Similarity in Flight Behaviour Between the Honeybee Apis Mellifera (Hymenoptera: Apidae) and its Presumed Mimic, the Dronefly Eristalis Tenax (Diptera: Syrphidae)

The mimicry of honeybees (Apis mellifera) by droneflies (Eristalis spp.) has been fooling humans for thousands of years (Atkins, 1948). The important question in considering whether it is a true case of Batesian mimicry, however, is whether it fools their predators, which are thought to be birds. Eristalis spp. are not considered to be the best visual mimics of honeybees; other British hoverflies such as Mallota cimbiciformis (Fallén), Criorhina asilica (Fallén) and Brachypalpus laphriformis (Fallén) are thought to be better (Stubbs and Falk, 1983; Howarth et al., 2000), while Morgan and Heinrich (1987) even categorise Eristalis spp. as non-mimetic. Previous work, however, found that, on a range of flowers throughout the season, the foraging behaviour of Eristalis spp. was more similar to that of Apis mellifera than to that of other more closely related flies (Golding and Edmunds, 2000), and this was confirmed by the general impression gained from watching them in the field. There were similarities between droneflies and honeybees in the times spent sitting on flowers and in the times flying between flowers, suggesting that droneflies may be behavioural mimics. This study investigates this possible flight mimicry further.

Studies of flight in hoverflies have been concerned mainly with aerodynamics (Ellington, 1984), with flight mechanisms, wing design and kinematics (Ennos, 1987, 1988, 1989) and with other behavioural aspects, such as the mechanisms by which hoverflies compute interception courses (Collett and Land, 1975) and manage to return to exactly the same spot (Collett and Land, 1978). Flight mimicry has not been studied specifically.

Mimetic resemblance of flying insects has, however, been observed in flower-visiting beetles of the genus Acmaeodera, which look very much like hymenopterans in flight (Silberglied and Eisner, 1969). Other examples include the mimicry between Lebia vittata (Carabidae) and Disonycha scapulari (Chrysomelidae), in which the chrysomelid flea beetle has evolved an effective escape mechanism so that a potential predator gets zero reward (Lindroth, 1971). Hespenheide (1973) described a case of Müllerian mimicry in which agile flies and their beetle models act in a similar way. Examples such as these have, however, been criticised by Brower (1995), who maintains that evidence for unpalatability of the models is often anecdotal. Hoverflies regularly appear in the diet of birds, e.g. swallows (Hirundo rustica) (Kozena, 1979), swifts (Apus apus) (Lack, 1956), house martins (Delichon urbica) (Bryant, 1973), wagtails (Motacilla spp.) (Davies, 1977), flycatchers (Ficedula spp.) (Bures, 1995), magpies (Pica pica) and tree sparrows (Passer montanus) (Kristin, 1986), and are therefore clearly palatable. Hoverflies have been referred to anecdotally as having bee-like flight (Wickler, 1968), bumblebee-like flight (Howarth, 1998) and lazy wasp-like flight (Azmeh, 1999). Morgan and Heinrich (1987) observed that the mimicry of many of the hoverflies they studied appeared to be most accurate during flight. They showed that hoverflies (including Eristalis spp.) were able to warm up using behaviour such as basking or ‘shivering’ which might allow them to behave like their endothermic models. These observed similarities in flight behaviour, however, have never been quantified.

Chai and Srygley (1990) and Srygley (1994) suggested that predation has selected for increased flight speed and manoeuvrability in palatable butterflies, whereas the morphology of distasteful species compromises flight performance. Srygley and Chai (1990) discuss the difficulties imposed on Batesian mimetic butterflies in that they may need to compromise their flight ability to mimic their unpalatable models. This may well be true for droneflies mimicking the flight of honeybees. Droneflies are generally regarded as having superior flight agility compared with hymenopterans because of the position of their centre of body mass (CM_body) (Ellington, 1984), their use of inclined-stroke-plane hovering and their apparent ability to move the aerodynamic force vector independently of the stroke plane (Ennos, 1989). Size is unlikely to be a factor because droneflies and honeybee workers are similar in size and body mass (Y. C. Golding, unpublished data), although female Eristalis spp. tend to be slightly larger than males.

Locomotor mimicry has been most clearly demonstrated in mimetic butterflies of the genus Heliconius in which it is achieved by morphological (Srygley, 1994), kinematic (Srygley, 1999; Srygley and Ellington, 1999a) and aerodynamic (Srygley and Ellington, 1999b) convergence. The mimics are more similar to each other than to other closely related species in their wing-beat frequency and the degree of asymmetry in wing motion, and it has been suggested that this is the first clear example of a mimetic behavioural signal for a flying insect (Srygley, 1999; Srygley and Ellington, 1999a,b). These behavioural characteristics are impossible for the human eye to distinguish, but they may be seen by birds. It is unlikely, however, that hoverflies use such strategies; the wing-beat frequencies of butterflies are far easier to see because the wings are large, conspicuously coloured and beat much more slowly than those of hoverflies and bees, which have transparent wings. The wing-beat frequencies of Eristalis tenax and Apis mellifera certainly overlap, varying between 150 and 200 Hz, but such frequencies are far too high to be detected visually by birds, although the frequency of their buzzing might be heard. These frequencies also overlap with those of a whole range of other hoverflies, large muscids and bumblebees, which also fly using asynchronous muscles (Ellington, 1984; Unwin and Corbet, 1984; Ennos, 1989).

The present study has concentrated on gross aspects of flight, including velocity, flight paths and time spent hovering, which will all be highly visible to potential predators. If there are similarities in flight behaviour in such phylogenetically distant organisms as Eristalis tenax and Apis mellifera, then this would be a more unexpected case of flight mimicry than that described in Heliconius butterflies.

This work was carried out on insects foraging on everlasting daisy flowers (Helichrysum bracteum) at Wythenshawe Park in Manchester, UK. Helichrysum is particularly suitable for this study as it has strong, erect stems and is uniform in height. Insects will therefore tend to move horizontally between flowers, allowing filming from above to capture their two-dimensional movement. In the park, it is grown in a 6 m×2m block, and it attracts a wide range of insects because the flowers provide good pollen and nectar rewards.

An area approximately 60 cm×40 cm within the patch was manipulated to present all flowers at the same height. A ruler was mounted at this level to act as a reference point and scale. A Panasonic video camera (×16 digital zoom; focal length 30–70 mm) was set up 2 m immediately above the patch, and insects were filmed at 25 Hz as they made foraging visits. No correction was required for perspective since all the insects foraged within a zone 10 cm above or below the ruler and moved in a more-or-less horizontal plane. A commentary made onto the audio channel at the time helped later with the identification of different species. The filming was carried out on warm sunny days in July 1997.

The commonest insects were the hoverflies Eristalis tenax (L.) and Syrphus ribesii (L.), the honeybee (Apis mellifera L.) and a muscid fly (Musca sp.). Eleven film sequences were obtained for each species and analysed using a time-lapse video machine. The duration of each sequence varied depending on how long the insect stayed in the field of view, so several trips were recorded for some individuals. The routes (trajectories) taken by the insects between flowers were established by stopping the film frame by frame and marking their position on transparent OHP film, together with the position of the ruler and the flowers they visited.

From the OHP films, the horizontal distance flown between each flower was measured by following the path with a map measurer. The horizontal distance between adjacent flowers visited was also measured, and the ratio of the horizontal distance travelled by the insect to the actual distance between flowers was calculated. This ratio is a measure of deviation from a direct flight path between flowers.

To measure the speed at which the insects approached the flower patch, flew between flowers and departed from the flower patch, the horizontal distance traversed for each flight was divided by the flight time, knowing that each frame represented a time lapse of 0.04 s. Approaching and departing flights were examined over a 10 cm distance from the flower. Time spent hovering was excluded from the calculations of flight velocities. Hovering was identified as episodes during which the insects moved less than 2 mm per frame. The percentage of flight time each individual insect spent hovering during each flight was also calculated.

For each measure, the mean result for each insect was calculated (although, for a few sequences, flights approaching or leaving the field of view or between flowers within the field of view were missing). Results for each species were then tested for normality using the Kolmogorov–Smirnov test. All except hovering time showed no significant difference from a normal distribution. The means for the different species were compared using analysis of variance (ANOVA), with two-sample t-tests (assuming unequal variances) for specific comparisons. Hovering time was analysed using the Kruskal–Wallis test and the Mann–Whitney U-test.

The mean durations of flights between flowers for each species are shown in Fig. 1. A one-way ANOVA on the log-transformed data showed that there were significant differences among the species (F_3,35=14.87, P<0.0001). A two-sample t-test showed that there was no difference between the mimic and the model (t₁₁=0.76, P=0.37), but there were significant differences between the mimic and both Syrphus ribesii (t₁₁=4.71, P=0.0005) and the muscids (t₁₆=3.40, P=0.002).

What accounts for the differences in time taken to fly from one flower to another? It is possible that insects might travel at different speeds, that they might fly to flowers that are different distances apart, that they might take different routes between flowers or that they might spend different times hovering. The velocities at which the insects flew between flowers are shown in Fig. 2. A one-way ANOVA on the data showed that there were significant differences among the species (F_3,35=11.82, P<0.0001): the muscids flew approximately 50 % faster than the other species. Therefore, the times insects spent flying between flowers were partly explained by the speeds at which they flew.

Fig. 3 shows that the percentage of time the different insects spent hovering while flying between flowers explains more of the difference. The proportion of time spent hovering was significantly different among the species (Kruskal–Wallis H_3,35=31.04, P<0.0001). The difference in the proportion of time spent hovering between the mimic and the model was significant (Mann–Whitney U=6, P=0.0326), but small. There were much bigger differences in times spent hovering between the mimic and S. ribesii (U=11, P=0.0043) and muscid flies, which were never seen to hover. S. ribesii hovered much more than the other insects when foraging: it spent approximately 45 % of its flying time hovering. In contrast, the muscids did not hover at all, while Eristalis tenax hovered for 18 % of the time and Apis mellifera for approximately 11 %.

When examining distances travelled, a one-way ANOVA showed that there were no differences between the straight-line distances between the flowers that the insects visited (F_3,35=0.27, P=0.845) (Fig. 4). However, there were differences among species in the directness of their flights. A one-way ANOVA of the ratio of the distance that insects travelled between two flowers to the shortest distance between them was significant (F_3,35=6.32, P=0.002) (Fig. 5). There was no difference between the mimic and the model (t₇=1.57, P=0.16), but there was a significant difference between the mimic and Syrphus ribesii (t₁₁=4.59, P=0.0008), although the difference between the mimic and the muscids was not significant (t₁₇=1.71, P=0.11). These results further help to explain the times spent flying between flowers shown in Fig. 1. There is no difference between droneflies and honeybees in this ratio, as would be expected from the flight times. In contrast, muscids take the shortest, most direct routes between flowers, which correlates with their shorter flight times, while Syrphus ribesii takes the most convoluted routes (see Fig. 6).

Typical trajectories for each insect species are shown in Fig. 6. These show that both droneflies and honeybees often perform loops along their routes. Of the 11 individuals of each species, the honeybees performed 17 loops and the droneflies 10 loops, whereas S. ribesii performed only one loop and the muscids performed no loops at all. The numbers of loops were significantly different among species (Kruskal–Wallis H_3,40=12.86, P=0.005), but median numbers of loops were small (and some of the honeybees performed no loops) so that Mann–Whitney tests could not detect differences between particular species.

The velocities of insects approaching and leaving the flower patch are shown in Fig. 7. A one-way ANOVA showed that there were significant differences among the species (approaching F_3,21=7.49, P=0.002; leaving F_3,39=19.25, P<0.0001). All insects left flowers at a faster speed than they approached them, although the actual speed varied for each species. There was a significant difference in approach speed between Eristalis tenax and Apis mellifera (t₁₃=4.41, P=0.022), but no significant difference in leaving speed (t₉=0.25, P=0.81). The speeds of Eristalis tenax and Apis mellifera were lower than those of the other two species. S. ribesii flew off faster than the other insects, attaining speeds 0.72 s after leaving a flower of up to 0.43 m s⁻¹, although its approach speed averaged only half this. Muscids were able to reach speeds of 0.38 m s⁻¹ after 0.8 s when leaving a flower; this was similar to their approach speeds.

Previous work on insects visiting everlasting daisy flowers (Golding and Edmunds, 2000) showed that the foraging behaviour of dronefly mimics is more similar to that of their honeybee models than to that of other dipterans. A similar pattern was seen here: droneflies and honeybees have shorter flight times between flowers than S. ribesii but longer flight times between flowers than Musca sp. This study, by examining the flight pattern more closely, shows why this takes place. First, there were significant differences in flight speed between flowers, Musca sp. being faster than the other species. There were also two other reasons for this difference in flight times: the species vary in their flight path and in the extent to which they hover. S. ribesii has more convoluted flight paths and hovers more than Eristalis tenax and Apis mellifera, while the muscids have more direct flight paths and never hover.

The convoluted, looping flight of Eristalis tenax (Fig. 6) is surprising behaviour because droneflies are very adept fliers and can easily change direction (Collett and Land, 1975) without altering their stroke plane (Ennos, 1989). Humphries and Driver (1967) have suggested that looped and convoluted flight may be erratic behaviour that is difficult for a predator to follow and that thereby increases an insect’s chances of escaping. They describe this in small flying insects but do not specify which species. Similar tactics have been described and analysed in butterflies (Chai and Srygley, 1990). However, it is not clear why the well-protected honeybee shows this behaviour; perhaps looping is the best way for it to change direction or to orientate (see below). Also, as discussed by Brower (1995), erratic flight as an aversion tactic employed by insects and their Batesian mimics is unlikely to result in long-term learning by a predator. One explanation may relate to the aerodynamic flight mechanisms used when foraging. Both Apis mellifera and Eristalis tenax use a near-horizontal stroke plane while foraging that produces less precise flight than does the inclined stroke plane used by Eristalis tenax males when waiting for a mate (Ennos, 1989). Eristalis tenax is certainly capable of more precise movements, and the fact that it does not use them suggests that mimicry is a possible reason.

The loops in the flight paths of Eristalis tenax and Apis mellifera were not detectable to the human eye, but only became evident when the film was slowed down. The looped flight of honeybees between flowers may be connected with their ability to orientate in relation to the hive using the sun, as described by von Frisch (1967), but the mechanism for orientation in droneflies is different (Collett and Land, 1975). Eristalis tenax may perform loops when returning to its home territory after chasing females or other insects (Collett and Land, 1975). It is difficult to see why droneflies should perform loops when flying between flowers whilst foraging since they are also capable of sudden changes of direction without even altering their body position, performing sharp turns and even turning in their own length (Ellington, 1984; Ennos, 1989). Srygley and Ellington (1999a) present evidence that birds can perceive motion 2–4 times faster than humans. Hence, birds are probably able to detect these subtle differences in flight behaviour. The most likely explanation for the looping flight of foraging droneflies is therefore that, when foraging, they are particularly vulnerable to predation by birds (Dlusski, 1984) so their flight behaviour has been modified at this time to be more similar to that of the noxious honeybees.

The different proportion of time spent hovering is the other aspect of behaviour that helps to explain the differences in flight times shown in Fig. 1. The honeybees and droneflies spent a more similar, and much lower, proportion of their flight time hovering than S. ribesii, although both are well able to hover. Why, then, is there so much difference in the hovering behaviour of the two hoverflies and why are the dronefly and honeybee more similar? One explanation is that S. ribesii is particularly vulnerable to predation when foraging and, being a poor mimic of wasps, it can enhance its mimicry by hovering, thus presenting a blurred image of yellow and black that a bird may avoid, mistaking it for a wasp (Howarth, 1998). Another explanation suggested by Azmeh (1999) is that the colour pattern of poor mimics is aposematic rather than mimetic, advertising efficient escape by flight and zero reward for a possible predator. Support for this hypothesis is the fact that S. ribesii flew away from the flower patch faster than any other insect, which is clearly an advantage in avoiding predation. A third explanation is that hovering S. ribesii are looking for predatory crab spiders before alighting; however, this should apply equally to other hoverflies including Eristalis tenax, which hover much less. A fourth explanation is that droneflies, which have been considered to be poor or imperfect honeybee mimics (Howarth, 1998), have modified their behaviour by hovering less so as to be more similar to their honeybee model.

There is also some evidence of mimicry between droneflies and honeybees in the speeds at which they approached and left the patches, although these data are less reliable since, at these times, the insects were either decelerating or accelerating. Although all four species flew away from the flowers faster than they arrived (Fig. 7), the speeds of the droneflies and honeybees were more similar to each other, and they flew more slowly than the other two species of flies. Why should Syrphus ribesii and Musca sp. fly off at higher speeds? The obvious explanation is that this might help them escape from predators, whereas Eristalis tenax moves more slowly to mimic its model, Apis mellifera.

Dlusski (1984) measured flight speeds as flies flew away from flowers after being disturbed by a model bird and found the average speed to be 1.1 m s⁻¹ with a maximum of 1.63 m s⁻¹, but he found that these speeds could only be attained quickly if the flies were already airborne. Hoverflies are capable of very fast flight once airborne. Collett and Land (1978) measured speeds of up to 10 m s⁻¹ for Eristalis tenax, much faster than any insect measured in this study, but this species may require a considerable distance to generate maximum velocity. The conclusion of Dlusski (1984), after comparing the flight speeds of redstarts (Phoenicurus phoenicurus) and pied flycatchers (Ficedula hypoleuca) swooping down from a nearby tree to a patch of flowers, was that the insects would be very vulnerable to predation whilst foraging but difficult to catch once airborne. Slow flight around flowers might be expected as insects are looking for suitable flowers on which to forage.

In conclusion, the flight behaviour of droneflies when foraging around flowers, including the routes taken and the extent of hovering, is remarkably similar to the flight behaviour of honeybees. More research is needed on other flowers with a wider range of insects, but these results clearly demonstrate flight mimicry of honeybees by droneflies in addition to the behavioural mimicry already established (Golding and Edmunds, 2000).

We thank Dr Bill Bailey at the University of Manchester for use of the time-lapse video recorder and Dr Robert Srygley and a referee for valuable comments.