The role of experience in flight behaviour of Drosophila Free

The study of animal behaviour differentiates between two types of behaviours: innate genetically determined behavioural sequences, so-called stereotypic behaviour or fixed-action patterns, and acquired or learned behaviour. During the early half of the last century, the field of ethology was dominated by the theory of instinctive behaviour proposed first by Lorenz(Lorenz, 1937) and later supported by Tinbergen (Tinbergen,1951), which claimed that most of the observed behaviours in animals were instinctive and therefore not influenced by experience. However,the theory of instinctive behaviour was soon recognised as being too simplistic, at least for vertebrates where a large part of behaviour arises from interactions with the environment and with conspecifics (Lehrman, 1953). Thus today, it is recognised that the acquisition of complex motor skills in vertebrates results from the interplay between the physiological development of the locomotor apparatus and long-lasting motor practice.

In humans it typically takes several years of practice to become skilled in dexterous tasks and sport disciplines, and birds learn to fly efficiently within days or weeks after leaving the nest(Yoda et al., 2004; Hamer et al., 2002). Traditionally, flight in birds is assumed to be correlated with maturation and thus with the development of muscles and neurons, without any learning taking place (Grohmann, 1939). A recent study on the brown booby Sula leucogaster, however, showed that flight duration decreases and the proportion of time spent gliding increases, with the number of days since fledging, indicating that flight becomes more efficient with experience in this seabird(Yoda et al., 2004). Motor skills in insects and other invertebrates, by contrast, are still widely recognised as being predominantly innate fixed-action patterns. Evidence for this view comes from a large body of observations on a wide group of animals. Orb spiders, for example, build their webs by following a stereotypic set of rules without relying on any previous web-building experience(Witt et al., 1972). Other examples include the stereotypic set of locomotor actions used by many walking insects, such as stick insects (Blaesing and Cruse, 2004) and fruit flies(Pick and Strauss, 2005), when they encounter and cross gaps, the initiation of flight in locusts when sensory hairs on the head detect head wind(Wilson, 1961) and the landing response in flies occurring in response to expanding visual flow fields(Borst, 1990). The improvement in flight capacity of butterflies during the first days after emergence, by contrast, results from cuticle hardening and not from motor learning(Petersen et al., 1956).

During the past few decades several studies have repeatedly emphasized that these stereotypic behavioural patterns yield some degree of plasticity and are potentially shaped by previous experiences. Spiders, for example, make small changes to their web-building routine resulting in functional changes in the geometry of their webs in response to a large set of environmental and physiological factors including previous prey experiences(Venner et al., 2000). Similarly, the jumping spider Portia fimbriata shows trial-and-error learning, in which previous success of motor behaviours, such as the escape from islands by jumping or swimming, determines future actions in a similar context (Jackson et al.,2001). Insects also readily modify stereotypic motor sequences as a result of sensory input. Tethered locusts experience an immediate decrease in wing stroke frequency after ablation of hind wing proprioceptors but they regain the initial frequency within days, when the insects compensate for the loss by relying on other proprioceptors(Gee and Robertson, 1996). In another study, experimental manipulations of the feedback loop in tethered locusts showed that these animals adapt to an artificially created asymmetry between the wing steering muscles and that this adaptation is remembered in subsequent open-loop flight (Möhl,1988). Locusts also seem to possess an internal representation of the movement of their hind legs during scratching movements that remains fixed to body coordinates (Matheson,1997).

Flies display a remarkable repertoire of complex aerobatic behaviours including hovering, tracking, collision and escape responses, and an elaborate landing programme mediated by visual pathways and haltere feedback(Frye, 2007; Tammero and Dickinson, 2002a; Wagner, 1986). Forward flight is, furthermore, characterised by stereotypic rapid saccades in which the fly changes yaw heading between 90 deg. and 120 deg. in 50–130 ms(Mronz and Lehmann, 2008; Fry et al., 2003; Tammero and Dickinson, 2002b; Mayer et al., 1988). Thus,efficient flight in Drosophila requires substantial motor skills as well as strong coordination between sensory input and motor output. To our knowledge no previous studies have investigated to what degree flight behaviour in these flies, or indeed in any other insect, is acquired during post-pupal development. Recent studies, however, demonstrate that there is plasticity and learning in the motor system of adult fruit flies. Wang et al.(Wang et al., 2003), for example, found that the animals learn to modulate their torque output in response to a visual pattern when trained with heat punishment. Fruit flies are, moreover, able to compensate for wing damage and remain airborne despite having up to half of their wings removed(Bender and Dickinson,2006).

In this study we investigated the question of whether flight of the fruit fly is purely innate or improves with practice. We addressed this question by comparing a variety of flight parameters such as wing kinematics, flight speed and acceleration, yaw torque-equivalent stroke amplitude responses during object tracking, and saccadic turning angles and rates between experienced and naive flies without previous flight experience. This was done under both tethered flight conditions in a simulator and free-flight conditions in a free-flight arena.

The tested flies were selected from a laboratory colony, maintained at 24°C on commercial Drosophila feed (Carolina Biological,Burlington, NY, USA). Two days before hatching, five pupae were carefully transferred, without directly touching the pupae, into a standard 35 mm diameter Drosophila vial (control animals) and another five into a 80 mm×60 mm×2 mm (length×width×height) custom-built flat Perspex rearing chamber (naive animals, Fig. 1A). Both vials were kept under identical conditions in a breeding chamber under a 12 h: 12 h light:dark cycle. In both chambers hatched flies had access to food offered in a 35 mm diameter circular container. Free walking area was not very different between the two chambers, amounting to approximately 4790 and 9840 mm² for the standard vial and the custom-build chamber, respectively. By contrast,free total space was larger in the standard (37×10³mm³) than in the custom-built chamber (9.6×10³mm³). Owing to the low population density and the relatively large surface area of both chambers, the flies could readily walk around, carry out courtship behaviour and also mate. Although we cannot completely exclude subtle changes, we did not observe differences in the behaviours between the two groups, except that the flat chamber did not allow the naive group to gain flight experience. Observations of undisturbed flies in the standard vials showed that they initiated short flight bouts, thus giving them the chance of free flight prior to our experiments. We did not record flight time and distance during the flight bouts in the controls, thus it is possible that these flies had some variations in their actual flight practice prior to testing. All flight experiments were carried out on 4- to 5-day-old female Canton S wild-type Drosophila melanogaster Meigen.

A close visual inspection of the animals using image analysis software(Scion Image, Frederick, MD, USA) revealed no major morphological differences between the two groups. Thorax width, measured between left and right frontal notopleural setae amounted to 0.738±0.006 mm and 0.745±0.047 mm in control and naive flies, respectively (N=16, two-tailed t-test on means, P=0.67), and thorax length, measured between the most frontal row of hairs on the thorax and the tip of the scutellum was 0.939±0.062 mm and 0.951±0.061 mm for the two groups, respectively (means ± s.d.). Similar results were obtained for wing length (controls: 2.35±0.08 mm, N=19; naive animals:2.37±0.14 mm, N=33; two-tailed t-test on means, P=0.70) and wing area (controls: 1.77±0.14 mm², N=19; naive animals: 1.79±0.21 mm², N=33;two-tailed t-test on means, P=0.70). Although body mass was not significantly different between the two groups (controls: 1.12±0.25 mg, N=36; naive animals: 1.06.37±0.22 mg, N=60;two-tailed t-test on means, P=0.25), control flies tended to be heavier than the naive animals. Altogether, the above values suggest no major differences in flight muscle volume, power requirements for lift production and aerodynamic effects as a result of wing size for the two groups.

The flight simulator used in this study has already been described in detail by Lehmann and Dickinson (Lehmann and Dickinson, 1997), so only a brief introduction is given here. The tungsten rod on the tethered flies fitted into a holder that placed the fly in the middle of a cylindrical flight simulator, 125 mm high and 150 mm in diameter (Fig. 1B). The inclination of the holder ensured that the fly was in a hovering position,with a body angle of 60 deg., so that the stroke plane of the wings coincided with the horizontal (David,1978). An infra-red diode above the flight simulator cast shadows of the wings on an infra-red-sensitive mask connected to a wing stroke analyser that provided wing stroke amplitudes and frequency for each single stroke cycle. We calibrated wing stroke amplitudes by digitizing the wing positions on video images of the flying animal recorded by an infra-red-sensitive camera. The voltages coming from the infra-red light path were subsequently converted into degrees using linear regression on the digitised data. Both digitisation and the final calibration were done with custom-built Origin (Version 7, OriginLab corporation, MA, USA) routines.

A conventional computer generated a 12 deg. wide black bar foreground on a uniform green background (Fig. 1B). The image displayed in the simulator was updated every 8 ms and flickered with a frequency of 1000 Hz, which is well above Drosophila's flicker fusion rate of around 200 Hz(Autrum, 1958). The pattern could move in both an open loop, where the stripe rotated uniformly at a speed of 36 deg. s^–1, independent of the fly's behaviour, and in a closed loop, where the fly actively controlled the azimuth velocity of the pattern by changing the relative difference of the stroke amplitude between both wings (left minus right). Throughout the manuscript, a value of 0 deg. stripe position indicates the frontal azimuth position of the compound eyes,whereas –90 deg. (90 deg.) angle indicates the lateral position on the left (right) body side. The relationship between angular velocity of the image and the wing stroke difference was controlled by a physics engine that simulated physical conditions similar to free flight such as the frictional damping on body and wings and inertial moments of the fly. In all experiments we used a damping coefficient of 520×10^–12 Nm s and an inertial coefficient of 0.52×10^–12N m s²,because a previous study showed that these values yielded the most precise visual control feedback of tethered Drosophila(Hesselberg and Lehmann,2007).

The free-flight arena consisted of two concentric, 200 mm high acrylic cylinders with 140 mm and 190 mm diameter, respectively [for a detailed description see Mronz and Lehmann (Mronz and Lehmann, 2008)] (Fig. 1C). The inner translucent cylinder was immoveable and prevented the flies from landing on the second outer cylinder on which a visual 8 deg.×8 deg. wide random square pattern was attached. Three surrounding circular fluorescent light tubes illuminated the pattern cylinder from outside, while the cylinder's frosted surface established diffusive and almost uniform light conditions within the arena. When flies took-off from the central platform, a high-speed camera recorded the flight path of the fly from above at 125 frames per second with a resolution of 1020×1020 video pixels. For each animal we recorded an 8 s sequence using Pixoft video capturing software (Pixoft V3.0, Birmingham, UK). Since we aimed to evaluate the flies at their maximum locomotor capacity, the outer pattern cylinder rotated at 500 deg. s^–1 counter-clockwise to elicit visual optomotor behaviour (Mronz and Lehmann,2008). We started the rotation before the fly took-off so that a uniform speed of the cylinder was achieved when the insect initiated flight behaviour.

We analysed the recorded video images using custom-built software written in Visual C++ (Microsoft, Redmond, USA) and commercial imaging components(Matrox Imaging Library Mil 7.5, Dorval, Canada), extracting data on (i) the position of the fly's centre of gravity, (ii) the angular orientation of the longitudinal body axis, and (iii) the size of the fly using `blob' analysis from each frame. These data were subsequently converted into (i) total flight distance covered, (ii) lateral distance to the arena wall, (iii) flight altitude, (iv) horizontal velocity and (v) acceleration, and finally the animal's turning (angular) velocity around the yaw axis. Flight altitude was estimated from linear regression [y=–73+1.05x, where x is the `blob' size in pixel(Mronz and Lehmann, 2008)] and used to correct the fly's x/y position to obtain exact velocity estimates. Similar to previous studies, we scored flight manoeuvres as saccades when the turning rate of the animal exceeded 1000 deg. s^–1. From yaw orientation of the fly's longitudinal body axis, we estimated the following saccade parameters: (i) rotational direction,(ii) total turning angle, (iii) saccade frequency and (iv) angular velocity during the saccade.

The flies were transported in small vials and released individually into a tube leading to the centre of the free-flight arena by means of a microprocessor-controlled gate. The animals voluntarily walked through the tube and into the arena and usually took off within a second after emerging in the centre. By contrast, the flies used in the flight simulator were anaesthetised by cooling them to approximately 4°C and tethered between the head and the notum to a 7.3 mm long, 0.13 mm diameter tungsten rod using UV-light activated glue. The flies were taken to the flight simulator immediately after tethering in which they were allowed to recover. Only 13% of the controls (3 of 23) and 14% (5 of 36) of the naive animals did not initiate flight voluntarily within a 30 min time period after moving. In these cases,we initiated flight by gently blowing air on the animal's wings. Mean recovery time was similar in both groups (controls: 16.5±10.5 min, N=19; naive animals: 20.8±11.3 min, N=33; two-tailed t-test on means, P=0.18). The experiment started as soon as the animals initiated wing movements.

Each simulator experiment was 220 s long and divided into the following flight sequences: (i) an initial open-loop feedback sequence of 20 s split into 10 s, in which the stripe rotated one full circle in the clockwise direction and a further 10 s, in which it rotated a full circle in the counter-clockwise direction, (ii) 180 s of closed-loop feedback and (iii) a final open-loop feedback sequence similar to the initial score(Fig. 2). In the open-loop sequences the stripe always appeared directly behind the fly and then moved into its visual field. Flies that had an intermediate flight stop in one of the sequences were excluded from the analysis. We found no major difference in the relative number of exclusions between control and naive flies, suggesting that both groups had similar motivation for flight (controls: 42%, naive animals: 34%). In general, we carefully ensured that naive flies were given no opportunity to fly before they had been tested either in the flight simulator or in the free-flight arena. Ambient temperature was similar in the experiments; 22.9±1.32°C (control) and 23.3±1.36°C(naive; means ± standard deviation, two-tailed t-test, P=0.28).

In open-loop conditions, we found a distinct correlation between stripe position and the difference in stroke amplitude between both wings, as expected from previous work (Heisenberg and Wolf, 1984). For further comparison, we thus fitted a Boltzmann wave to the individual responses within a ±120 deg. visual range using the Boltzmann fitting function in Origin, which is based on the simplex search method (Lagarias et al.,1998). To test for changes in the angular velocity profile during flight saccades, we calculated the means derived from data pooled from–80 ms before to 80 ms after peak turning velocity. For statistical tests we used SPSS (version 10.0, SPSS, 1999) and Origin at a significance level of 5%. If not stated otherwise, data are given as means ±standard deviation.

After being placed in the arena, the majority of flies voluntarily initiated flight behaviour and attempted to fixate the stripe in the frontal region of their compound eyes. We analysed the behavioural data separately for the two open-loop and the longer closed-loop sequences to highlight the following aspects. First, data from the initial open-loop sequence (0–20 s) allowed us to determine the initial stroke amplitude-to-stripe position representation in the two groups and thus to estimate the effect of previous flight experiences. In this analysis, we evaluated wing kinematics and flight force production. Second, during closed-loop (20–200 s) flight we determined wing kinematics and fixation behaviour towards the black stripe in order to identify any potential differences in the ability of the two groups to visually control the azimuth velocity of the object. Third, we used the data recorded in the final open-loop sequence (200–220 s) to highlight potential changes in the flies' internal gain for stroke amplitude-to-stripe position, derived from the previous closed-loop experience. Fig. 3 shows typical time traces of the kinematic parameters for the three flight conditions: the relative difference between left and right stroke amplitude, mean stroke amplitude of both wings and stroke frequency, for an experienced(Fig. 3A–C) and a naive fly (Fig. 3D–F).

During open-loop flight, when the stripe rotated at constant speed across the flies' visual field, the flies typically steered towards the stripe by changing the relative difference between left and right stroke amplitude,producing a sinusoidal response curve (Fig. 4). Mean stroke amplitude and stroke frequency, by contrast, were less dependent on stripe position. We found only small, insignificant changes in mean amplitude difference between experienced and naive flies in both open-loop sequences (t-test, 0–20 s: P=0.59;200–220 s: P=0.65; Fig. 4A, inset). The same holds for mean stroke amplitude(t-test, 0–20 s: P=0.23, 200–220 s: P=0.35; Fig. 4B,inset) and stroke frequency (t-test, 0–20 s: P=0.75,200–220 s: P=0.96; Fig. 4C, inset). In addition, we found no significant difference for any of the three kinematic parameters between the two open-loop flight sequences [t-test, control (naive), stroke amplitude difference: P=0.57 (0.77), mean amplitude: P=0.12 (0.07), frequency: P=0.87 (0.58)]. To statistically evaluate the amplitude response to the moving stripe, we fitted a Boltzmann wave to the amplitude difference between ±120 deg. of the visual field and scored both the total angular response (A_P; Fig. 5A) and the offset of the fit curve (d_F; Fig. 5A). Fig. 5B shows that the response amplitude in the naive flies was 29.4±13.7 deg. (N=32 flies)and thus significantly smaller than the controls (39.1±14.8 deg., N=18, t-test, P=0.03). For the final open-loop sequence (200–220 s flight time), we obtained slightly smaller values,but did not find a significant difference between those measures (control:25.0±17.0 deg., naive: 22.8±17.4 deg., t-test, P=0.66). The curve offset (d_F) was not statistically different between control and naive animals or between both open-loop sequences (t-test, P>0.05; Fig. 5B).

A possible explanation for the decrease in response amplitude during open-loop flight of naive flies is a reduction in the animals' capacity to produce elevated flight forces. However, this seems not to be the case. We derived total flight force production from the squared product between mean stroke amplitude and stroke frequency using a conventional quasi-steady approach (Ellington, 1984). Since mean lift coefficient, C̄_L, in tethered flight depends on mean stroke amplitude, Φ, we adjusted this parameter accordingly[C̄_L=–5.46+43.5×10^–3Φ(Lehmann and Dickinson,1998)]. Mean total flight force averaged over the entire 220 s flight time was similar in both experimental groups (control: 7.97±2.55μN, naive: 7.06±1.99 μN, t-test, P=0.17)including peak force (maximum 5% of all force data) during open-loop(two-tailed t-test, 0–20 s: P=0.66, 200-220 s: P=0.37; Fig. 6, inset)and closed-loop (control: 11.1±3.24μN, naive: 10.0±3.22μN, t-test, P=0.24) flight. Moreover, we did not find any correlation between response amplitude and peak force production in both open-loop sequences of control animals (linear regression fit, 0–20 s: y=49.8–0.81x, R²=0.06, P=0.31, N=18 flies; 200–220 s: y=9.32+0.97x,R²=0.16, P=0.08, N=32 flies) and naive flies(linear regression fit, 0–20 s: y=35.3–0.41x,R²=0.02, P=0.46, N=20 flies; 200–220 s: y=11.1+1.30x, R²=0.08, P=0.13, N=32 flies; Fig. 6),altogether suggesting that response amplitude and thus maximum yaw torque as shown in Fig. 5 is independent of the animal's maximum locomotor capacity.

While fixating the black stripe under closed-loop visual feedback condition in the frontal region of their visual field, both groups exhibited similar mean stroke amplitude (control: 151±11.7 deg., N=20 flies;naive: 149±9.84 deg.; N=30 flies; t-test, P=0.40) and almost identical stroke frequencies (control:206±11.4, naive: 205±15.3 Hz, t-test, P=0.95). However, similar to what had been measured during the open-loop sequences, we found subtle changes in steering control behaviour between the two groups. The examples in Fig. 7 show that naive flies used more corrective steering when fixating the stripe, which resulted in a slight shift in the fast Fourier transformed (FFT) spectrum of left-minus-right stroke amplitudes towards higher frequencies above approximately 2 Hz and a pronounced decrease in FFT amplitude at frequencies below this value (Fig. 8). On average, naive flies exhibited a 53% higher absolute amplitude difference between wings (control: 2.30±1.30 deg., N=20; naive:4.84±1.57 deg., N=30; t-test, P<0.001)during closed-loop flight than the controls. Consequently, stripe fixation behaviour was less precise in naive flies as shown by the means of absolute stripe position (control: 26.1±14.7 deg., N=20; naive:37.2±7.73 deg., N=30; t-test, P=0.001,20–200 s flight). Nevertheless, the overall differences between both animal groups were relatively little, again supporting the assumption of a genetically pre-programmed locomotor apparatus.

Since potential differences in aerial performance should be most visible at elevated motor activity of the animals, when most of the locomotor reserves are needed, we compared flight of the two groups in a free-flight arena rotating at 500 deg. s^–1 (see Materials and methods). Previous data have already shown that in Drosophila flying in a stationary arena, flight speed and turning rate is typically 40 and 70% below the performance measured during optomotor response, respectively(Mronz and Lehmann, 2008). In this regard, we here score the animals at extreme situations, comparable, for example, to escape manoeuvres from aerial predators in the wild.

The flies initiated flight almost as soon as they had entered the arena and followed the rotating panorama on circular flight paths(Fig. 9A–F). We observed no major differences between naive and experienced flies in take-off latency or transfer probability inside the arena(Fig. 9G,H). Naive flies,however, flew approximately 23% slower than experienced flies and thus covered a significantly shorter distance (Table 1). Despite the lower flight speed, absolute horizontal acceleration was 3.3 times higher in the naive animals whereas peak cruising speed was not significantly different between the two groups (0.84 vs0.73 m s^–1). Flight by the naive flies was thus either more erratic or manoeuvrable, with higher fluctuation in forward speed. To achieve optomotor balance, the control group closely adjusted its mean turning rate(519 deg. s^–1) to arena velocity (500 deg. s^–1), which is similar to what had been found in previous experiments (Mronz and Lehmann,2008). By contrast, naive flies compensated for only approximately half of the stimulus speed (270 deg. s^–1), although their absolute turning rate was similar to the rate of the control group (739 vs 782 deg. s^–1, P=0.46; Table 1). Superficially, the latter finding suggests that naive flies were handicapped in maintaining their flight heading during smooth and saccadic turns, similar to what is found in tethered flight. We observed no difference in mean distance between the flight path and the arena centre – a measure that is thought to reflect the production of centripetal forces during turning flight in a circular arena(Mronz and Lehmann, 2008). Moreover, none of the tested parameters was significantly different in the two groups during the initial take-off period (0–0.5 s; Table 1).

In general, both groups performed saccadic flight turns in the counter-clockwise direction (ccw) but also, less frequent, clockwise (cw),against the direction of the rotating arena(Table 1). Fig. 9I,J shows that the angular velocity profiles were almost symmetrical around peak turning rate for both saccade directions (0 ms), but yielded differences between groups of flies during clockwise saccadic turns (Fig. 9I). Clockwise rotations of the controls were preceded by small counterturns in the direction of the visual panorama (black arrows in Fig. 9I), whereas this subtle modulation in velocity was missing in the naive flies. In contrast to horizontal velocity, peak turning rate within the saccade was similar under all conditions but clockwise total turning angle was significantly larger in the naive (157 deg.) than in the control group (108 deg.; Table 1).

In the present study we investigated the role of experience in the flight behaviour of the fruit fly Drosophila. The tethered and free-flight data suggest that most flight behaviours seem to be genetically predetermined because all animals exhibited similar mean stroke amplitude and stroke frequency, and similar open-loop responses to visual stimulation (Figs 4, 5, 6, 7, 8). In short free flight bouts,peak saccadic turning rate, angular acceleration, peak horizontal speed and flight altitude were also similar in naive and control flies(Fig. 9, Table 1). Nevertheless, flight in naive flies without previous flight experience differs from that of control flies in several key parameters. A consistent feature of all tested flies in the simulator was the position-dependent modulation in yaw torque towards the stripe moving in an open loop (Fig. 4A). Thus, even animals without prior flight experience obviously have an internal representation of how the movement of their wings translate into real world body movements. This visuo-motor gain – or response strength – is not a constant within the visual field of the animal but increases when the stripe moves in the frontal ±90 deg. visual field,which corroborates previous results (Fig. 5) (Heisenberg and Wolf,1984; Reichardt and Poggio,1976). In contrast to the control group, however, naive flies showed inferior flight control when fixating the stripe in open-loop conditions (25% less amplitude difference; Fig. 5) and performing optomotor behaviour during free-flight conditions(Table 1), both indicating a flight-dependent adaptation of the visuo-motor gain in the experienced animals. However, the inferior fixation in the first open-loop sequence vanished after flying the animals under closed-loop conditions, because in the final open-loop sequence the behaviours of naive and control flies were no longer significantly different. Moreover, naive fruit flies had less precise yaw torque-equivalent amplitude control than the experienced animals(Fig. 7).

In the following sections we discuss the internal representation of motor actions in greater detail, presenting two alternative hypotheses to explain our results: the `neuronal experience' hypothesis suggesting that there are some elements of learning and fine-tuning involved during the first flight experiences in Drosophila and the `muscular exercise' hypothesis. Owing to the lack of experimental data, we will not further discuss other hypotheses such as changes in viscoelastic properties of the thorax and/or wings or in the concentration of neuromodulators (`neuromodulatory hypothesis'). Octopamine, for example, has profound effects on the physiology of muscles and neurons, and could be elevated in controls as a result of pre-experimental flight activity (Chapman,1998; Brembs et al.,2007). Although we cannot completely exclude this explanation, the hypothesis assumes that flight bouts at least 25–30 min prior to the experiment affect the initial open-loop testing.

In addition to the difference in the initial visuo-motor gain (0–20 s flight time) between the two experimental groups during open-loop flight(Fig. 5B), naive flies also showed subtle differences in closed-loop flight when flown under tethered-(20–200 s) and free-flight conditions(Table 1). Most notably, naive flies showed a higher degree of active steering during fixation of the stripe when tethered than the controls, which resulted in a significant higher absolute difference of wing stroke amplitude between the two wings. Naive flies may potentially compensate for this reduction in fine control of their motor actions (stroke amplitude) by increasing the temporal rate of steering manoeuvres. The distortion of the FFT spectrum shown in Fig. 8E might indicate such a process. The latter behaviour partly restores the reduction in turning efficiency for yaw control and thus flight heading stability. This result may suggest that the changes in yaw control and thus the changes in fixation behaviour result from an active behavioural process supporting the `neuronal experience' hypothesis, rather than from a reduction in the ability of the naive animals to maximise their stroke amplitudes.

Although Fig. 5B clearly shows that the difference in left-minus-right stroke amplitude quickly ceases with flight time, the underlying sensu-motor mechanism for this decay is less clear. There are at least two possible explanations for this finding. First,the response difference vanishes because the naive animals experience closed-loop feedback conditions between both open-loop sequences, allowing the flies to adjust their neuromuscular system similar to that of control animals. This would require short-term adjustments via physiological changes in neural activity within 3 min of closed-loop flight. In this scenario, we would also expect that the response difference between naive and control flies persists when repetitively testing the animals under open-loop conditions,without the application of closed-loop feedback conditions. Second, an alternative explanation would be that the initially scored response difference(0–20 s; Fig. 5B)decreases with increasing flight time, independently of the experience collected during the closed-loop feedback flight sequence. Experiments planned in the future should allow us to distinguish between these long- and short-term adjustments, and thus to possibly clarify the nature of the measured changes.

The above conclusion is also supported by two further findings. First,freely flying naive flies had a more erratic flight path due to an apparent loss of precise heading control and showed insufficient optomotor response during free flight in the rotating arena, although they flew at the same mean altitude and thus lift production (Table 1). Second, naive animals exhibited changes in relative frequency and velocity profile of clockwise and counter-clockwise free-flight saccades,respectively. Experienced flies, for example, elicited approximately 90% of the saccades in the counter clockwise direction and thus in the direction of the rotating arena. By contrast, naive flies more often turned clockwise (73%counter clockwise rotations), contrary to what would be expected from optomotor stimulation, and also yielded higher variances in their responses. A possible explanation for the observed changes is that flight saccades in naive flies are more often triggered by retinal slip due to image expansion or contraction on the lateral eye regions, arising from more frequent approaches towards the visual panorama (Duistermars et al., 2007).

The finding that in the entire free-flight sequence, mean horizontal speed in naive flies is approximately 23% less than in the control group(Table 1), may support the muscular exercise hypothesis mentioned before. This theory suggests that the weaker response in naive flies results from either a decrease in mechanical power output of the indirect flight muscle (IFM) or by physiological changes in flight control muscles caused by lack of exercise, rather than from lack of previous experience. Although our free-flight data are consistent with both the neuronal experience and the muscular exercise hypotheses, there is little well-founded evidence for the latter for the following reasons. First,previous studies on flight activity in house and fruit flies have repeatedly shown that an increase in muscular exercise causes an increase in mortality rate as a result of increased oxidative damage(Magwere et al., 2006; Yan and Sohal, 2000). Consequently, it is less likely that exercise improves muscle performance of the IFM. Second, although mean horizontal speed appeared to be weaker in naive flies than in the controls, peak performance was nearly identical in both groups and mean body acceleration was even approximately three times higher in the naive flies (Table 1). Third, under tethered flight conditions, we obtained slightly higher measures for peak aerodynamic force production in the controls than in the naive animals, superficially supporting the muscle exercise hypothesis(Fig. 6). However, statistical tests did not yield any statistically significant differences between the two measures. Collectively, our results indicate that naive animals appear to be capable of producing a similar maximum muscular mechanical power output by the IFM and control muscles as the controls, which implies, again, that their flight behaviour was shaped by the lack of previous experience and not by power attenuation.

Conventionally, it is not believed that invertebrates such as insects are capable of showing motor learning based on sensory input or from experience,as has been found in vertebrates (Yoda et al., 2004). Undoubtedly, most flight behaviour in Drosophila seems to be genetically predetermined, because naive flies show seemingly `normal' flight behaviour when randomly released into the laboratory. However, since relevant studies have repeatedly emphasised the evidence of behavioural plasticity and learning in insects, we should revise our simplistic view on insects as genetically determined flight machines. More recently, proof of complex motor learning in insects has been increasing and includes the learning of specific motor sequences in ants that had to navigate through a maze (Macquart et al.,2008), the ability of fruit flies to associate visual cues with a specific set of motor commands [yaw torque(Wang et al., 2003)], and pilot experiments on the update of long-lasting adjustments of the visuo-motor gain in wild-type and learning mutants of Drosophila, trained at various feedback conditions during tethered flight (T.H. and F.O.L.,unpublished results).

In general, data on the acquisition of everyday complex motor skills from previous experience are rare in insects. Our data on the plasticity of the effect of motor commands on body movements in Drosophila thus helps to understand the concepts of motor learning in model systems for flight. Although flying vertebrates apparently rely more strongly on the advantage of motor learning to improve their flight efficiency, agility and manoeuvrability– even small adjustments in control properties of the feedback cascade in insects may advance their fitness, for example, during escape from predators, mating flights or aerial combats. Therefore, studies on the plasticity of motor skills are also of interest for engineers who aim to design the future generation of micro-aircrafts based on flapping wing propulsion.