Resisting majesty: Apis cerana, has lower antennal sensitivity and decreased attraction to queen mandibular pheromone than Apis mellifera

In highly social bees, queen mandibular pheromone (QMP) is vital for colony life. Both Apis cerana (Ac) and Apis mellifera (Am) share an evolutionarily conserved set of QMP compounds: (E)-9-oxodec-2-enoic acid (9-ODA), (E)-9-hydroxydec-2-enoic acid (9-HDA), (E)-10-hydroxy-dec-2-enoic acid (10-HDA), 10-hydroxy-decanoic acid (10-HDAA), and methyl p–hydroxybenzoate (HOB) found at similar levels. However, evidence suggests there may be species-specific sensitivity differences to QMP compounds because Ac workers have higher levels of ovarian activation than Am workers. Using electroantennograms, we found species-specific sensitivity differences for a blend of the major QMP compounds and three individual compounds (9-HDA, 10-HDAA, and 10-HDA). As predicted, Am was more sensitive than Ac in all cases (1.3- to 2.7- fold higher responses). There were also species differences in worker retinue attraction to three compounds (9-HDA, HOB, and 10-HDA). In all significantly different cases, Am workers were 4.5- to 6.2-fold more strongly attracted than Ac workers were. Thus, Ac workers responded less strongly to QMP than Ac workers, and 9-HDA and 10-HDA consistently elicited stronger antennal and retinue formation responses.

These species differences could arise from multiple factors. However, a logical first step is to examine the sensory input (antennal olfactory sensitivity) and an immediate behavioral output, physical attraction and movement towards a component, which will, in turn, expose a worker to higher levels of that compound. Our goals were therefore to compare the antennal responses and retinue formation behaviors of Ac and Am workers to identical presentations of major-component QMP blends and individual QMP components. We focused on the major QMP components that are known to play a key role in retinue formation in both species 7,16 .

Results
Am had stronger antennal responses than Ac. To measure antennal sensitivities, we used electroantennograms (EAG), which are commonly used to measure olfactory stimulus sensitivity in honey bees 18 and have been employed to measure worker responses to Am QMP 1 . In both species, the slope and shape of EAG antennal responses to different compounds were similar, and exhibited a fast recovery to baseline (Fig. 1). However, there were differences in the peak magnitudes of responses to different compounds.
We tested responses to major QMP blends that contained the most abundant components found in the QMP of each species (Table 1) and individual compounds. As predicted, Am had significantly stronger antennal responses to the major QMP blends and to some individual compounds than Ac. Ac never had significantly stronger responses than Am (Fig. 1). For the blends, there were significant effects of bee species (F 1,32 = 51.95, P < 0.0001) and compound (F 1,34 = 21.95, P < 0.0001), but no significant interaction of bee species*compound (F 1,34 = 0.41, P = 0.53). Colony accounted for 17% of model variance. Am workers had a significantly higher EAG response to the Am QMP blend than to the Ac QMP blend (Tukey's HSD test, P < 0.05). Similarly, Ac workers had a stronger response to the Am QMP blend than to their own Ac QMP blend (Tukey's HSD test, P > 0.05). Ac workers had consistently lower EAG responses than Am workers (Fig. 1A).
We next focused on testing EAG responses to individual compounds. In the full model (Table 2), there were significant effects of compound, dose and multiple interactions (P ≤ 0.004) since species responded differently to different compounds and doses (Fig. 1B). We therefore next considered each compound separately, using a Bonferroni-corrected alpha = 0.025 (k = 2) for all tests conducted on these data.
Am was more strongly attracted to individual QMP compounds than Ac. We measured the attraction of individual QMP compounds with a retinue bioassay. We counted the number of Am and Ac workers that moved across a comb towards the test compounds (Fig. 2). Three compounds (9-HDA, HOB, and 10-HDA) attracted significantly more Am than Ac workers. Ac was never significantly more attracted than Am (Fig. 2).  38 with the exception of HVA (data from Slessor et al. 13 . Per compound, the mean of both species is shown because these quantities were used for the comb bioassay experiment. P-values are from Univariate ANOVA tests reported in Tan et al. (2009) and show that the compounds are found at similar levels in both queens of both species 38 . For HVA, it is not possible to compare levels statistically because only a single data point is available for Am. For 9-HDA, there was a significant effect of dose (Likelihood-Ratio χ 2 1 = 10.05, P = 0.007), no effect of species (L-R χ 2 1 = 2.33, P = 0.13), but a significant interaction (L-R χ 2 2 = 12.08, P = 0.002) because significantly more Am than Ac workers were attracted to the 100 μ g dose (L-R χ 2 1 = 19.30, P < 0.00001). For HOB, there were significant effects of dose (L-R χ 2 2 = 1.41, P = 0.50), species (L-R χ 2 1 = 4.08, P = 0.04), and the interaction dose*species (L-R χ 2 2 = 11.33, P = 0.004) because significantly more Am than Ac workers were attracted to the 100 μ g dose (L-R χ 2 1 = 6.95, P = 0.008). Higher doses of 10-HDA attracted more bees: there was a significant effect of dose (L-R χ 2 2 = 8.36, P = 0.015) but no significant effect of species (L-R χ 2 1 = 2.33, P = 0.13) or the interaction dose*species. (L-R χ 2 2 = 4.01, P = 0.13). However, graphical inspection of the data suggested the following contrast tests: significantly more Am than Ac workers were attracted to the 100 μ g dose (L-R χ 2 1 = 12.81, P = 0.0003) and the 10 μ g dose (L-R χ 2 1 = 7.78, P = 0.005).

Discussion
Two main results emerge from these experiments. First, Am workers consistently exhibited higher antennal sensitivity than Ac to the main components of QMP and the QMP blends. Am workers responded more strongly to the QMP blend of their own species, and Ac showed uniformly lower antennal responses. Ac did show a slightly higher response to the Am as compared to the Ac QMP blend, perhaps because the Am blend contained HVA but the Ac blend did not. HVA is not found in Ac QMP, but Ac workers show antennal responses to this compound. Second, Am also consistently showed a stronger retinue attraction than Ac to these QMP components. Such retinue attraction is important because it plays a major role in colony life, mediating care of the queen and helping to disperse QMP, which is a primer and a releaser of multiple important colony activities 7 . In our individual compound tests, Ac workers never had higher antennal responses than Am to any compounds, and 9-HDA, 10-HDA, and 10-HDAA elicited higher amplitude antennal signals from Am than Ac workers (Fig. 1). The retinue bioassay likewise showed that only Am was more attracted to individual compounds (9-HDA,10-HDA, and HOB) than Ac (Fig. 2). The match between antennal sensitivity and behavioral attraction was not exact. Both 9-HDA and 10-HDA elicited strong antennal responses and attracted more workers. However, 10-HDAA elicited a stronger Am antennal response, but did not increase worker attraction. Likewise, HOB increased Am worker attraction but did not result in a higher antennal response. These differences likely arise from the role that higher order neural processing plays in worker attraction to QMP. However, the overall results of the retinue bioassay and the EAG measurements matched: Am consistently showed stronger responses than Ac.
It is possible that Ac and Am worker responses may depend upon the full blend of QMP compounds, including those found only in trace quantities. However, our goal was to provide comparative data by testing bees with identical doses of the same major compounds. Moreover, our Ac QMP blend contained the key compounds (9-ODA, 9-HDA, and HOB) that Plettner et al. 16 found were sufficient to elicit a full Ac worker retinue response 16 .
What could cause these antennal sensitivity differences? They may have arisen from saturation differences in antennal responses. However, our major QMP blend tests (Fig. 1A) showed significant species differences at the level of one queen-equivalent. At this biologically relevant level, Am workers consistently had 1.8-fold higher antennal responses than Ac workers. Bees may also have perceived the compounds presented in isolation differently from a full QMP blend. Our results do not support this interpretation because the responses to the blends (Fig. 1A) are similar in amplitude to the sum of responses to compounds individually presented (Fig. 1B).
Ac antennae may be less sensitive to some QMP compounds (9-HDA, 10-HDAA, and 10-HDA) than Am antennae. In honey bees, differences in EAG antennal responses are associated with multiple morphological and electrophysiological properties of antennae 19 . The size and surface area of the antennae in Ac workers and Am workers appear to be identical, but the distributions of sensory hairs on the antennae are significantly different 20 . Four classes of olfactory sensilla (placodea, trichodeum types A and B, basiconica) are significantly more abundant on Ac than on Am worker antennae 20 . However, Am workers have a greater abundance of other sensilla, such as sensilla campaniformia, s. coeloconica, s. ampullaca, and s. chaetica, than Ac workers 20 . Insect sensilla abundance may affect antennal olfactory sensitivity 21,22 , although more studies are required to demonstrate this  The mean per-trial difference in the number of bees that were attracted to the treatment as compared to the control is shown. Significantly more Am than Ac workers were attracted to higher quantities of 9-HDA, HOB, and 10-HDA (*). Means, standard errors, and significant contrast tests are shown. Compounds are grouped into three rows, corresponding to the average amounts found in one queen equivalent of QMP (see Table 1).
in honey bees. The number of different sensory neurons per sensilla also influences sensitivity, and the EAG response is a measure of summed responses of all chemosensory neurons in all antennal sensilla 23 . Odorant receptors may differ between Am and Ac. Major groups of chemoreceptor genes include odorant receptors (Or's) 24 . In Am, the number of odorant receptors has been estimated at 160-170, which is approximately equal to the number of glomeruli in the Am antennal lobe 24,25 . The Am receptor, Or11, has been functionally characterized as responding to 9-ODA 26 . In contrast, markedly fewer odorant receptors (119) have been characterized in the Ac genome 27 . Am may therefore have a better ability to detect odors than Ac in a variety of contexts. Finally, neuromodulators such as serotonin, octopamine and dopamine can alter the sensitivity of invertebrate chemosensory neurons 28,29 and higher level processing should play a role. The evidently greater antennal sensitivity of Am as compared to Ac workers may derive from multiple factors.
Because QMP is distributed throughout a colony, average colony sizes are also relevant. Ac colonies contain fewer workers than Am colonies 30 . In our experiments, average Ac and Am colonies consisted of approximately 15,000 and 20,000 bees, respectively. Given that the QMP of healthy, fertilized queens of each species contains, on average, the same amount of 9-ODA, 10-HDA, and HOB (Table 1), one might expect that Ac workers would be exposed, per bee, to higher levels of QMP than Am workers. However, even at higher levels QMP component doses, Am workers consistently exhibited stronger antennal responses (Fig. 1) and greater attraction (Fig. 2) than Ac workers.
The greater attraction of Am as compared to Ac workers towards QMP blends and their components may have implications beyond retinue formation and suggest future studies. In both species, laying workers have higher amounts of QMP components than non-laying workers 31 . Ac workers also seem slightly more tolerant of worker-laid eggs as compared to Am workers 12 . Thus, reduced Ac attraction to QMP compounds and blends, as compared to Am, may reflect an increased tolerance of workers with developed ovaries. It is also possible that decreased Ac worker sensitivity to QMP could partially account for the higher ovarian activation levels seen in Ac workers in queen-right colonies. However, testing these hypotheses will require a different set of experiments that examine worker ovarian development and the effects of long-term exposure to major QMP components.
What are the evolutionary reasons behind these differences? A review of social insect pheromones suggests that social insect queen pheromones have evolved to provide an honest indication of queen quality rather than as coercive agents that chemically sterilize workers 16 . Tan et al. 17 provided evidence for such honest QMP pheromone signaling in Ac 31 . Given this, we hypothesize that the higher levels of ovarian development in Ac as compared to Am workers reflects a beneficial adaptation that facilitates colony fitness. For example, queenlessness may occur more often in Ac than in Am because swarming and absconding are more common in Ac than in Am 32 . However, if this is an honest queen-worker signal, Ac queens could simply have evolved a lower level of QMP compounds rather than workers evolving a higher response threshold. Because both Ac and Am have similar levels of most QMP compounds (Table 1), we therefore wonder if other constraints, such as worker evaluation of and regulation of replacement queens 33 , are at play.

Materials and Methods
Study site and colonies. The experiments were conducted from May 2015 to August 2016, at Yunnan Agricultural University, Kunming, China. Three queenright Ac colonies and three Am colonies were kept in standard Langstroth hives. Each hive consisted of four frames covered with adult workers, two frames of brood and two frames of honey and pollen.

Exp. 1: EAG.
Over a wide variety of compounds, including 9-ODA, the electrophysiological responses of Am olfactory antennal cells increases up to 4 days of age after adult emergence and then remains fairly constant for the rest of adult life 34 . For example, Pham-Delegue et al. (1993) tested the EAG response of Am workers to Am queen-head extracts or to synthetic QMP (9-ODA, 9-HDA, HOB, and HVA in natural proportions) and found no significant differences in the magnitude of EAG responses for adult bees over the range of 2-21 days of age 35 . Similarly, Allan et al.(1987) tested the EAG responses of Am workers of different ages (1-60 days of adult age) to two components of Am QMP (9-HDA and 9-ODA) and showed that EAG responses increased in magnitude with age, but were roughly similar for bees between 6-40 days, particularly for 9-HDA 36 . Our approach assumes that, in Am and Ac workers of similar age, the same EAG response indicates a similar sensitivity. This assumption may not be correct, but we believe that this is a reasonable initial approach given that we have no a priori expectation of differences or the direction (higher or lower sensitivity) of any putative differences.
We used plastic boxes to carefully capture adult workers (returning foragers) of both species as they returned to their nest entrances. Based upon the age of first foraging in these species, these bees were likely more than 22 days of adult age. We chose foraging-age bees because one of the few other comparative studies on QMP in Am and different Asian honey bee species, including Ac, also used bees of foraging age 37 .
We tested worker EAG responses to a blend of the major components found in the QMP in each species. We used the average amount of each compound per species (Table 1). The major-component Ac QMP blend therefore contained 243 μ g 9-ODA, 33 μ g 9-HDA, 31 μ g HOB, 1.3 μ g 10-HDA, and 0.9 μ g 10-HDAA, and the Am QMP blend contained 237 μ g 9-ODA, 67 μ g 9-HDA, 26 μ g HOB, 1.2 μ g 10-HDA, 4 μ g 10-HDAA, and 2 μ g HVA 13,37,38 . Thus, the Am QMP blend contained HVA, which is not found in the QMP of Ac (Table 1). Each blend was diluted in dichloromethane (Aladdin, CN), which is highly volatile and elicits a minimal EAG response 39 , and presented at a quantity equivalent to one queen. To compensate for individual variation, each bee was exposed to three treatments: the blank control (dichloromethane), the Ac QMP blend, and the Am QMP blend. We used workers from three Ac colonies (n = 18 different bees: 6 bees from each colony) and three Am colonies (n = 18 different bees: 6 bees from each colony).
We also measured EAG responses to the individual QMP components that we used in our blends. The mandibular glands of a mated, egg-laying queen contain approximately 200 μ g 9-ODA, which comprises about 60% of the total secretion in Ac and Am queens 13,37 . We therefore used doses ranging from 1 ng to 100 μ g. Each compound was diluted in dichloromethane and presented as the following concentration series: 0 ng/μ l (solvent control), 1 ng/μ l, 10 ng/μ l, 100 ng/μ l, 1000 ng/μ l, 10,000 ng/μ l, and 100,000 ng/μ l. Each sample was applied to a filter paper strip. The paper strip was then placed inside the odor pipette after the solvent was evaporated for 10 s. This should not have resulted in appreciable evaporation of the test compounds because all tested QMP components have vapor pressures below 0.0001 kPa. We used multiple bees from three Ac colonies (n = 106 different bees: 35, 35 and 36 from colonies 1, 2, and 3 respectively) and three Am colonies (n = 103 different bees: 34, 34, and 35 from colonies 1, 2, and 3 respectively).
To record the antennal response, we gently removed one worker at a time from the box in which it was captured, cut off a randomly-chosen antennae (left or right), and placed it between glass electrodes filled with honeybee Ringer's solution 39 . Blends were presented in the order described above. Individual compounds were tested in an ascending concentration series on each antennal preparation. In the QMP blend tests, we tested each antenna with both blends. However, in the individual compound tests, we only tested one compound type per antenna. We waited for 30 s between stimulations to provide sufficient recovery time and then provided a 3 s stimulus. We measured the baseline-peak amplitude (mV) 39 . The EAG recording system was the same as described in Wang et al. 39 . The antennal preparation was placed 1 cm away from the outlet of a odor pipette (1 cm inner diameter, 15 cm long) that provided the test odor by combining a clean and wet continuous air flow (15 ml/s, 90% relative humidity) and a pre-filtered and wet pulsed air flow (5 ml/s, 90% relative humidity). The test odor stimulus was presented for 3 s. All measurements were conducted at 25 °C. To record the antennal responses, a modified EAG amplifier fed the amplified (21X) and filtered EAG signal into an HP34465A Digital Multi Meter (Agilent, USA) and BenchVue software (Keysight, USA) running on a PC for signal recording.
For the chemical standards, commercially available HOB, HVA, and 10-HDAA were obtained from the Aladdin Reagent Database Inc. (Shanghai, China). Isomerically pure 9-ODA and 10-HDA were synthesized using the Doebner-Knoevenagel condensation method 40 . The 9-HDA was synthesized by selective reduction of the keto group of 9-ODA with NaBH 4 41 .

Exp. 2: Behavioral attraction (retinue formation).
We conducted a retinue bioassay to test the attractiveness of six major QMP components. We individually tested 9-ODA, 9-HDA, HOB, 10-HDAA, 10-HDA, and HVA, each at a quantity which was an average of the amount found in queens of both species (see Table 1).
For each trial, we collected 30 foragers (see Exp. 1 methods) from the entrance of the focal colony. We used CO 2 applied for 5 s to briefly anesthetize the bees, and then placed them in the center (4 cm diameter circle) of a processed beeswax comb foundation (41 cm × 19.5 cm) inside a box. Prior studies with Am workers used a similar approach and counted the number of workers entering an elliptical space around the test lure 16,42 .
Centered and separated by 20 cm apart, we fastened two clean pieces of filter paper (each 0.8 cm × 2 cm) with insect pins (Fig. 2A). One paper was the solvent control (1 μ l of dichloromethane) and the other was the treatment compound at different doses (1 μ g, 10 μ g, or 100 μ g in 1 μ l of dichloromethane). We then counted the number of bees entering a circle circumscribing each of the paper strips over a 5 min trial ( Fig. 2A). Based upon preliminary trials run for 20 min, we found that maximal choice was achieved within 5 min. We used new comb foundation, filter papers, and pins for each trial. Control and treatment positions were alternated between trials to avoid potential side bias. Each trial tested one compound at one dose and used a different set of workers. We conducted four replicates of each condition with each colony and used three colonies of Ac and three colonies of Am.
The beeswax foundation was commercially purchased and was produced from the wax of Am colonies melted at a minimum of 45 °C and maintained in a liquid state for an extended duration. Before use, this foundation had also been sitting at room temperature (21 °C) for over one year. Because of this processing, high levels of speciesand colony-specific volatiles were likely not retained. This foundation therefore provided a neutral, yet more natural base upon which we could conduct our bioassay. Moreover, any remaining species-specific volatiles would have been dispersed throughout the wax foundation and therefore should not have affected attraction to the test compound as compared to the solvent control. Finally, for half of the test compounds (9-ODA, 10-HDAA, and HVA), Am showed the same level of attraction as Ac, suggesting that the wax foundation did not bias our results in favor of Am.
Statistics. Because bees exhibited variance in their individual baseline response to control exposure (0 μ g), we calculated a rectified (corrected) response, obtained by subtracting the control response (CR) from all other responses. We used JMP Pro v12.0.1 for all tests. All models met parametric assumptions, as determined by analyses of residuals.
We first used Repeated Measures Analysis of Variance (ANOVA, REML algorithm) to analyze the effects of compound, dose, species, and all interactions on the log-transformed rectified response. In this model, colony and bee ID were random effects, and all other effects were fixed. Each bee was exposed to different doses of only one compound type, and thus we nested bee identity within compound. Because there were significant interactions between species and compound and between species and dose, we then analyzed each species separately to explore these effects in detail. We used limited post-hoc Least-Squares (L-S) Means Contrast tests to make comparisons between the responses of differences. We applied a Sequential Bonferroni correction (k = 2) and alpha = 0.025 to these data and reported non-significant results as NS.
To analyze the comb bioassay results, we calculated the difference between the number of bees attracted to the treatment filter paper as compared to the blank control (∆ number of bees). We first tested for potential colony effects by running a General Linear Model (GLM, Poisson distribution, Identity link, Maximum Likelihood Estimation, Overdispersion corrected) for each species with colony as a fixed effect. Given that we found no significant colony effects (Likelihood-Ratio χ 2 2 ≤ 3.50, P ≥ 0.17), we then pooled the colony data for each species and ran the same GLM model with fixed effects for dose, species, and the interaction dose*species. Based upon data inspection, we used a limited number of Likelihood Ratio (L-R) Chi-square contrast tests to compare the attraction between the two species.