₃=86.60, P<0.0001, Fig. 4B). There was a significant effect of colony (L-R χ²₂=106.80, P<0.0001) because some colonies had stronger responses. However, for each colony, the overall response pattern was consistent. All fractions and their combination received significantly more attacks than the control (contrast tests, L-R χ²₁≥48.90, P=0.00001<α_Bonferroni=0.017, Fig. 4B).

Each attacking hornet could potentially deposit additional alarm pheromone on the target. If so, then the number of attacks should increase over time on each fraction and the combination of both fractions. However, the number of attacks per minute did not increase over the 3 min trial (no effect of time for all fractions separately or in combination: F_2,22≥2.75, P≥0.09).

Hornet antennae did not respond strongly to four major identified compounds that we tested (heptan-2-one, undecan-2-one, nonan-2-one and non-8-en-2-one). The EAG response difference threshold, the lowest quantity that elicited a statistically different response from exposure to the blank control was 1000 ng for all these compounds (Dunnett's test, P<0.05, Fig. 5A). With one exception, 1000 ng is far greater than the quantity released by a single hornet venom gland, suggesting a low biologically relevant sensitivity. However, nonan-2-one does occur at 852 ng per hornet (Table 1).

There was a significant effect of compound type (L-R χ²₃=18.51, P=0.0003), colony (L-R χ²₂=86.00, P<0.0001, but similar trends when colonies examined separately), quantity (χ²₄=54.39, P<0.0001), and the interaction compound type×quantity (L-R χ²₁₂=27.73, P=0.006) on the number of attacking hornets (Fig. 5B). For non-8-en-2-one there were no significant contrasts (L-R χ²₁=0.70, P=0.40). However, contrast tests revealed a quantity effect in heptan-2-one (0 vs 10⁴ ng, L-R χ²₁=6.87, P=0.009), nonan-2-one (0 vs 10⁴ ng, L-R χ²₁=8.19, P=0.004) and undecan-2-one (0 vs 10⁴ ng, L-R χ²₁=13.39, P=0.0003 and 0 vs 10³ ng, L-R χ²₁=6.07, P=0.01). For all of these contrast tests, α_Bonferroni=0.025. Thus, these four compounds, presented in isolation, did not elicit a strong attack response because levels far greater than those found in one venom gland are needed to elicit attacks.

Alarm pheromones play a key role in social insects, but their function in social wasps and hornets remains poorly understood compared with other social insects. We provide the first evidence that V. velutina, a widespread and invasive species, uses an alarm pheromone. Our analyses of volatiles produced by alarmed hornets revealed that this alarm pheromone is produced by the sting gland, and increasing quantities of sting gland extract increased aggressive attacks. We then used GC-MS analysis and authentic standards and identified 13 of the 16 major compounds found in these volatiles, all ketones. Using micro-scale derivatization, we narrowed the possibilities for two compounds and hypothesize that they are a type of undecen-2-one and a type of undecene-2,10-dinone. Worker hornets had antennal responses to four of the identified compounds (non-8-en-2-one, heptan-2-one, nonan-2-one, and undecan-2-one) and exhibited aggression to three of these compounds (heptan-2-one, nonan-2-one and undecan-2-one). The alarm pheromone fractions that we tested (monoketones, diketones and their combination) all significantly elevated attacks (Fig. 2B). Thus, as in other Vespidae species (Akre et al., 1982; Ali and Morgan, 1990; Downing, 1991; Landolt and Akre, 1979; Ono et al., 2003), ketones are important components of hornet and wasp alarm pheromones, which are commonly volatile venom components.

Two of the major components (non-8-en-2-one and nonan-2-one) that we tested did not elicit strong responses at biologically relevant doses (one hornet equivalent). However, nonan-2-one did elicit a significant antennal response at 1000 ng (Fig. 5A) and occurs at 852 ng/sting gland. Nonan-2-one appeared to trigger more hornet attacks at 1000 ng, but this was not significantly different from responses to the blank control. There was a clear trend with increasing quantities, but only 10,000 ng significantly elevated attacks (10 hornet equivalents, Fig. 5B). Thus, further studies of nonan-2-one, the remaining 12 compounds, and combinations of these may narrow down the key attack triggers. Synergistic interactions between compounds may be important.

Saslavasky et al. (1973) reported that multiple ketones elicit strong alarm responses in Vespa orientalis, an Asian species related to V. velutina. Recently, Jimenez et al. (2016) demonstrated that ketones in Dolichovespula maculata venom also acted as an alarm pheromone, with some ketones eliciting attacks directed at a target, much as we observed. With respect to the V. velutina ketone, nonan-2-one, we observed significantly increased attacks at 10 eq, which is similar to the effects observed by Jimenez et al. (2016) with natural venom or candidate synthetic ketones tested at 5 eq (5 venom sac extract equivalents). We also tested lower concentrations of natural venom extracts and found exceptionally high sensitivity, with hornets attacking the target significantly more often at only 0.01 eq (Fig. 1B).

Because we used a single target in our assays, not multiple ones (Jimenez et al., 2016), we cannot distinguish between targeting (spatially localized) versus the general alarming (not tightly localized) effects of individual compounds. However, given the relatively low behavioural responses of hornets to most of our pure synthetic compounds, even to nonan-2-one, compared with natural venom, further investigations of the remaining 12 venom compounds are warranted. It is possible that a less abundant component in venom (Table 1) could elicit the observed responses to 0.01 eq of natural venom. Alternatively, a combination of compounds may be required.

Recent findings suggest that persistent (lower vapour pressure) components in honey bee venom pheromone (sting alarm pheromone) play an important role because they provide longer-lasting information (Wang et al., 2016). For V. velutina, dimethylnonen-2-one (peak 9), the unknown undecen-2-one (peak 10) and the unknown undecene-2,10-dinone (peak 15) are therefore potentially components that can mark attack targets. These three compounds have lower vapour pressure and can persist to mark a predator or indicate danger.

Our results contribute to a growing body of evidence that alarm pheromones in social insects are closely linked to toxic or venomous components, as predicted by the evolutionary ritualization hypothesis. Alarm pheromones should easily evolve from volatiles associated with defensive chemical weapons, like venoms, because these are associated with attack and defence by receivers and because this information enhances colony fitness by rallying nest mates to attack. An unresolved evolutionary question is the function of the complex volatile blends that we see today. In some cases, these mixes may be metabolic ‘spandrels’ (byproducts of a common metabolic pathway). Multiple compounds may also provide more reliable information in noisy chemical environments. Finally, individual components may be selected based upon volatility to provide information over time. A more complete understanding of the functions of individual components and mixtures is required, although it is interesting that the use of venom in the alarm response is largely parallel in ants, bees and wasps.

We thank the Central Laboratory at Xishuangbanna Tropical Botanical Garden for their assistance in chemical analysis.