Flight activity and age cause wing damage in house flies

Insect wings are delicate structures which often break over the course of an animal's lifetime. Multiple reasons for wing damage have been proposed, including interactions between conspecifics during sexual competition and mating (Alcock, 1996; Burkhard et al., 2002; Pinheiro, 1990; Ragland and Sohal, 1973), injuries during oviposition, building and defending a nest, and intraspecific and interspecific aggression, competition and predation (Benson, 1972; Nalepa, 2012; O'Neill et al., 2015; Rajabi et al., 2020). Besides the above reasons, wing damage predominantly results from the collision of wings with objects (Burkhard et al., 2002; Foster and Cartar, 2011; Rhainds and Brodersen, 2012) and thus increases with increasing flight activity (Hayes et al., 1998). In general, adverse habitats (Lee et al., 2006) and high population densities of animals in captivity favour wing damage (Ragland and Sohal, 1973). Wing damage also depends on the sex ratio within an insect group (Rhainds and Brodersen, 2012) and is sex specific (Burkhard et al., 2002; Kiritani et al., 2013; Lee et al., 2006; Mueller and Wolf-Mueller, 1993; Ragland and Sohal, 1973; Rhainds and Brodersen, 2012). The relevance of wing damage for the fitness of flying insects is due to the negligible healing capacity of broken wings, although puzzling findings previously suggested that wings may regrow after complete extirpation (Kammerer, 1907; Werber, 1907, 1911). These findings, however, were also explained by the deformation of dead cuticle material and not a regeneration of living tissue (Kříženecký, 1914). Wing damage in holometabolic animals thus accumulates over the animal's lifetime and is used for estimations of chronological (Allsopp, 1985; Hargrove, 2020; Mueller and Wolf-Mueller, 1993) and physiological age (Burkhard et al., 2002; Hayes et al., 1998).

There is ongoing controversy on the development of wing damage in insects and its dependency on flight activity. Several authors reported an increasing rate of wing damage with increasing lifetime (Foster and Cartar, 2011; Higginson and Barnard, 2004), while others found the opposite (Mueller and Wolf-Mueller, 1993; O'Neill et al., 2015). For example, Mountcastle and Combes (2014) proposed that insect wings with broken tips accrue less additional damage. They argued that frayed and shorter wings are more compliant than intact wings when hitting a solid object and thus the rate of wing wear should decrease with increasing lifetime. Studies on ‘lifetime potential’ concepts, for example in bumblebees, suggest that flight with damaged wings leads to faster physiological senescence and premature death (Cartar, 1992). Vance (2009) observed an inability of older animals to compensate for wing area loss by an appropriate increase in wing flapping frequency, which implies a loss of fitness. There are experimental findings on male flies that supposedly showed a reduction in longevity when the wings were artificially removed from the body. This result has been attributed to an overuse of leg motion as a result of mating attempts because it was not present in solitary males (Ragland and Sohal, 1973).

The behavioural and fitness consequences of wing area loss for flight are manifold and the animals' compensatory mechanisms remarkable. For example, (i) bumblebees with damaged wings often have unchanged flight behaviours compared with intact conspecifics (Haas and Cartar, 2008), (ii) dragonflies with a missing hindwing are still able to turn, land on a moving object and catch flies, with little difference in locomotor force production compared with intact animals (Kassner et al., 2016), and (iii) butterflies with wing damage still defend territories against intruders with intact wings (Monge-Nájera et al., 1998). The last, however, may be explained by the highly ritualised nature of butterfly fights, in which the intruder usually loses (Davies, 1978). Nevertheless, the loss of wing area typically attenuates flight capacity. Insects with damaged wings fly more slowly than those with intact wings (Combes et al., 2010; Fischer and Kutsch, 2000; Haas and Cartar, 2008; Jantzen and Eisner, 2008; Muijres et al., 2017) and maximum acceleration and manoeuvrability are usually reduced (Combes et al., 2010; Haas and Cartar, 2008; Mountcastle et al., 2016). It has been suggested that this reduction increases the risk of predation (Cartar, 1992; Combes et al., 2010; Dukas and Dukas, 2011; Jantzen and Eisner, 2008; Vance and Roberts, 2014) and social insects such as bees may suffer from reduced performance of wing-damaged foragers (Dukas and Dukas, 2011; Higginson and Barnard, 2004; Johnson and Cartar, 2014). Similar results have been reported for small solitary carpenter bees (Rehan and Richards, 2010). In general, wing damage reduces the animal's load-lifting capacity, although neither the shape nor the symmetry of damage seems to be significant (Buchwald and Dudley, 2010; Fernández et al., 2017; Roberts and Cartar, 2015). Data show that insects lose their flight ability if wing area loss exceeds ∼33–40% of the initial area (Haas and Cartar, 2008; Ragland and Sohal, 1973). Some four-winged insects may still actively fly with only two wings, i.e. a ∼50% reduction in total wing area (Kassner et al., 2016), and butterflies without hindwings produce enough lift for flight but at the cost of lower flight and turning speeds (Jantzen and Eisner, 2008).

To compensate for the loss of wing area, many insects such as moths, bumblebees and dragonflies adapt their wing kinematics to the smaller lift-generating surface, increasing or decreasing wingbeat frequency (Danzer, 1956; Fernández et al., 2012, 2017; Hargrove, 1975; Hedenström et al., 2001; Jantzen and Eisner, 2008; Kassner et al., 2016; Kingsolver, 1999; Muijres et al., 2017; Roberts and Cartar, 2015; Sotavalta, 1952a; Vance and Roberts, 2014) and increasing wingbeat amplitude (Buchwald and Dudley, 2010; Fernández et al., 2017; Johnson and Cartar, 2014; Lyu et al., 2020; Muijres et al., 2017; Vance and Roberts, 2014). While in Diptera the changes in wing beat frequency are thought to result from the reduction in mass load on the mechanical thoracic oscillator (Sotavalta, 1952b), the changes in wingbeat frequency of insects with synchronous flight muscles necessarily result from neural adjustments in muscle contraction (e.g. Wilson and Weis-Fogh, 1962). Besides wingbeat frequency and amplitude, insects also change their wing tip path during upstroke and downstroke and body posture in response to wing wear (Fernández et al., 2012; Lyu et al., 2020; Muijres et al., 2017) that moreover alter aerodynamic force production and moment control around yaw, pitch and roll axes of the insect body.

To investigate the dependency of wing damage on both flight activity and age in an insect, and to estimate the significance of wing area loss for flight capacity, this study experimentally investigated the time evolution and behavioural impact of wing damage in the house fly Musca domestica. We scored flight activity throughout the entire lifetime of adult animals, mapped their wing wear at regular intervals and tested the ability of animals with damaged wings to support their body weight in free flight. The statistical analyses suggest that wing area loss in flies follows a logistic function and depends equally on flight activity and age. We also found that the animal's active flight capacity ceases when ∼10–34% of total wing area is lost.

This study encompasses two experimental approaches: first, an assay with which we determined the progression of wing wear and flight activity in small groups of flies kept in small transparent cylindrical plastic containers (height, 72 mm; diameter, 48 mm; Fig. 1C) with ∼130 cm³ volume (container experiment); and second, a behavioural assay, in which we scored wing wear and the flies' ability to freely fly (free-flight experiment) in a ∼2000 times larger mesh flight cage (52×52×93 cm³, ∼0.25 m³ volume). Both approaches comprised lifetime measurements (container experiment, total experimental time: 153 days; free-flight experiment, total experimental time: 94 days). If not stated otherwise, all data are given as means±s.d. Data are available through Dryad (Lehmann, 2021).

In all experiments, we used house flies Musca domestica L. from a laboratory strain obtained from the German Federal Environmental Agency (Fig. 1A,B). The flies were kept under a 12 h:12 h light:dark cycle and supplied ad libitum with sugar, powdered milk and water; females had the opportunity for oviposition on a moist medium. Larvae were reared in separate plastic containers, and transferred into plastic tubes after pupation, in which the adults hatched. For experiments, all animals were marked with small amounts of paint on their thorax (Pelikan, Hannover, Germany).

For container experiments, the animals were sorted according to sex, and moved from their hatching container into the experimental container at a median age of 0.5 days, where they were kept until they died. There was no significant difference in transfer age between males and females (Welch two-sample t-test, P=0.39). We estimated body mass when the flies were ∼4.5 days old. Dead flies were replaced by flies of the same age from a control container in order to keep group sizes constant in the experimental containers (see below). We did not collect wing damage data from flies that had been added as a replacement. In total, we tested 72 flies in 18 groups (4 flies per container) and the final number analysed was 27 females and 36 males. Body mass was ∼18.6±2.8 mg in females and ∼14.1±2.4 mg in males (Fig. 2A). Females were significantly heavier than males (Welch two-sample t-test, P<0.001). Wing area was also greater in females than in males (Fig. 2B,E) in both experimental approaches, and significantly increased with increasing body mass (Fig. 2C). For free-flight experiments, we placed 10 females and 10 males immediately after hatching into the mesh cage and kept them there until they died. As three flies were accidentally killed during the experimental period, the data in this approach stem from 8 females and 9 males. In this case, dead flies were not replaced.

For age-related phenomena, the physiological age can be more relevant than the chronological age (Burkhard et al., 2002; Hayes et al., 1998). We thus used physiological age in this study and estimated its value in degree days (Kelvin days, K days) above a threshold temperature of 11°C (Skovgård and Nachman, 2004). Temperature was continuously scored and eventually averaged for each 12 h light or dark period. Missing values were interpolated. Fly hatching was monitored twice a day. Physiological and chronological age are linearly correlated with a Pearson's correlation coefficient of >0.99. Mean linear regression slope between chronological and physiological age was 15±3 K and varied between ∼10 K and 18 K (container experiments). The slope in the free-flight experiment was ∼13 K. For correlations between physiological and chronological age, see Supplementary Materials and Methods 1. The theoretical maximum precision in physiological age estimation was 6.75 K days and mean ambient temperature for all measurements was ∼24.5°C.

The experimental setup consisted of nine cylindrical plastic containers that were mounted in 3×3 holes of an acrylic glass rack. All containers except for the water supply were surrounded by a printed random dot visual environment and equipped with an electret microphone at the bottom (Fig. 1C). Other containers served as the water supply, reference for sound recording (empty container) and fly reservoir to replace flies that died in one of the six experimental containers. All containers were closed with foam stoppers. The rack was illuminated from above by two fluorescent lamps (16 W, colour temperature 3400 K) on a 12 h:12 h day:night cycle (07:00–19:00 h). Measured illuminance inside the containers varied between 380 and 840 lx. We monitored ambient temperature (mean 24.7±2.8°C) and relative humidity (mean 38±6%) every 5 min with a data logger (BL-30, Trotec GmbH, Heinsberg, Germany).

One of our initial ideas was the reconstruction of wing beat frequency from the sound recordings during flight. However, simultaneous high-speed video analysis on wing motion and sound recording revealed that the major frequency component obtained from fast Fourier analysis did not reliably map wing flapping motion. Possible explanations include sound interference between the two beating wings, sound being reflected within the container and Doppler effects. We thus excluded the frequency analysis from this study.

To score the relationship between wing damage and locomotor capacity, flies were kept and tested in the mesh flight cage under similar environmental conditions to those used for the container experiments. Mean illuminance inside the cage was ∼350 lx at the bottom and 42,000 lx at the top, mean ambient temperature was 24.1±0.8°C, and relative humidity was 41±7%. Each time we removed the flies from the cage in order to take photographs of the wings, we tested their ability to support their body weight in free flight over a flight distance of 26 cm by releasing them in the middle of the cage. Weight-supporting active flight was considered when the altitude of the landing site was equal to or above the altitude of the start post.

Our statistical analyses cover approaches on two topics of this study. The first topic was the relationship between wing area loss, physiological age and flight activity (container experiments) and the second topic was the relationship between wing area loss and flight capacity (free-flight experiments). The statistics on the first topic consist of three steps: (i) a pre-test procedure using Pearson correlation, (ii) a principal component analysis (PCA) with Kolmogorov–Smirnov test, and (iii) a comparison of model parameter variability to assess the significance of physiological age and flight activity on wing area. The statistics on the second topic encompassed PCA only.

To pre-evaluate any relationship between wing area loss and the two predictors (physiological age and cumulative flight activity), we tested whether t_f values at 0.85 derived from the logistic fits are clustered compared with a reference distribution. A clustered value would suggest that wing area loss starts at a specific combination of age and activity during the animals' lifetime. A limitation of this approach is that it excludes all wings for which we could not conduct a logistic fit, for example all wings with less than 15% area loss (see above). Area loss of less than 15% prevented fitting a curve in ∼80% female and ∼35% male wings. For analysis, we reduced the two dimensions age and activity to one dimension using PCA. The magnitude of the first principal component (PC1) for all area measurements in all flies was used as a reference. As the units of the two predictors are different (age, K days; activity, h), age and activity were z-transformed by subtracting means and normalising to standard deviation before PCA. The reference distribution takes into account the various lifespans because the number of age and activity data points decreases with increasing age and activity in the tested flies (Fig. 2D,F). In other words: it considers the changes in local density of PC1-values due to the decreasing number of flies alive with increasing time. To directly compare reference values along PC1 with t_f values from the logistic fits, we z-transformed t_f values for age and activity using the means and standard deviations of the reference dataset, copied the PC1 eigenvector of the reference PCA into the t_f coordinate system instead of performing a new PCA, and estimated PC1 t_f values along the eigenvector. To test for statistical differences between the reference and t_f distributions along PC1, we used a Kolmogorov–Smirnov test with Bonferroni correction for multiple testing.

PCA cannot provide evidence for the strength of each predictor for wing damage. We approached this problem by comparing the variance of the beginning of wing damage (t_f) between both predictors. The t_f values were derived from the logistic wing models fitted according to age and activity, respectively. If the models are more uniform along one dimension, this dimension would appear to be a better predictor of wing damage. As the two predictors possess different scales, we transformed the two sets of t_f values using their medians for centring them on zero and normalising their scale. After transformation, data variability was relative to the original scale and the variance of both datasets was compared by the Fligner–Killeen test on homogeneity of variances. The above approach is limited to those wings for which a logistic curve could be fitted successfully (∼15% for activity as a predictor and 20% for age as a predictor for female wing models, ∼50% for male models in both cases). This excludes wings with less than 15% wing area loss (which is the case for 80% of female wings and 35% of male wings) and wings where the fitting process was unsuccessful for other reasons. An alternative approach that allowed us to include all measured data such as multiple linear model regression analysis was not suitable because of the logistic development in wing area loss.

In the container experiment, females reached physiological (chronological) ages of between 310 and 880 K days (17.5 and 59.5 days); mean age was 480±140 K days (32.0±11.5 days; Fig. 2D). Females lived ∼1.5 times longer than males (two sample t-test; physiological age, P<0.001; chronological age, P<0.01), which reached maximum ages of between 30 and 660 K days (2.0 and 49.5 days); mean age of these flies was 320±170 K days (22±12 days; Fig. 2D). In the free-flight experiments, female flies lived 400–1210 K days (31 to 92.5 days) with a mean of 860±270 K days (65.5±20.5 days; Fig. 2F). In contrast to the container experiment, the life expectancy of males in the free-flight cage was relatively shorter but not significantly different from that of females (two-sample t-test, P>0.09 for both ages) and ranged from 400 to 930 K days, with a mean age of 650±180 K days (31 to 70.5 days; 50.0±13.5 days). In sum, flies in the free-flight experiment tended to live twice as long as those in the container experiment.

In total, we recorded 7,776,670 flights of which ∼30% were measured in containers with females and ∼70% with males (Fig. 3). Fig. 3A shows frequency and length of flight bouts and how they were distributed over one day. Mean duration of flight bouts in females was 0.20±0.23 s, significantly longer than that in males (0.17±0.17 s; Wilcoxon rank sum test with continuity correction, P<0.001). Fig. 3B highlights that ∼75% of all flight sequences were equal to or shorter than ∼200 ms (females, ∼74.8%; males, ∼74.6%). The 99% quantiles in flight activity bout durations thus yield 0.92 s in females and 0.79 s in males. The ethogram in Fig. 4 shows that activity bouts were not equally spaced in time, owing to the animals' circadian rhythm (Fig. 4A). Flight activity during the light period was significantly higher than activity during the dark period (paired t-test, P<0.001; Fig. 4B). The data also show that males were significantly more active than females during both the light (females, ⁠; males, ⁠) and dark period (females ⁠; males, ⁠; Welch two-sample t-test, P<0.05; Fig. 4C).

Although the two experimental approaches (container, free flight) provided separate datasets for wing damage, we summarise these findings in this section. Fig. 5 shows two examples of wing damage for six observations of two different flies for container (male; Fig. 5B) and free-flight experiments (female; Fig. 5A). In total, we were able to score wing damage in 534 and 656 photos in container experiments and 190 and 157 photos in free-flight experiments of females and males, respectively. Wing wear typically started in both sexes by a loss of wing area at the wing tip or the distal parts of membrane area at the trailing wing edge. At this initial state, we did not observe any area loss at the leading edge in either sex. With increasing age, the wings also broke at the stiff leading edge and typically lost all distal areas. Thus, wing length decreased with increasing age and flight activity.

As mentioned in Materials and Methods, we used a logistic fit function to describe wing area loss in a subset of all data. The examples of this analysis show fast (Fig. 6A–C) and slow (Fig. 6D,E) damage progress. In general, the results of the logistic fit suggest that wing area loss occurs suddenly and quickly progresses with age and activity (Table 1). As already mentioned, the analysis condition excluded single wings (container experiment) and sets of wings of individual animals (free-flight experiment) that experienced less than 15% wing area loss, i.e. 33 female and 24 male wings in container experiments and 2 females and 1 male in free-flight experiments. These values correspond to ∼83% and ∼34% of the tested female and male wings in container experiments and ∼25% of females and ∼11% of males in free-flight experiments, respectively. For the predictors physiological age and cumulative activity, we eventually derived data from 7 and 6 female wings, and 37 and 37 male wings, respectively, which is 40% of all tested wings in container experiments. For the 6 females and 8 males available for fitting a curve in the free-flight experiment, one model per fly could be fitted successfully (∼82% of all flies).

We used these models as the basis for further analysis. In the container experiment, we found that wings lost on average 69% of their initial area within the flies' lifetime (Â_f=0.31±0.23 initial area). Area loss in single animals was widely scattered, ranging from ∼100% (complete area loss) to a minimum of ∼15% loss of initial area (container experiments) at death. We obtained similar data for flies in the flight cage (free-flight experiments) in which the mean sum of left and right wing area at death (Eqn 4) was ∼27% of the initial area and ranged from ∼100% to ∼16% loss of initial area. The time at which the flies lost 50% of their wing area loss, i.e. the inflexion point of the logistic fit function, was ∼258±165 K days for physiological age and ∼6.72±5.13 h for flight activity (container experiments), compared with ∼614±266 K days in free-flight experiments (Table 1). The data in Table 1 also highlight that flies in small containers lost wing area ∼20 times more rapidly compared with their relatives in the larger flight cage. We found that the rate of wing area loss at the inflexion point was ∼−31.4(±36.8)×10⁻³ K⁻¹ day⁻¹ for physiological age and ∼−2.68±4.46 h⁻¹ for accumulative flight activity, compared with ∼−1.59(±1.05)×10⁻³ K⁻¹ day⁻¹ in free-flight experiments. Logistic growth rates were broadly independent of the onset of wing damage (Fig. 1D) in both container (linear regression, flight activity versus k; R²<0.16) and free-flight experiments [flight activity versus k, R²<0.01; where k is the logistic growth rate (see Eqns 6,7)].).

Wing damage may depend on age or activity, a combination of the two, or neither. We approached this question by scoring the variance of the relationship between age and activity in each animal. Without variance, age and activity would be indistinguishable. This was checked using a linear interaction model with activity as outcome variable, age as covariate and individual fly identity as explanatory factor (Fig. 7A). We estimated regression slopes and intercepts of 53 animals and found a mean slope of ∼0.020±0.009 h K⁻¹ day⁻¹ for all flies. The slope varied between ∼0.005 h K⁻¹ day⁻¹ (Fig. 7A, yellow) and 0.042 h K⁻¹ day⁻¹ in single animals (Fig. 7A, cyan). The mean regression intercept was ∼0.21±0.71 h and close to zero, with a variability ranging from −1.36 to 2.07 h. Age, individual fly identity and their interaction were significant model terms, suggesting that both slopes and intercepts were generally different among flies (P<0.001). Pearson's correlation coefficient on pooled flies was ∼0.71 (625 data points, N=20 females, N=35 males) and smaller than the coefficients for single animals, which ranged from ∼0.97 to ∼1.0.

Firstly, we combined the two predictors and analysed to what degree this one-dimensional combination is correlated with wing damage. Because of the elevated degree of collinearity between the two predictors, this approach allowed us to analyse dependencies regardless of whether the two predictors could be disentangled. PCA on the combined predictors was done on 1250 data points of which 558 were recorded in females and 692 in males. The lower and upper quartiles of physiological age range from 120 to 370 K days in females and from 80 to 290 K days in males, and flight activity from 0.92 to 3.99 h and from 1.96 to 8.53 h, respectively. The first principal component (PC1) explained ∼86% of the data variance. Fig. 7B shows that after z-transformation, approximately 50% of PC1 values were distributed between −1.13 and 0.57 in females and between −1.08 and 1.09 in males. Calculations further yielded t_f=450±60 K days and 4.68±1.13 h in females, and 210±130 K days and 6.32±3.88 h in males, respectively. After projecting these values into PCA space, ∼50% of PC1 varied between 0.69 and 1.50, and −0.82 and 1.07 in females and males, respectively. We found no significant difference between the reference distribution and the distribution of t_f in pooled flies (P=0.172). This result, however, was sex specific because in females the difference was significant (P<0.05) while in males it was not (P=1.00). We also found that females and males differed in the reference distribution (P<0.05) but not in the distribution of t_f (P=0.2, Bonferroni correction, N=5 tests).

Secondly, we tested the two predictors as separate variables by comparing the relative variability of t_f for age and activity (see Materials and Methods). We found that the median of model deviance divided by degrees of freedom was 1.15×10⁻³ for physiological age (N=44 wings) and 1.14×10⁻³ for cumulative activity (N=43 wings). These values were not significantly different (two-sided two-sample Kolmogorov–Smirnov test, P≈0.89). This result suggests that wing damage might depend more on age than activity or vice versa if variability of model parameters is different for the two predictors, because the modelling did not obviously work better along one dimension than the other. We thus compared the relative variability of t_f between predictors and found no significant difference between the two values (P=0.807, Fligner–Killeen test). This means that wing damage in the subset of successfully modelled wings cannot be attributed to a single predictor.

Fig. 8 shows how wing area loss developed in house flies that lived in the large flight cage (free-flight experiment; analysis for this experiment was done at the animal level). These data potentially link wing damage to the animals' free-flight capacity. According to quasi-steady aerodynamic theory (see Discussion), lift production depends on wing area and increases with the wings' 2nd moment of area (Weis-Fogh, 1973). In general, our data show that normalised wing area [linear regression, ⁠, R²=0.05, P_slope<0.01, N=347; Fig. 8A] and mass-specific 2nd moment of area [linear regression, ⁠, R²=0.03, P_slope<0.01; Fig. 8B] decrease with increasing age (θ). In most flies, the same also held for normalised right minus left differences of these measures (Fig. 8C,E). Fig. 8D highlights the linear relationship [linear regression, ⁠, R²=0.85, P_slope<0.001] between wing area and mass-specific 2nd moment of area – flies with wing damage thus suffer from simultaneous changes in lifting-surface and wing aspect ratio with increasing age. The relative right minus left difference in wing area is largely independent of total normalised wing area until area drops below a threshold of ∼0.8 area (Fig. 8F).

Free-flight performance was sorted into only two categories (‘flight’, ‘non-flight’). We scored wing area for 347 out of 351 flight trials (192 trials, N=8 females; 159 trials, N=9 males) that split into 308 ‘flight’ and 43 ‘non-flight’ scores (Fig. 9). Fig. 9A shows that the age distribution of flying flies (median: ∼370 K days) is significantly different from the age distribution (median: ∼560 K days) of non-flying flies (two-sided two-sample Kolmogorov–Smirnov test, P<0.001, Bonferroni-corrected for 6 tests, N=351). The same holds for normalised wing area and mass-specific 2nd moment of area in both categories (medians: ∼0.97 versus ∼0.63, Fig. 9C; 33.1 versus 19.9 mm⁴ mg⁻¹, Fig. 9D; one-sided two-sample Kolmogorov–Smirnov test, P<0.001, Bonferroni-corrected for 6 tests, N=347). The reverse applies to absolute normalised differences of both parameters (medians: ∼0.01 versus ∼0.41, Fig. 9E; 0.02 versus 0.64, Fig. 9F, P<0.001). Similar to the analysis of the container experiment, we calculated the PC1 as a predictor for wing area loss using PCA on z-transformed normalised wing area, body mass-specific 2nd moment of area and absolute relative differences. PC1 accounted for 88% of the variance and medians were −0.78 (flight) and 3.43 (non-flight; Fig. 9B). We found that flight and non-flight distributions of PC1 were significantly different (two-sided Kolmogorov–Smirnov test with Bonferroni correction, P<0.001, N=347).

Flight by active wing flapping is one of the most costly forms of locomotion (Casey and Ellington, 1989). Consequently, any change in the flight apparatus such as wing area and shape may compromise flight ability and efficiency and thus have serious consequences for the animal's fitness. To highlight the significance of wing area loss, we here investigated the factors that determine wing damage in house flies and scored the ability of these flies to fly with damaged wings. In contrast to previously published experiments in which wings of flies were artificially cut (e.g. Higginson et al., 2011; Mountcastle et al., 2016; Muijres et al., 2017; Roberts and Cartar, 2015), we investigated natural progression of wing damage under two housing conditions. One focus of the study was to map the progression of damage and to describe wing area loss throughout the animal's entire lifetime as a function of age and flight activity. We found that (i) wing damage typically starts to occur at the wing tip and progresses in a characteristic pattern (Figs 5 and 6), (ii) the rate of wing damage is independent of both age and activity at the initial area change, (iii) physiological age and flight activity are both predictors for wing damage, (iv) flight activity patterns are different between females and males (Figs 2–4), and (iv) the flies' ability to produce forces for body mass support decreases with increasing age and decreasing wing area, shape and asymmetry between the two wings (Figs 8 and 9). Noteworthy, as age, flight activity and wing damage simultaneously change, the loss of flight ability probably results from various factors of which we only scored a small number.

Investigations on flies kept in cages previously showed that behaviour and life history change depending on housing conditions. The life expectancy of house flies depends strongly on activity, which is shaped by temperature, population density and sex ratio (Ragland and Sohal, 1973, 1975; Sohal and Buchan, 1981). To account for these factors, we monitored temperature throughout the measurements and used physiological age which considers temperature rather than the chronological age as a descriptor. Sex ratio was considered by keeping females and males in separate containers. It is obvious that higher population density increases the likelihood of wing collisions in flying animals. For this reason, we selected two housing conditions with vastly different population densities. Fig. 2 shows that house flies kept in the larger cage lived considerably longer and tended to have larger wing areas. The latter result might be due to the variance in rearing conditions because all tested flies stem from the same population. The longer lifespan of flies in the free-flight experiment is consistent with the idea that crowded housing conditions, elevated levels of activity and an elevated number of males in a group result in earlier senescence. Ragland and Sohal (1973), for example, showed that under certain conditions, male house flies without wings perish earlier than intact animals. The authors contributed this finding to leg muscle failure because of extended mating behaviour (see Introduction). This is also consistent with elevated longevity in less-active flies (Sohal and Buchan, 1981). Even if flies live longer, older flies do not necessarily retain their flight ability for longer. Lane et al. (2014) showed that flies with forced activity stopped flying earlier in their life than flies that were voluntarily active. Our data on flight activity might be shaped by these findings because the flies inside the container and the cage were continuously stimulated by flying conspecifics. This might explain the earlier onset of wing damage in the container compared with the free-flight experiment. Altogether, males are more likely to experience wing damage than females, especially when males are kept in groups (Sohal and Buchan, 1981).

In nature, many insect wings are damaged, but in only a few cases is the damage severe (Anderson and Keyel, 2006; Rajabi et al., 2017; Rehan and Richards, 2010). To compare the damage pattern in our house flies with that of animals living in nature, we collected flies from the field between July and September 2018 using nets and traps at different horse stables around the University of Rostock, in Rostock Zoo, and around a house in Güstrow, Germany. Wing damage in these flies showed similar patterns to those in our experiments but damage was typically less pronounced. Typical patterns showed a loss of wing area at the wing tip and the membranous wing's trailing edge, similar to the wings shown in Figs 5 and 6. In total, 65% (161) of the trapped animals had intact wings, 24% (60) had only small damage and only 4% (10) of animals showed elevated wing damage. The finding that wing wear starts at the wing tip and trailing edge is also consistent with previous findings on bumblebees (Cartar, 1992). Moreover, Rajabi et al. (2017) reported wild dragonflies with both small (∼10% area loss) and severe damage in single wings with a loss of up to ∼3/4 the initial wing area. Noteworthy, insect wings may be damaged by losing surface material without losing wing area. This is due to cracking and tearing of veins and wing membranes, which has recently been reviewed by Rajabi et al. (2020).

As wing damage accumulates over time, the speed of area loss often changes with increasing age of an insect. Several previously published studies found evidence for an increasing rate of damage with increasing time, resulting in an exponential loss in area (Foster and Cartar, 2011; Higginson and Barnard, 2004). However, other authors reported an opposite trend in which wing area loss asymptotically saturates with increasing age. Some of these findings may be explained by the use of an ordinal instead of an interval scale for scoring damage (e.g. Mueller and Wolf-Mueller, 1993; O'Neill et al., 2015). Other authors suggested that a frayed wing tip becomes more compliant and is thus less prone to damage (Mountcastle and Combes, 2014). We observed that in house flies wing damage typically stabilised after an exponential loss. Our logistic function in Fig. 6 combines these two trends, showing an exponential increase at an early stage of wing area loss and an exponential decrease at the late stage. This finding is consistent with (i) the initial increase (∼5% area loss) in honey bees (Higginson and Barnard, 2004), (ii) the assumption that existing damage spreads more easily than initial damage (Foster and Cartar, 2011), and (iii) an increased number of wing collisions owing to changes in wing kinematics and manoeuvrability (Foster and Cartar, 2011). If haemolymph-carrying veins are damaged, wings may even desiccate (Pass et al., 2015; Steppan, 2000; Vincent and Wegst, 2004). Dried wings are more brittle, more stiff, have altered mechanical properties, and should thus exhibit more damage (Hou et al., 2015). As haemolymph circulation also decreases with increasing age of an insect, desiccation is thus a factor that might explain our findings (Arnold, 1964; Salcedo and Socha, 2020).

However, the steepness of wing area loss at t_tp (slope) suggests that wing wear in house flies flying in the cylindrical containers is approximately 20 times faster, once begun, than in flies flying inside the larger free-flight cage (Table 1). This result supports the previous assumptions that collisions are the main source of damage to appendages, as suggested for fish (Latremouille, 2003), birds (Francis and Wood, 1989) and several insect species (Burkhard et al., 2002; Foster and Cartar, 2011; O'Neill et al., 2015; Rhainds and Brodersen, 2012). This finding is in line with the difficulties of using wing damage for age grading in wild-caught specimens (Allsopp, 1985; Hayes and Wall, 1999; Lee et al., 2006; Nalepa, 2012; Rhainds and Brodersen, 2012). The cases in which age was successfully derived from wing damage in insects may potentially be explained by flight activity and the number of object collisions (Hargrove, 2020; Hayes et al., 1998; Irvin and Hoddle, 2009; Mueller and Wolf-Mueller, 1993). We have no evidence that our handling procedure and anaesthesia facilitated wing damage.

Insects compensate for the loss of wing area by changes in wing kinematics. As already mentioned in the Introduction, pronounced changes include an increase in both wingbeat frequency and amplitude in the damaged wings. If the two wings are differently damaged, amplitude may also decrease in the wing having the larger surface area. Wing damage often attenuates flight performance such as speed (Combes et al., 2010; Fischer and Kutsch, 2000; Haas and Cartar, 2008; Jantzen and Eisner, 2008; Muijres et al., 2017), acceleration and manoeuvrability (Combes et al., 2010; Haas and Cartar, 2008; Mountcastle et al., 2016), and load-lifting capacity (Buchwald and Dudley, 2010; Fernández et al., 2017; Johnson and Cartar, 2014; Roberts and Cartar, 2015). The degree of compensation is limited by several factors, including mechanical constraints of the wing hinge, maximum frequency of muscle contraction, power output of flight muscles, limits involved in left–right asymmetries, but also by the loss of sensory feedback from the wings (Combes et al., 2010; Lehmann and Dickinson, 1997).

Conventional quasi-steady aerodynamic theory in insect flight links lift production to wing area and characteristic wing velocity. Previous studies suggested maximum lift production in insects of twice the body weight (Lehmann and Dickinson, 1998; Marden, 1987). We thus predicted that an area loss of more than ∼50% should hinder active flight in house flies. Fig. 9C shows that the lower quartile of flight trials corresponds to =0.89 (∼10% loss) and the lower whisker to =0.76 (∼25% loss). The minimum for an actively flying animal was =0.67 (∼33% loss). These data are somewhat below the predicted value but the last of these is similar to the 33% found by Ragland and Sohal (1973) for flying insects. Moreover, our finding is clearly within the range from ∼20% to more than ∼50% area loss that has previously been published for various insect species (Fischer and Kutsch, 2000; Haas and Cartar, 2008; Jantzen and Eisner, 2008; Kassner et al., 2016). As asymmetrical wing area loss may also unbalance aerodynamic force production on the left and right body side (Figs 8C,E,F and 9E,F), flight ability may additionally be attenuated by the induction of turning moments around the yaw, pitch and roll axis of the fly body. Fig. 9E shows that the area difference for ‘flying’ animals was ∼0.02 median and ∼0.26 maximum for a single fly. In ‘non-flying’ animals, median area asymmetry was ∼0.40. If flight capacity is mainly constrained by asymmetrical force generation, these differences might reflect the limit that a house fly is able to compensate by neuromuscular-induced changes in wing kinematics such as alterations in wingbeat amplitude, angle of attack, and rotational speed and timing at stroke reversal. We found similar values for the left–right difference in 2nd moment of area with a threshold of 0.4–0.6 between flight and no flight (Fig. 9F). This result is in good agreement with the finding in Drosophila hydei that a ∼50% unilateral loss of the wing's 2nd moment of area still allows active flight (Muijres et al., 2017). In house flies, 50% of the 2nd moment of area is 18 mm⁴ mg⁻¹, which is close to the lower whisker of the box blot for ‘flying’ flies (Fig. 9D).

In sum, our study has confirmed previous findings on wing damage and its significance for flight in insects. A missing link between wing damage and flight ability is elaborate data on aerodynamic forces and moments, including the changes in power requirements for flapping damaged wings. This is of ecological interest because previous studies on fish, birds, bats and insects showed that damage to appendages does not necessarily increase energy consumption during locomotion (Fu et al., 2013; Kingsolver, 1999; Voigt, 2013; Webb, 1973). Computer simulations of a phorid fly with unilateral wing damage, for example, showed that total aerodynamic power did not differ much between intact and damaged animals (Lyu et al., 2020). Similar findings have been reported for bumblebee wings. Although wing wear in bumblebees leads to changes in aerodynamic force coefficients and power requirements for flight (induced and profile power), the overall effect was small compared with intact conspecifics (Hedenström et al., 2001). By contrast, flight with moulting wings in humming birds and clipped wings in large moths requires higher metabolic rates than flight with intact wings (Chai, 1997; Fernández et al., 2017). Thus, to further investigate the significance of wing damage in house flies, the presented data are currently being used in a follow-up study to determine the changes in aerodynamic forces, moments and power expenditure in flight with damaged wings using a computational fluid dynamics approach.

The authors appreciate help with the experiments and sound analysis algorithm from student intern Evan Grant. We also thank the technicians Bärbel Redlich-Witt, Birgit Wobith and Annegret Mahrwald for their help with fly breeding. Anna Grießbach helped with wing area tracing. H.-N.W. is grateful to Wolf Hanke, Alexander Meister, Sixin Zhang and Kai Schneider for discussions on data analyses. The authors also received helpful input from three anonymous reviewers and appreciate the work put into copy-editing this article.