Wing wear, aerodynamics and flight energetics in bumblebees (Bombus terrestris): an experimental study



  • 1 Previous work has shown that wing wear increases mortality rate in bumblebees. Two proximate explanations have been suggested to account for this: increased energy flight costs and increased predation risk due to reduced manoeuvrability.
  • 2 Wing wear was mimicked by experimentally clipping the forewing distal trailing edge, causing a 10% wing area reduction. Experimental and sham control bumblebees were induced to hover in a flight respirometry chamber for measuring metabolic rate of hovering. Simultaneous video and sound recordings were taken for wingbeat kinematic data required for an aerodynamic analysis.
  • 3 In the experimental group with reduced wing area we measured increased wingbeat frequency, lift coefficient and induced power, but a reduced profile power. The mechanical power output, assuming perfect elastic storage in the flight system, remained largely unchanged after the wing-trimming treatment.
  • 4 Metabolic flight costs (CO2 production rate) did not increase significantly in the reduced wing area group, which is in line with the aerodynamic power output.
  • 5 Our results indicate that an increase of flight cost due to wing wear is not a likely explanation for increased mortality rate in bumblebees. Wing wear may, however, affect escape performance from predators.


Foraging behaviour in social insects seems to be adjusted so that the ratio of net energy benefit to cost is maximized, usually called the foraging efficiency (Schmid-Hempel, Kacelnik & Houston 1985; Kacelnik, Houtson & Schmid-Hempel 1986; Schmid-Hempel 1986; Cartar & Dill 1990; Dyer & Seeley 1991). This is to be expected if there is only a finite amount of flying time available, and it has been argued that the wings have a limited life before they become too abraded for proper function (Rodd, Plowright & Owen 1980). It has been shown that the life span of worker honeybees (Apis mellifera L.) is correlated with total flight duration (Neukirch 1982). In a wing-clipping experiment, Cartar 1992) showed that worker bumblebees (Bombus melanopygus Nylander) with a reduced wing area experienced a lower life expectancy compared with controls. Also, the amount of natural wing wear was associated with elevated mortality (Cartar 1992). Hence, wing wear seems to be a proximate factor responsible for an increased mortality in bumblebees, but the direct mechanism remains unknown. One hypothetical mechanism is that the reduced wing area could affect the energetic cost of flight, so that it becomes progressively more expensive to fly with increasingly worn wings. An alternative is that the reduced wings could lower manoeuvrability so that worn bees are more susceptible to predation (Rodd et al. 1980). We investigated if the first hypothesis is applicable to bumblebees by experimentally reducing the wing area and measuring the flight cost in hovering bumblebees (Bombus terrestris L.) using a respirometer. The rate of energy expenditure shows a rather flat relationship with forward speed in bumblebees (Ellington, Machin & Casey 1990), and hence hovering flight cost should be a reliable measure of the overall flight cost. We also conducted an aerodynamic analysis of the effects of reducing the wing area on hovering flight performance.

Materials and methods


An artificially bred bumblebee B. terrestris colony was purchased (Biological Crop Protection Ltd, Kent, UK) and installed in the laboratory. The colony nest box was connected to a small Perspex flight cage (35 cm × 25 cm × 20 cm) through a 15-cm long plastic tube. In the flight cage the bees were offered a sucrose solution (66% w/w) ad libitum.


Bumblebees were captured in the flight chamber and transported in small glass tubes to another room for respirometry measurements and recording of flight kinematics. Every second bee was assigned to either the experimental group or the sham control group. Experiments were carried out between 27 March and 19 April 1996. The experimental treatment comprised the trimming of the outer trailing edge of the forewing by fine scissors, resulting in a mean wing area reduction of 10% (Fig. 1; Table 1), which is similar to the wing area lost due to natural wing wear in wild bumblebees (cf. Cartar 1992). The sham treatment was similar to the trimming procedure, but the scissors only touched the trailing edge of the wing without removing any wing area. Bees were flown in the respirometry chamber (see below) before and after experimental treatment, hence we did a repeated measure design experiment. After the second flight measurement the bees were sacrificed, wings cut off and weighed to the nearest 10−5 g using a Sartorius R200D balance (Sartorius AG, Göttingen, Germany), and preserved together with the cut-off pieces in glass slides for later analysis (see below).

Table 1.  Wing area (S, mm2), aspect ratio (AR), wingbeat frequency (n, Hz) and wingbeat amplitude (Φ, deg) before and after experimental treatment for the sham control and wing reduction groups. Values are mean ± 1 SD. Also given is the outcome of a repeated ancova analysis. N = 17 for the sham control and N = 18 for the experimental cut-group. 1 is first measurement and 2 is after experimental treatment
MeasureSham 1Sham 2Cut 1Cut 2F1,33P
S 94·1 ± 15·4 94·1 ± 15·4103·2 ± 18·1 92·8 ± 15·7137·0<0·001
AR 6·45 ± 0·15 6·45 ± 0·15 6·48 ± 0·19 6·49 ± 0·33  0·015   0·902
n182·9 ± 9·1181·5 ± 9·3175·1 ± 8·9186·6 ± 9·7 92·3<0·001
Φ114·5 ± 3·3114·0 ± 2·9115·3 ± 3·1116·5 ± 4·0  2·22   0·146
Figure 1.

Bumblebee wings showing how the forewing area was experimentally reduced to mimic natural wing wear. On average 10% of the wing area was removed by clipping the distal trailing edge, an amount and position typical of natural wing wear.


An open-flow respirometry system was used for measuring rates of CO2 production during hovering flight (see Wolf et al. 1996). Room air was scrubbed of CO2 and water vapour using soda lime and silica gel, respectively, and pumped through the hovering chamber at 1 l min−1. Rate of flow was measured using a Honeywell (AWM3300 V) mass-flow sensor, and the pump (ADC type 124) was controlled by a custom-built flow controller. The hovering chamber was a cubic (side length 13 cm, volume 2·2l) sealed Plexiglas chamber where the bees were induced to hover for periods of up to 20 min by the aid of a circular UV lamp placed above the chamber. The intensity of the UV light was manually controlled by a dimmer so that the bees were hovering near the middle of the chamber. Gas samples were drawn from the exit port through an infrared CO2 gas analyser (ADC 225 Mk3). The reference gas (dry CO2-free room air) was drawn directly into the second port of the CO2 analyser. For calibration at the beginning and end of each experiment, 50 ml of 8·0 ± 0·08 mol% CO2 (BOC special gases) injected into the system using a motorized syringe pump (Harvard ‘33’; Harvard Apparatus, South Natick, MA, USA) at a flow rate of 4 ml min−1 was used. CO2 levels were monitored on an SE 460 chart recorder (BBC Goerz Metrawatt, ÖVE, Austria) until an asymptotic plateau level was reached, and that was compared with the level of the calibration CO2 injection. Room temperature was controlled at 22–23 °C and ambient air pressure monitored at each flight measurement. All results were corrected to standard temperature and pressure, dry.


Immediately before and after a flight experiment, body mass was measured to the nearest mg on a Mettler BB240 balance (Mettler Toledo, Columbus, OH, USA). After experimental manipulation and before the second flight measurement, the bee was fed with sugar water until it had the same initial mass (<1 mg) as during the first measurement. After the second flight experiment both wing pairs were cut from the body at the wing base and weighed as described above. Wing length of both wings was measured to ±0·01 mm using vernier callipers. Information concerning wing parameters, necessary for the aerodynamic analysis (Ellington 1984a) was measured as follows. The cut-off wings and the pieces removed were mounted in glass slides and placed on a light box. The image was captured using a Macintosh Quadra 650 computer with a Neotech Image Grabber 124 (Neotech Ltd, Eastleigh, UK) connected to a Panasonic video camera (wv-BL600). The pieces were outlined using the auto-trace function in Canvas, and the original wing reconstructed by moving the outlines. Wing areas were analysed following Ellington (1984a).


Two video cameras (Hitachi CCTV HV735K) were mounted for simultaneous top and side view recording of the hovering bumblebee when flying in the respirometry chamber. Background light for both cameras was achieved by Fresnel lenses focusing DC lights onto the cameras. The two cameras were synchronized to the monitor of a BBC Master microcomputer by a switch-box, which also multiplexed the camera signals. The combined output alternated between top and side views 50 times per second. Each field (1/50 s) recorded about 3–4 wingbeats, from which the wing-stroke amplitude could be measured from the blurred image of the beating wing (see Cooper 1993). The computer monitor displayed a time signal, experimental information and a frame counter, which was keyed onto the video signal by a Videomatte VMI Mix-and-Wipe box (Videomatte, London, UK). The video output was recorded on a Sony U-matic KCA-60K videotape using a JVC CR-6060E video recorder. The wingbeat frequency was recorded onto the videotape using a small microphone fixed inside the flight chamber.

Dorsal and ventral edges of the wing-blur from the top view were measured in each field to give the maximum and minimum positional angles, φmax and φmin, of the wing. From these values the mean positional angle φ and wingbeat amplitude Φ were derived. For the aerodynamic analysis it was assumed that the stroke plane was horizontal (Ellington 1984b; Dudley & Ellington 1990a, 1990b; Cooper 1993). Wingbeat frequency n was obtained by fast-Fourier transform (FFT) analysis of the digitized sound track.


An aerodynamic analysis of hovering flight was performed before and after experimental treatment according to Ellington (1984c). Morphological and kinematic data were used to calculate lift coefficient (CL), profile drag coefficient (CD,pro), induced power (Pind), profile power (Ppro), and the total mechanical power (P0) assuming perfect elastic storage (Ellington 1984c, 1985). Perfect elastic storage assumes that all kinetic energy is stored during the deceleration of the wing and then released to accelerate the wing on the next half-stroke. This requires zero inertial power and the mechanical power output of the muscles is equal to P0 = Pind + Ppro. The value of CD,pro calculated according Ellington (1984c) is currently being revised, and Ellington (1999) suggested that a ratio between CL and CD,pro of about 1·7 might be more appropriate. This relaxation was used to calculate alternative values of Ppro and P0. Both the old and these new values were used for comparisons before and after experimental treatment. Mechanical power P0 from the aerodynamic analysis was divided by the power input as measured from the CO2 production rate to yield the overall conversion efficiency. The aerodynamic calculations were made by an Excel spreadsheet, ‘Hover’, implementing the aerodynamical theory of hovering flight due to Ellington (1984c).


The experimental design made the statistical analyses straightforward. Most flight parameters scale with size in insects, and so the derived aerodynamic properties and the measured flight costs for the sham control and experimental groups were analysed by repeat measurement ancova’s, with body mass as covariate and experimental treatment as factor (Sokal & Rohlf 1995). The statistical tests were carried out using STATISTICA version 5·1 (StatSoft Inc, Tulsa, OK, USA). Posthoc statistical power estimates for the experimental design used were calculated, i.e. the probability of rejecting the null hypothesis when it is false and the alternative hypothesis is correct, using the software G*Power (Erdfelder, Faul & Buchner 1996). The calculated power for a medium effect (f2 = 0·15; Cohen 1988) was 0·604, while for a large effect (f2 = 0·35) it was 0·925.



The wing area cut off of the experimental bumblebees averaged 10·4 ± 3·67 mm2 (mean ± SD, range 4·40–17·70 mm2, N = 18; Table 1), which was on average 10 ± 2·5% of the total wing area (range 4·7–13·8%). Wing length was reduced from 12·9 ± 1·18 mm to 12·2 ± 1·08 mm due to the wing clipping. The shape of the wing measured as the aspect ratio remained unchanged due to the experimental manipulation (Table 1).


There was a significant negative relationship between wingbeat frequency and body mass for the first measurements of control and experimental groups combined as n = 292m−0·092 (Hz; r2 = 0·17, P = 0·013, N = 35), where m is in mg. Hence, body mass is an important variable affecting the wingbeat frequency and the reason for comparing the control and experimental groups by ancova with body mass as covariate. Wingbeat frequency increased significantly after the experimental reduction of wing area (Fig. 2), but remained unchanged in the control group (P < 0·001; Table 1). Wing-stroke amplitude was, however, unaffected by the wing area reduction (P > 0·05; Table 1). These parameters together with direct and derived morphological parameters were used for the aerodynamic analysis.

Figure 2.

Wingbeat frequency in relation to body mass measured before treatment (○) and after experimental or sham treatments (●): (a) experimental group where the wing area was reduced; (b) control group where no wing area was removed. The increase in wingbeat frequency after wing clipping treatment was statistically significant (P < 0·001; F1,33 = 92·3).


The means (±SD) of the derived parameters before and after experimental manipulation with the outcome of the statistical tests are given in Table 2. The lift coefficient increased significantly due to the wing area reduction (P < 0·001; Table 2). This plus the reduction in wing length explains the associated increase of induced power (P < 0·001; Table 2a). The profile drag coefficient was not significantly affected by the reduced wing area, but the profile power decreased significantly (P < 0·001; Table 2a). The mechanical power output assuming perfect elastic storage, did not change when reducing the wing area, but the work done per wingbeat (W0) decreased significantly (P < 0·001). The latter effect is, however, only due to the increased wingbeat frequency with reduced wing area. Similar results were obtained with the new assumption concerning the profile drag coefficient (Table 2b), although it should be noted that profile power did not change significantly.

Table 2.  Aerodynamic measures and the effect of experimentally reducing the wing area of bumblebees. The measures are lift coefficient (CL), profile drag coefficient (CD,pro), induced power (Pind), profile power (Ppro), power output for perfect elastic storage (P0) and work per wingbeat (W0). Panel (a) is based on CD,pro estimates from Ellington (1984c). Panel (b) assumes CL/Cd,pro = 1·7. Values are mean ± 1 SD. Also given is the outcome of a repeated ancova analysis. N = 17 for the sham control and N = 18 for the experimental cut-group. 1 is first measurement and 2 is after experimental treatment
MeasureSham 1Sham 2Cut 1Cut 2F1,33P
CL 1·36 ± 0·22 1·39 ± 0·21 1·18 ± 0·28 1·35 ± 0·2918·4<0·001
CD,pro 0·15 ± 0·011 0·15 ± 0·011 0·15 ± 0·013 0·15 ± 0·012 2·69   0·110
Pind 4·58 ± 1·45 4·60 ± 1·44 3·99 ± 0·97 4·34 ± 1·0247·0<0·001
Ppro 2·05 ± 0·68 1·99 ± 0·64 2·29 ± 0·75 1·97 ± 0·6514·6<0·001
P0 6·64 ± 2·00 6·59 ± 1·98 6·28 ± 1·50 6·31 ± 1·48 2·43   0·129
W00·037 ± 0·0120·037 ± 0·0120·036 ± 0·0090·034 ± 0·00863·5<0·001
CD,pro 0·80 ± 0·13 0·82 ± 0·12 0·69 ± 0·16 0·79 ± 0·1718·4<0·001
Ppro10·90 ± 3·6210·79 ± 3·5710·45 ± 2·7410·20 ± 2·67 0·989   0·327
P015·48 ± 4·9915·38 ± 4·9614·44 ± 3·6214·54 ± 3·61 3·28   0·079
W00·086 ± 0·0290·086 ± 0·0290·083 ± 0·0210·078 ± 0·02160·8<0·001


Measurements of carbon dioxide (CO2) were obtained before and after control and experimental treatments for 17 and 18 bumblebees, respectively. The wing area reduction had a small but not significant effect on the rate of CO2 production rate and hence on power input (Table 3). There was no significant effect on the metabolic energy per wingbeat either (P > 0·05; Table 3). Using the ‘old’ assumption regarding the profile drag coefficient, the conversion efficiency of metabolic power into mechanical power output was about 7% (Table 3), which was not significantly affected by the wing reduction. With the ‘new’ assumption (CL/CD,pro = 1·7) the conversion efficiency became higher at about 15–16% (Table 3), but still unaffected by the wing reduction.

Table 3.  Measured rate of CO2 production (ml h−1), metabolic power input (Pi, mW) and conversion efficiencies for the standard CD,pro (η, %) and the new assumption of CL/CD,pro = 1·7 (ηn,%). Values are mean ± 1 SD. Also given is the outcome of a repeated ancova analysis. N = 17 for the sham control and N = 18 for the experimental cut-group. 1 is first measurement and 2 is after experimental treatment
MeasureSham 1Sham 2Cut 1Cut 2F1,33P
CO˙215·6 ± 3·2315·8 ± 2·6616·0 ± 4·0117·0 ± 3·801·140·293
Pi90·9 ± 18·892·1 ± 15·593·2 ± 23·499·4 ± 22·11·140·293
η7·25 ± 1·167·09 ± 1·366·90 ± 1·476·37 ± 0·791·150·290
ηn16·8 ± 2·9916·5 ± 3·5215·8 ± 3·4214·6 ± 1·951·010·322


Most surprisingly no significant effect of reduced wing areas on measured CO2 production rate and hence the metabolic cost of flight was found. The aerodynamic analysis provides the following probable explanation for the lack of significant effects on the energetic cost. Reducing the wing area increased the wingbeat frequency, presumably because the wing moment of inertia decreased. For weight support, the reduction in wing area was partly compensated by the increased frequency and also by an increase in lift coefficient. The wing length was also reduced by a small amount by clipping, which increased the induced power. However, the profile power decreased or remained constant, depending on the relation assumed for CD,pro, because of the reduction in the third moment of wing area (Weis-Fogh 1973; Ellington 1984a). The wing area manipulation was distal and the third moment of wing area is very sensitive to distal changes. For the ‘old’CD,pro relation, the two opposing effects on Pind and Ppro largely cancel, giving a more or less constant power output (P0) in spite of the wing area reduction. For the ‘new’CD,pro relation, profile power does not change significantly and is much larger than induced power, so the power output is again unchanged. Consequently, there were no significant increases in metabolic rate (Pi or CO2 production rate), which is in agreement with the near constant aerodynamic power output. Hence, there are no significant energetic consequences of wing clipping and therefore presumably of wing wear in bumblebees.

Cartar (1992) also experimentally reduced the wing areas in a North American bumblebee species and measured a reduced survival compared with controls. Cartar also found a correlational evidence for natural wing wear being associated with increased mortality. He proposed two hypotheses to account for the increased mortality rate, one being that the experimental bees reach their finite physiological limit at an earlier time due to an increased metabolic rate. This hypothesis is at variance with our findings. The alternative hypothesis is that the wing clipping experiment affects manoeuvrability and thereby the escape capacity from predators (Morse 1986; Rodd et al. 1980), and further investigations should concentrate on these factors.


This work was carried out when A.H. was supported by a postdoctoral exchange fellowship from the Swedish Natural Sciences research Council.

Received 4 August 2000; revised 9 November 2000; accepted 16 November 2000