Chasing Behavior and Optomotor Following in Free-Flying Male Blowflies: Flight Performance and Interactions of the Underlying Control Systems

The chasing behavior of male blowflies after small targets belongs to the most rapid and virtuosic visually guided behaviors found in nature. Since in a structured environment any turn towards a target inevitably leads to a displacement of the entire retinal image in the opposite direction, it might evoke optomotor following responses counteracting the turn. To analyze potential interactions between the control systems underlying chasing behavior and optomotor following, respectively, we performed behavioral experiments on male blowflies and examined the characteristics of the two flight control systems in isolation and in combination. Three findings are particularly striking. (i) The characteristic saccadic flight and gaze style – a distinctive feature of blowfly cruising flights – is largely abandoned when the entire visual surroundings move around the fly; in this case flies tend to follow the moving pattern in a relatively continuous and smooth way. (ii) When male flies engage in following a small target, they also employ a smooth pursuit strategy. (iii) Although blowflies are reluctant to fly at high background velocities, the performance and dynamical characteristics of the chasing system are not much affected when the background moves in either the same or in the opposite direction as the target. Hence, the optomotor following response is largely suppressed by the chasing system and does not much impair chasing performance.


INTRODUCTION
Visual pursuit of small objects is an important task to be solved by many animals including humans, for instance when catching a baseball (e.g. McBeath et al., 1995;Land and McLeod, 2000;Shaffer and McBeath, 2002). Especially with respect to rapidity and virtuosity also various insect species show an impressive performance in pursuit tasks, for instance in the context of predation behavior (dragonfl ies: O'Carroll, 1993;Olberg et al., 2000Olberg et al., , 2005 praying mantis: Rossel, 1980) or mating behavior (various fl y species: Land and Collett, 1974;Collett and Land, 1975;Collett, 1980;Wagner, 1986b;Land, 1993a,b;. Visual pursuit requires the target to be detected, then to be fi xated in the frontal visual fi eld and eventually to be followed by appropriate movements of the eyes, the head and/or the entire body. In primates visual pursuit is generally assumed to be based on a smooth control system mediating continuous rotations of the eyes, sometimes in combination with head and body movements (Miles, 1997;Schweigart et al., 1997;Krauzlis, 2004). However, pursuit responses are not always smooth. If target velocity is too fast, a rapid saccadic eye movement shifts the image of the object of interest from an eccentric retinal location to the fovea before smooth pursuit commences again. Smooth pursuit is not a distinguishing ability of primates, since male blowfl ies as well as other insects employ very precise smooth head and body movements to keep the image of a moving target fi xated in the frontal visual fi eld even during highly aerobatic fl ight maneuvers. During pursuit the angular velocity of the male blowfl y was concluded to be Chasing behavior and optomotor following in free-fl ying male blowfl ies: fl ight performance and interactions of the underlying control systems schemes that the turning commands mediated by the pursuit system and the optomotor system, respectively, are added to the overall turning response of the animal. However, the interaction schemes differ considerably with respect to the way they cope with the interference of optomotor following with pursuit (Figure 1).
(1) Additive scheme ( Figure 1A): This scheme is the most parsimonious one, as the outputs of the two pathways are just added and do not interact in any other way. As a consequence, the angular velocity of the pursuer depends on both the error angle and the wide-fi eld motion caused by target pursuit in a stationary environment. The interference of the optomotor system is the smaller the larger the gain of the pursuit system relative to the optomotor gain (Virsik and Reichardt, 1976;Collett, 1980). It has been proposed that pursuit effi ciency may be enhanced if the two interacting systems differ with respect to their dynamical properties, the pursuit system being mainly active at high and the optomotor system mainly at low frequencies (Collett, 1980;Egelhaaf, 1987Egelhaaf, , 1989).
(2) Efference copy scheme ( Figure 1B): A so-called 'efference copy' is generated by every turning command of the pursuit system and thought to cancel out by subtraction the responses of the optomotor system to the visual consequences of target pursuit (von Holst and Mittelstaedt, 1950;Collett, 1980;Webb et al., 2004). Still, the pursuing fl y's angular velocity depends on both the error angle and wide-fi eld motion (Collett, 1980). (3) Suppressive scheme ( Figure 1C): In its most extreme version this scheme assumes that optomotor following is simply inoperative during turns evoked during pursuit; however, not as consequence of the limited dynamic range of the optomotor system, but as a result of active inhibition mediated by the pursuit system. Hence, any optomotor information is largely suppressed during pursuit (Webb et al., 2004). This disadvantage might be balanced by the advantage of simplicity in that the size of the expected optomotor signal does not have to be perfectly predicted as is the case for the efference copy scheme (Webb et al., 2004).
Although the wiring principles underlying these different interaction schemes differ considerably, it is not easy to distinguish between them experimentally when analyzing behavioral performance in stationary environments (Collett, 1980;Webb et al. 2004). However, if the environment is artifi cially moved around the animal in an experimental setup, the suppressive scheme, at least in its perfect form, may be distinguished from both the additive and the efference copy scheme. For the latter schemes the angular velocity of the animal and, thus, its pursuit performance should depend on the angular velocity of the background in contrast to the suppressive scheme.
Indeed, pursuit performance has been found to be affected by optomotor following in a number of studies on both primates and insects The steady-state pursuit eye velocity of primates induced by a small moving target was found to increase when the background moved in the same direction as the pursued target, and decreased when the background moved in the opposite direction as the target (Masson et al. 1995). Other behavioral studies indicate that pursuit eye movements are variously affected by a stationary or dynamic visual background (Collewijn and Tamminga, 1984;Keller and Khan, 1986;Kimmig et al., 1992;Masson et al., 1995;Mohrmann and Thier, 1995;Niemann and Hoffmann, 1997;Spering and Gegenfurtner, 2007). Despite discrepancies in detail, these studies indicate that the control systems mediating wide-fi eld following and small-fi eld pursuit in primates do not work independently of each other, at least at the behavioral level. In insects the possible impact of optomotor following on the performance of target pursuit has been addressed, so far, in only few studies. In praying mantis target pursuit is strongly affected by retinal background motion (Rossel, 1980). Similarly, in freefl ying hoverfl y Syritta or in tethered fl ying female housefl ies target fi xation and pursuit behavior are infl uenced by a simultaneously presented large-fi eld background motion (Virsik and Reichardt, 1976;Collett, 1980).
Here we show by systematic behavioral analysis that male blowfl ies chasing after moving targets behave differently in this respect than most other animals analyzed so far. Although free-fl ying blowfl ies respond strongly with following responses when exposed to external wide-fi eld motion, the same wide-fi eld motion does not much deteriorate the male fl y's performance in catching a target during extremely virtuosic chasing fl ights and appears not even to affect its fl ight style. In particular, it does not affect the fl y's angular velocity as might be expected by the additive and the efference copy scheme. Rather, it is concluded that during chasing the optomotor system is largely suppressed. This result is extraordinary as it reveals the high effi ciency of the blowfl y chasing system to successfully cope with situations that in several other systems deteriorate performance.

MATERIALS AND METHODS
All experiments were performed with several sets of about 20 blowfl ies (Lucilia sp.). Each set was used for several days to conduct all types of behavioral experiments with the same animals. The fl ies (6-to 10-day-old) were released in a cylindrical fl ight arena (diameter: 0.4 m, height 0.67 m) made of clear Perspex; the ceiling was homogeneously white. The arena was illuminated by boards of green LEDs that were used as a panoramic stimulus (see below). Two tungsten light heads (DLH4, 150 W, Dedo Weigert Film, Germany) additionally illuminated the arena from below. To ensure that these lamps did not much reduce the contrast of the visual stimulus and because red light is almost invisible for blowfl ies (Hardie, 1985), both lamps were fi tted with dichroic red-light fi lters (DFCOL2R, Dedo Weigert Film, Germany). The temperature in the fl ight arena ranged between 20°C and 29°C.
The upper 0.32 m of the fl ight arena was surrounded almost completely (see below) by a cylindrical stimulus device that consisted of two rows of ten circuit boards each, stacked over each other. Each board contained 48 columns and 30 rows of LEDs (Toshiba TLG234P, 2.5 × 5 mm 2 , emitted wavelength 565 nm). Each column could be switched on and off independently. The horizontal angular extent of each LED amounted to approximately 0.7° as seen from the center of the arena at the same height of the respective LED. The time until an LED reached a constant luminance value after switching it on or off was 20-50 µs. The device was programmed to generate apparent motion of a vertical grating with a spatial wavelength of 30° (corresponding to a spatial frequency of 0.033 cycles per degree) and 85% contrast. This grating was either stationary or rotated clockwise or counter-clockwise at 45°/s or 365°/s. These pattern velocities corresponded to temporal frequencies of 1.5 Hz and 12.2 Hz of the drifting grating. Generating one frame, i.e. addressing all groups of LED-columns serially, took approximately 370 µs. The cylindrical LED-array spanned 330° in azimuth and approximately 75° in elevation as seen from the center of the arena. 30° of the cylinder were left open for the lateral camera to see the inside of the fl ight arena.
A black sphere (diameter: 5 mm) was used as a chasing target ('dummy fl y'). If the dummy was fi xated by the chasing fl y at a distance of 50 mm it had an angular size of approximately 6°, i.e. much larger than the interommatidial angle measured for frontally looking ommatidia (Land and Eckert, 1985). The sphere was attached to the tip of a translucent perspex stick (length: 65 mm) that was fi xed to a non-transparent, white circular disk (diameter: 0.395 m) which could be rotated horizontally directly beneath the top end of the arena (vertical distance from upper rim of arena 45 mm; total vertical distance of the target from the upper rim of the arena 0.11 m). In the following the white disk will be referred to as 'ceiling' of the fl ight arena. The dummy was moved clockwise on a circular track (radius: 80 mm) at a speed of 1 m/s (i.e. about 700°/s), which is well within the speed range of fl ies. The dummy and the visual background, generated by the LED-array, could be moved individually or in combination.
Flying fl ies were fi lmed with two orthogonally arranged digital high speed cameras (MotionPro 500, Redlake, San Diego, CA, USA, spatial resolution: 1024 × 1024 pixels 2 ) at a sampling rate of 250 Hz. One camera viewed the upper part of the arena from the side through the gap of the LED array. The other camera was placed underneath the arena and covered the entire arena. For both views, the 2D positions of the fl y and -if present -of the target were determined frame by frame with custom-made software. The longitudinal body axis orientation of the fl y was determined from the bottom view only (for details of the procedures see Lindemann, 2006). Briefl y, for both camera views a background image was calculated as the mean of 10 frames from different parts of a movie. This background image was subtracted from every frame, yielding difference images. In principle, difference images highlight moving objects and suppress all stationary items in the image. Unfortunately, the output of the camera chip suffers from pixel noise, and the noise pixels introduce false object pixels which have to be eliminated from the difference images. The difference images were binarized employing a user-defi ned threshold above noise level. The program seeks for connected object regions in the binary image (segmentation) and prunes the resulting region set by testing certain criteria such as the size (number of pixels constituting the region) and form of the region (cf. roundness). The position of the fl y or dummy is assumed to correspond to the center of gravity of a respective region and its orientation to the orientation of its major axis. Because the position and orientation of the fl y (region) are evaluated from a large number of pixels in the binary image (typically well above 100), errors are small in general.
Knowing the relative positions of the two cameras, 2D image coordinates were transformed into an orthographic 3D coordinate system (Zeil, 1983;. To assess methodological errors, the position and orientation of a perched fl y was reconstructed. The apparent yaw velocity caused by orientation errors had a standard deviation of 45°/s. Due to the high sampling rate of the cameras (250 fps) this standard deviation corresponds to a change of the estimated fl y orientation between two consecutive frames of only 0.18°. The position error that is caused by distortions caused by the camera optics, increased with increasing eccentricity of the fl y in the fl ight arena, but was always below 2 mm. These data allowed the analysis of the fl ight trajectories with respect to the fl y's yaw velocity (calculated from the changes in yaw angle, as determined from the data of the bottom camera, within 4 ms time intervals), forward velocity (calculated from the changes in 3D position within 4 ms time intervals), turning frequency, the distance to the target and the deviation angle of the target direction from the fl y's body long axis orientation. This deviation angle approximates the retinal error angle of the target quite well, since the yaw angle of the head deviates only by few degrees from the orientation of the body length axis, and the roll angle does not change much during free-fl ight maneuvers despite considerable body roll during saccadic turns . The reconstruction of the 3D-trajectories and all further data analysis was done in Matlab 6.5 (Mathworks, Natick, MA, USA).
For the analysis of the time-structure of the fl ight trajectories only those fl ights were used that met certain criteria. (1) Optomotor and cruising fl ights lasted for at least 150 ms. (2) Chases were classifi ed according to whether the target was caught ('C-chases') or pursued ('P-chases') without catching it . To ensure that males did not coincidentally fl y for some time in the same direction as the target, but really chase it, fl ights were classifi ed as P-chases only if the male followed the target for at least one lap of the dummy (i.e. for at least 510 ms).
To evaluate the frequency of yaw turns executed during fl ights, it had to be defi ned which fl uctuations of the yaw velocity profi le are classifi ed as turns. Although the yaw velocity traces change continuously, turns were defi ned by applying threshold operations. First, the mean angular velocity was subtracted from the angular velocity trace. In general, turns were detected, if the angular velocity exceeded a certain threshold and fell below this threshold after some time. Rightwards turns corresponded to positive thresholds, leftwards turns to negative ones. During chases and optomotor following the absolute value of the angular velocity frequently did not immediately fall down again to the mean angular velocity, but started increasing again from a more elevated level to reach a second peak. This peak was counted as a separate turn if the angular velocity decreased below one of fi ve thresholds of increasing absolute size (150°/s, 300°/s, 500°/s, 800°/s, and 2000°/s). For instance, the angular velocity may increase monotonically from its mean value to a value above the threshold of 500°/s and then fell down below the threshold of 300°/s, before it increased again to reach a second peak above the threshold of 800°/s. In this case, two turns were counted. However, if the angular velocity did not decrease between the two peaks below the threshold of 300°/s, only one turn was counted. The lowest threshold was chosen to be above three times the standard deviation of the methodological velocity error (45°/s, see above).

RESULTS
Three distinct fl ight behaviors of blowfl ies were analyzed as well as their interactions, i.e. chasing behavior, optomotor following and, as a reference, cruising fl ight. A small black sphere moving in the upper part of the fl ight arena served as a target for chasing behavior. The fl ight sequences were recorded with a pair of highspeed cameras in a cylindrical fl ight arena surrounded by a grating pattern that was either stationary or moved horizontally at different velocities in either direction.

DIFFERENT FLIGHT BEHAVIORS
The term cruising fl ight is used here for fl ights which do not have an obvious goal and are not elicited by changes in the environment, for instance, by a moving target or the background. The dynamical features of cruising fl ights are distinguished by a series of rapid changes in the orientation of the body long axis (Figure 2Ai). The angular velocities generated during these rapid turns may reach frequently more than 2500°/s (Figure 2Aii). These rapid body turns exhibited during cruising fl ight are called -by analogy to rapid human eye movements -saccades (Collett and Land, 1975;Schilstra and van Hateren, 1999;van Hateren and Schilstra, 1999). Between saccades the orientation of the fl y's body long axis remains relatively stable.
The behavior of male flies dramatically changes when a black sphere mimicking a female fly is set into motion. Such a female dummy, even when moving at constant velocity (700°/s) on a circular track, is pursued from below and behind. During chasing flights, the male frequently flies slightly outside the track of the target (Figure 2Bi). We classified the chases by their catching success : In successful chases the target was caught after short time ('C-chases'; mean duration = 371 ms, STD = 172 ms, n = 50). In unsuccessful chases, the target was pursued for at least one lap of the target (510 ms) without catching it; the fly either approached but missed the target, or it gave up chasing and retired. While pursuing the target in front of a stationary background, the mean angular velocity of the fly is close to that of the target (Figure 2Bii). The fly continuously changes its yaw orientation and performs relatively smooth body turns (compare with body saccades during cruising flight; see above) of varying amplitude. As these yaw velocity fluctuations are not as transient and rapid as the saccades found during cruising flight, chasing behavior of blowflies has been concluded to be mediated by a smooth control system Egelhaaf, 2003, 2005;. Relatively large turning velocities are frequently observed at the beginning of a chase when the fly makes an initial turn towards the target and at the end of the chase when the fly tends to orient itself almost orthogonally to the target's direction of movement. Because we want to concentrate on the flight characteristics during on-going pursuit, these large turns were not included into the detailed quantitative analysis.
The saccadic fl ight style of blowfl ies appears to be abandoned not only when males chase moving targets, but also when the entire background is moving. When the vertical grating surrounding the fl ight arena rotated horizontally at a constant velocity either slowly (45°/s) or fast (365°/s), the fl ies tended to follow the visual wide-fi eld motion on roughly circular tracks by continuous body rotations. In this way fl ies tend to reduce the retinal slip velocity induced by the imposed wide-fi eld motion ( Figure  2Ci). This so-called optomotor following is generally assumed to compensate for asymmetries in the animal's sensory and motor systems during locomotion (reviews: Wehner, 1981;Collett et al., 1993). Because the fl y follows more or less closely the movements of the background, its mean yaw velocity is close to the angular velocity of the background and the fl uctuations around this mean are smaller than many saccades during cruising fl ights (Figure 2Cii).
The differences in fl ight style as shown in Figure 2 for examples of cruising behavior, chasing and optomotor following were quantifi ed across different fl ies by evaluating fl ight parameters, such as the average yaw velocity, the peak yaw velocity deviations during turns from the average yaw velocity, the forward velocity as well as the turning frequency (Figure 3).
Both the mean yaw velocity and the peak yaw velocity deviations from average yaw velocity refl ect the fl ies' fl ight mode ( Figure 3A). During cruising fl ights the average yaw velocity is close to 0°/s, although male blowfl ies perform saccadic turns with often high peak velocities (Figure 2Aii). This is because during the intersaccadic intervals yaw velocity is close to 0 (Wagner, 1986b,c;Schilstra and van Hateren, 1999;van Hateren and Schilstra, 1999) and saccadic turns are distributed approximately equally in both directions during cruising fl ight. Thus, positive and negative velocities may occur with an approximately equal share (Trischler, 2009).
During optomotor following fl ies tend to turn according to the direction of the moving background. The average yaw velocity is increased as compared with that during cruising fl ights, i.e. to 279°/s during slow and to 426°/s during fast background rotation ( Figure 3A). Given that the background angular velocity amounted to 45°/s and 365°/s, respectively, this indicates the somewhat paradoxical situation of a closed-loop optomotor gain, as given by the ratio of the angular velocity of the fl y and that of the background (Collett, 1980), of more than 1. Note, however, that the fl ies do not only rotate under our experimental conditions as is implicitly assumed when estimating the optomotor gain, but that they also translate. Combined rotation and translation results in complex optic fl ow patterns on the two eyes depending, apart from the current motion vector of the animal, on its position and orientation in the fl ight arena. In any case, during optomotor following the peaks of the angular velocity fl uctuations around the average yaw velocity are much smaller than during cruising (compare also Figure 2Aii with Figure 2Cii). Since during optomtotor following the fl ies fl ew in the upper part of the arena (on average 115 mm below its ceiling), mainly regions below the equator of the eyes were stimulated by wide-fi eld motion. This fi nding indicates, in accordance with a previous study on tethered fl ying fl ies that optomotor following can also be elicited in the equatorial and ventral parts of the visual fi eld (Borst and Bahde, 1987). In a similar way, stimulation of the ventral eye is needed in bibionid fl ies to evoke optomotor responses, whereas only the dorsal eyes promote male-specifi c visually guided behavior (Zeil, 1983).
During chasing after the dummy, the males turn continuously at high velocities. Only rarely the yaw velocity assumes 0°/s for a longer time. By contrast, during cruising fl ights the fl y does not turn much between saccades. The peak angular velocity deviations from average yaw velocity are again much smaller during chasing than is characteristic of cruising fl ights. Hence, when pursuing a target that moves on a circular path and when following a continuously moving optomotor stimulus, blowfl ies follow the respective visual stimulus by changing their body orientation much more smoothly than during saccadic cruising fl ights.
The average forward velocity during cruising in our fl ight arena amounts to 0.4 m/s. This value increases during optomotor following with increasing background velocity to 0.55 m/s and 0.8 m/s during slow and fast background motion, respectively ( Figure 3B). This fi nding indicates that optomotor following under free-fl ight condition is not only accomplished by turning responses but, in addition, by adjusting the translation velocity. While pursuing a target moving at 1 m/s the chasing fl y has an average forward velocity of 1.18 m/s ( Figure 3B). It is remarkable that during chasing, males can triplicate the forward velocity adopted during cruising fl ights in the same fl ight arena. This fi nding is in accordance with previous conclusions that the translational velocity during chasing is increased until a critical retinal target size is reached . Independent of the exact environmental conditions and the fl ight mode, the velocity fl uctuations around the mean forward velocity range between 0.11 m/s and 0.18 m/s. Hence, at least during optomotor following and during chasing the variations of forward velocity are small relative to the respective mean velocities.
Blowfl ies execute on average 16-18 turns/s during cruising, optomotor following and chasing (Figure 3C), although the peak velocity of turns might differ tremendously for the different behaviors. This fi nding indicates that the number of turns is quite independent of the fl ight behavior and of the environmental conditions, although the characteristics of the turns differ considerably.

CHASING BEHAVIOR DURING BACKGROUND MOTION
Chasing behavior and optomotor following that were investigated separately, so far, may interfere with each other under normal fl ight conditions. When the male fl y turns during a chase towards its moving target, the retinal image of the background inevitably moves in the opposite direction on the fl y's retina and, if it is textured, may activate optomotor following. Optomotor following may then counteract fi xation of the target mediated by the chasing control system. Consequently, the two control systems may be in confl ict with each other.
To elucidate the potential impact of the optomotor system on chasing behavior, we conducted a set of further behavioral experiments. Since it is hardly possible to eliminate any background textures and thus some sort of background stimulation during self-motion, we employed an alternative approach and presented simultaneously a dummy fl y moving at 700°/s on a circular track and wide-fi eld motion of the vertical grating surrounding the fl ight arena. This background grating moved either slowly (45°/s) or fast (365°/s) in the same direction as the target ('positive background motion') or in the opposite direction ('negative background motion'). As reference condition, the background was held stationary. We analyzed the potential impact of background motion (1) on catching success of the pursued target and (2) on various relevant fl ight parameters, such as turning and translational velocity as well as turning frequency but also the error angle under which the target is fi xated by the chasing fl y. During chasing, when the pursuing fl y is close to the dummy target the average altitude of chasing fl y is 50 mm below the dummy track and, thus, about 115 mm below the ceiling of the fl ight arena. Then the optomotor stimulus is seen -depending on the pitch angle of the head -mostly in eye regions below the equator. The stimulated eye region thus overlaps largely with that that was stimulated, on average, during optomotor following (see above). To assess a potential impact of background motion on the catching success of chasing fl ies we determined the average number of catches occurring within 250 independent 40 s time-windows while 20 male fl ies were in the fl ight arena. These 20 male fl ies, on average, caught the dummy between 14 and 17 times per timewindow when the background was stationary or moved slowly in either direction (Figure 4A). Even at high background velocities male fl ies were still able to catch their target frequently. However, the catching frequency decreased by about one third to just below 10 catches per time-window. One potential reason for the decrement in catching success is a reduction in the readiness of fl ies to fl y at all during fast background motion, rather than the performance of the chasing system. This issue was addressed by determining the average number of fl ies that fl ew simultaneously at each instant of time for the different background conditions (Trischler, 2009). Since fl ight sequences usually last for only few seconds, many different fl ies fl ew during the counting period even if the mean number of fl ying fl ies at each instant of time ranges, on average, only between 5 and 2. The number of fl ies fl ying simultaneously at any time in the arena was similar in the stationary and slowly moving arena, but is approximately halved when the background pattern moved fast ( Figure 4B). Hence, fast background motion obviously diminishes the readiness of fl ies to fl y. This result suggests that the catching success does most likely not deteriorate as a consequence of an interference of background motion with the chasing system after the fl y has initiated a chase. Nevertheless, these data do not exclude that background motion infl uences the fl ight behavior during the chase in a more subtle way.
To quantify a possible impact of wide-fi eld motion on the fi ne structure of the chasing fl ights, we evaluated the temporal structure of chasing trajectories. The analysis was done separately for C-and P-chases. According to predictions based on both the additive and the efference copy scheme of how pursuit and optomotor following might interact, background motion should affect the turning velocity of the chasing fl y, although these predictions take only the rotational degree of freedom of movement into account. In contrast to expectations, the average yaw velocity during both Cand P-chases ranges between 800°/s and 900°/s for all background conditions (Figures 5A,B). The results for the C-and P-chases are similar, although P-chases are less variable than C-chases. For both types of chases the angular velocity did not depend consistently on FIGURE 4 | (A) Box-Whisker plots of the number of catches during time windows of 40 s. The target was moving at 700°/s under various background conditions. For each background condition the catches were counted in 50 time-windows (each lasting 40 s). The background was either moving slowly (45°/s) or quickly (365°/s) in the same (positive sign) direction as or in the opposite (negative sign) direction to the target. As a reference, the background was held stationary (0°/s). The catching frequency is signifi cantly decreased at high background velocities with reference to the number of captures obtained while the background was stationary (Wilcoxon signed rank test; P < 0.01). Catches: n = 2954. (B) Number of fl ies fl ying simultaneously at any time in the arena under the three different background conditions. The background was either stationary (0°/s, reference), or moved at 45°/s or at 365°/s (positive and negative direction values pooled). The number of fl ies fl ying was determined during each 1 s time bin of a 30-s time-window and averaged over the 30-s time-window. The averages shown in the fi gure are obtained from 10 trials. Since fl ight sequences usually last for only few seconds, many different fl ies fl ew during the counting period even if the mean number of fl ying fl ies at each instant of time ranges, on average, only between 5 and 2. The number of fl ights was found to be signifi cantly smaller at the high background velocity with reference to the stationary background (Kruskal-Wallis test; P < 0.01). 1 s time-bins: n = 1800.

FIGURE 5 | Yaw velocity, forward velocity and turning frequency determined separately for C-and P-chases for fi ve different background conditions: the grating was held stationary as a reference (0°/s) or the grating either moved slowly (45°/s) or fast (365°/s) in the same direction as the target (positive sign) or in the opposite direction to the target (negative sign). (A,B) Time-averaged (±SD) yaw velocity (black) and mean (±SD) deviation peaks from average yaw velocity (red) determined for P-chases (A) and C-chases (B). The target moved at 700°/s (asterisk). (C,D)
Time-averaged (±SD) forward velocities (black) and the average (±SD) velocity fl uctuations (red) determined for P-chases (C) and C-chases (D). The fl ies followed the target that moved at 1 m/s (asterisk). (E,F) Overall average (±SD) turning frequency determined for P-chases (E) and C-chases (F). Mean and standard deviation were calculated across fl ies. 0°/s: C-chases n = 7, total fl ight time (TFT) = 1896 ms; P-chases n = 10, TFT = 2844 ms. +45°/s: C-chases n = 7, TFT = 2288 ms; P-chases n = 8, TFT = 2492 ms. −45°/s: C-chases n = 4, TFT = 748 ms; P-chases n = 8, TFT = 3912 ms. +365°/ s: C-chases n = 9, TFT = 2788 ms; P-chases n = 7, TFT = 3108 ms. −365°/s: C-chases n = 9, TFT = 2928 ms; P-chases n = 9, TFT = 4024 ms. background velocity. Moreover, background motion does not affect the peaks of the fl uctuations around the average yaw velocity during both C-and P-chases (Figures 5A,B). The frequency of turns ranges in C-and P-chases for all background conditions between 12 and 18 turns per second in P-chases and in C-chases without consistent dependence background motion (Figures 5C,D).
Since background motion in a non-chasing situation was found to affect not only the rotational but also the translational velocity of fl ies, we also scrutinized potential changes in translational velocity induced by background motion during chasing behavior. Again, wide-fi eld motion does not affect systematically the translation velocity during both C-and P-chases. Moreover, in both chasing modes the velocity fl uctuations are small around average velocity (on average around 0.2 m/s) and increase only slightly with increasing background velocity (Figures 5E,F).
To assess potential consequences of wide-fi eld motion on the fi nal velocity before catching the target during C-chases, the time course of the fl ies' decreasing distance to the target was determined for the last 200 ms of successful chases. Since the translation velocity of fl ies is much infl uenced by background in the non-chasing fl ight mode, one could imagine that a chasing fl y is retarded from catching the target by negative background motion and pushed towards the target by positive background motion, respectively. Although there is some variability in the time course of distance reduction with background velocity, no consistent dependence on background velocity is observed (Figure 6A). The standard errors of the time-dependent distance curves overlap widely, suggesting that even the fl ies' performance during the fi nal fl ight phase before catching the target is not affected by background motion.
Since the error angle of the target on the chasing male's retina was concluded to be one visual parameter controlling chasing behavior , we investigated whether it is affected by background motion. Again, the error angle is evaluated only for C-chases which eventually terminated with catching the target. The target is, on average, seen in the frontal visual fi eld and only slightly displaced -on average by 0° and +10° -into the direction of target motion (Figure 6B). Does background motion have an impact on the target's error angle during a blowfl y's chase? There is a slight increase in the fi xation error at slow background velocities. However, this tendency is no longer obvious at high background velocities (Figure 6B). The fl uctuations of the error angle are, on average, somewhat larger at high background velocities. These values range between 7° and 24°, which indicates that the target is still fi xated within the frontal part of the visual fi eld. Altogether, we do not fi nd indications of pronounced and systematic dependencies of chasing performance on background velocity.

DISCUSSION
In behavioral experiments we examined two fl ight control systems of blowfl ies: the chasing system of males and the optomotor system which is implemented in both sexes, as well as the potential interactions of the control systems. Three fi ndings are particularly striking. (i) The characteristic saccadic fl ight and gaze strategy as is a distinguishing feature of cruising fl ights is largely abandoned when the entire visual surroundings move around the fl y; then fl ies tend to follow the moving pattern in a relatively continuous and smooth way. (ii) When male fl ies engage in following a small target, they also employ a smooth pursuit strategy (see also . (iii) The performance and dynamical characteristics of chasing are not much affected when the background moves in either the same or in the opposite direction as the target. Hence, optomotor following is overridden by the chasing system and does not much deteriorate chasing performance. It is concluded that during chasing after a moving target the optomotor system is largely suppressed.
Following moving targets is highly relevant for male blowfl ies, as they have to catch fl ying females in order to reproduce. In contrast, following moving wide-fi eld patterns will occur in nature only rarely. Normally, wide-fi eld motion on the eyes is only induced by the animal's own movements. Thus optomotor following is thought, under normal fl ight conditions, to stabilize the fl ight against unintended disturbances that cause asymmetries in the optic fl ow across the two eyes. A saccadic fl ight and gaze strategy as is a characteristic of many insects' cruising fl ight behavior (e.g. Wagner, 1986a,b,c;Land and Collett, 1997;Schilstra and van Hateren, 1999;van Hateren and Schilstra, 1999;Kern and Egelhaaf, 2000;Tammero and Dickinson, 2002;Mronz and Lehmann, 2008) is abandoned during chasing behavior and optomotor following (see also Mronz and Lehmann, 2008). There is evidence in blowfl ies that the translatory optic fl ow generated during the straight gaze segments between saccadic turns is extracted by the visual system to obtain spatial information about the environment Lindemann et al., 2005Lindemann et al., , 2007Karmeier et al., 2006 ). This information is no longer easily available during chasing behavior and optomotor following. It will be discussed below why this may not lead to severe problems for fl ies.

PURSUIT OF A SMALL TARGET
Chasing behavior of male blowfl ies belongs to the fastest visually guided behaviors found in nature. Generally, the chasing system is viewed as a feedback control system that minimizes deviations smooth pursuit is interrupted by saccades to center the target again (Boeddeker and Egelhaaf, 2005;review: Land, 1999). Smooth pursuit of primates plays its most important role in everyday life when fi xating stationary objects during locomotion. Then the retinal image of the object can only be kept in the fovea, if it is pursued by smooth eye movements. In contrast to fl ies, that may fi xate a rapidly moving target during their aerobatic chases for several seconds, the phases of smooth pursuit in primates are relatively short and require the eyes to turn by only some tens of degrees. Model simulations revealed that the smooth chasing system of blowfl ies may, as a consequence of inevitable time constants in the fi xation controller, even generate the catch-up saccades of the head and the entire body if the target is displaced too rapidly on the pursuing fl y's retina (Wagner, 1986b;. The model employed is somehow similar to the proposal of a kind of integrator in the pursuit system that becomes operative when the target cannot be fi xated as a consequence of background motion and, after reaching a threshold, induces a fi xation saccade (Kirschfeld, 1997). The smooth and the saccadic components of pursuit eye movements in primates are usually thought to be controlled by distinct neural systems. However, recent fi ndings suggest that saccades and pursuit are two outcomes of a single sensorimotor process that aims at orienting gaze direction (De Brouwer et al., 2001Gardner and Lisberger, 2001;Orban de Xivry and Lefèvre, 2007).

OPTOMOTOR FOLLOWING
Whenever an animal moves in its environment, the retinal images move continually across the eyes. This so-called optic fl ow is exploited by visual systems in various ways. Components of the optic fl ow are assumed to form an input to the optomotor control system. When freely fl ying fl ies are confronted with a large-fi eld rotating environment, they compensate to some extent the rotation of the background (corresponding to an apparent unintended self-rotation) by turning responses in the direction of the visual motion stimulus (Figures 2C and 3A; e.g. Götz, 1968;Collett, 1980;Heisenberg and Wolf, 1984;Warzecha and Egelhaaf, 1996;Mronz and Lehmann, 2008). During optomotor following under freefl ight conditions where the animal experiences a complex mixture of rotational and translational optic fl ow, depending on its own movements but also the 3D structure of the environment, both the yaw and forward velocity of blowfl ies are increased with increasing background velocity (Figure 3A). A similar result was obtained in Drosophila (Mronz and Lehmann, 2008) and the hoverfl y Syritta where the yaw velocity increased roughly linearly with the slip speed of the pattern across the fl y's retina (Collett, 1980). In addition, the optomotor stimulus infl uenced the translational velocity of both Drosophila (Mronz and Lehmann, 2008) and Syritta (Collett, 1980). It should be noted that the rotation of the arena in these experiments on free-fl ying fl ies does not only induce rotational optic fl ow on the eyes, but, since the animals are usually not in the center of the fl ight arena, also strong translational optic fl ow, depending for the different eye regions on the relative location of the animal in the arena. It has been concluded from behavioral experiments that the translational velocity of insects appears to be controlled quite generally by the translational optic fl ow component on their eyes even in stationary environments where the of the images of small objects from the midline of the visual fi eld. The chasing blowfl y keeps the retinal position of the target in the frontal fi eld of view by predominantly smooth rotations about the vertical body axis. Thus, the average error angle of the target on the retina is small during chases (Figure 6; . The fl ies' forward and the yaw velocities are adjusted to the target's fl ight dynamics. Hence, under our experimental conditions, the average forward velocities of fl ies pursuing a quickly moving target (700°/s) are increased threefold compared to the forward velocities during cruising fl ight in the same fl ight arena ( Figure 3B). During chases, forward and yaw velocity are controlled by the angular size and the error angle of the target, respectively Egelhaaf, 2003, 2005). A set of sex-specifi c neurons in the third visual neuropil of the male blowfl y's brain has been concluded to represent the neural substrate for target specifi city during chasing (Hausen and Strausfeld, 1980;Gilbert and Strausfeld, 1991;Strausfeld, 1991;Hausen, 1993, 1994). On the basis of naturalistic stimuli it was found that at least one prominent neuron of this ensemble shows a distinct direction selectivity and complex nonlinear response characteristics to the joint occurrence of different visual parameters of the target including size, position and velocity and the changes of these parameters over time (Trischler et al., 2007).
A control system that guides visual pursuit of a small target by visually induced changes in gaze direction mediated by movements of the eyes, the head and/or the entire body, is not only a characteristic of male blowfl ies, but has been described in other insects, such as other fl y species (e.g. Land and Collett, 1974;Collett, 1980; praying mantids (Rossel, 1980), dragonfl ies (Olberg et al., 2000) and bees (Gries and Koeniger, 1996). Not surprisingly, neurons have also been found in hoverfl ies and dragonfl ies, which are extremely sensitive to small moving objects and virtually do not respond to background textures (Olberg, 1981;O'Carroll, 1993;Frye and Olberg, 1995;Nordström and O'Carroll, 2006;Nordström et al., 2006;Barnett et al., 2007). However, targets can be pursued also on the basis of other sensory cues, such as auditory information as is employed, for instance, during phonotactic orientation of some insect groups (e.g. Stout et al., 1987;Weber et al., 1987;Böhm et al., 1991).
Our conclusion that in blowfl ies chasing behavior is mediated by a smooth control system is in accordance with previous studies on chasing behavior of other fl y species. Smooth chasing interrupted by body saccades, if the retinal image of the target moves outside the frontal part of the visual fi eld, has been shown also for the small housefl y Fannia canicularis (Land and Collett, 1974), for the hoverfl y Syritta pipiens (Collett and Land, 1975) and for the dolichopodid fl y Poecilobothrus nobilitatus (Land, 1993a). In contrast, a saccadic chasing system was concluded to account for chasing in the male housefl y Musca domestica (Wagner, 1986b).
In primates, including humans, a smoothly moving target evokes a combination of smooth and saccadic eye movements (review: Orban de Xivry and Lefèvre, 2007) depending on the target speed and on the target displacement with respect to the fovea, similar as in chasing blowfl ies (Boeddeker and Egelhaaf, 2005). At low target speeds, the target is kept centerd in the fovea by slow eye movements that follow the target smoothly. When the target is too rapid at high target speeds and displaced outside the fovea, optic fl ow depends, apart from the animal's movement vector, on its distance to objects in the environment (Srinivasan et al., 1991;Kimmerle et al., 1996).
In primates (including humans), coherent wide-fi eld motion as may be induced on the eyes when the animal rotates around its vertical body axis constitutes the input to the so-called optokinetic system which evokes following movements of the eyes. This optokinetic refl ex serves a similar function as the optomotor response of insects: By counter-directed eye movements the optokinetic refl ex compensates image slip induced by large-fi eld motion, such as occurs during self-motion of the observer. Together with the vestibulo-ocular refl ex, which provides fast compensation for head rotation, the optokinetic refl ex helps to stabilize gaze when the head and the body move, such as during self-motion. The optokinetic refl ex is characterized, like pursuit of small targets, by smooth eye movements intermitted by relocating saccades (Ilg, 1997).

INTERACTION BETWEEN PURSUIT AND OPTOMOTOR CONTROL SYSTEM
During turns towards a target, such as during chasing, the actively generated large-fi eld image displacements may contain strong rotational components and, thus, activate the optomotor following system. This system may then try to stabilize the fl ight path by generating turns away from the pursued target. Thus, the two control systems may be in confl ict with each other, and this might impair the chasing performance. Moreover, since the translational velocity of fl ies was shown to be affected by background motion, the performance of a fl y chasing after a target might also be affected by the translatory component of background motion.
How do chasing male blowfl ies deal with the potentially confl icting situation? Obviously they do it very effi ciently. Although the readiness of blowfl ies to fl y at all decreases at high background velocities, neither the performance of catching the moving target deteriorates much when the background is moved artifi cially with various velocities in either direction, nor could we pinpoint any impact on the error angle under which the target is fi xated in the frontal visual fi eld and the time course of chasing fl ights. These fi ndings are remarkable, since background motion on its own does not only affect the overall rotational and translational velocity of the fl y, but also the dynamic structure of fl ights by evoking largely smooth optomotor following responses. Hence, wide-fi eld background motion appears to reduce the likelihood of saccades to occur.
The robustness of blowfl y chasing behavior against large-fi eld retinal image displacements is in contrast to other systems mediating fi xation and pursuit of small targets. Female fl ies fi xate stationary and slowly moving objects (Virsik and Reichardt, 1976;Reichardt et al., 1983;Kimmerle et al., 1997;Duistermars et al., 2007). However, in contrast to the male chasing system, the retinal error angle under which the target is fi xated increases in females when the background is continually displaced at a constant velocity (Virsik and Reichardt, 1976). Furthermore, in praying mantids smooth pursuit deteriorates in front of a textured background, and the animal often switches to a saccadic pursuit strategy (Rossel, 1980;Kral, 2003). Such a shift from a smooth chasing mode of operation to a saccadic mode as has been hypothesized for target fi xation and pursuit in fl ies (Kirschfeld, 1997). Our fi ndings, however, reveal that such a shift does not occur systematically even when the background is moving.
The interaction between wide-fi eld motion and target pursuit has been also addressed in primates. Numerous studies indicate that target pursuit is clearly infl uenced by wide-fi eld stimulation, although the details of the infl uences are diverse (Collewijn and Tamminga, 1984;Keller and Khan, 1986;Kimmig et al., 1992;Masson et al., 1995;Mohrmann and Thier, 1995;Spering and Gegenfurtner, 2007).
How can the extraordinary performance of male blowfl ies in chasing small moving targets even in front of a moving background be explained? A wide range of mechanisms have been proposed so far to account for the interaction of two, potentially confl icting, behavioral responses (Crapse and Sommer, 2008). Three such interaction schemes are widely discussed also with respect to insect behavior, the additive scheme, the efference copy scheme and the suppressive scheme (see Introduction, Figure 1). Can any of these schemes explain the interactions between the blowfl y chasing system and optomotor following, and also the control system mediating the saccadic fl ight style when the animal is cruising around?
In order to explain by the additive scheme (see Figure 1A) that chasing performance is largely independent of background motion the gain of the chasing system had to be much larger than that of the optomotor system. In the hoverfl y Syritta the gain of the chasing system differs from that of the optomotor system in a frequency dependent way: it remains constant within a broad range of tested frequencies, whereas the optomotor system responds strongly only at low frequencies (Collett, 1980). Female housefl ies, Musca, revealed similar dynamic differences in its object-fi xation system and the optomotor system (Egelhaaf, 1987;Egelhaaf et al., 1988). Here, the object-detection system shows its strongest responses to transient object movements, i.e. at high frequencies, whereas the optomotor system responds best to much lower frequencies due to low-pass fi ltering in the visual motion pathway (Hausen, 1982a,b;Egelhaaf, 1987;Egelhaaf et al., 1988;Hausen and Egelhaaf, 1989). Recently, dynamic properties of these two control systems resembling those of hoverfl ies and housefl ies were also found in the fruit fl y Drosophila (Sherman and Dickinson, 2003;Duistermars et al., 2007). With dynamic separation of the visual consequences of object fi xation and of optomotor following, both systems can combine additively and still operate relatively independently in the different dynamic ranges. Dynamic separation could be a simple strategy to almost eliminate the unwanted optomotor infl uence on active turns, since the generation of rapid turns (head or body saccades) during object fi xation may induce retinal motion of the environment beyond the dynamic sensitivity range of the optomotor system (Collett, 1980;Egelhaaf, 1987;Duistermars et al., 2007). Nonetheless, the additive scheme with a frequency separation of the chasing system and optomotor following cannot easily explain the chasing performance of male blowfl ies. This is because in our systems analysis, where the male fl ies chased a constantly moving target smoothly, the background, as a consequence, was also displaced smoothly in the opposite direction. If no other computational measures were taken by the fl y's brain, the optomotor system would respond under these conditions very well to the large-fi eld motion and, thus, may impede the chasing performance by compensatory yaw body turns.
Also the efference copy scheme (von Holst and Mittelstaedt, 1950) cannot account for the interaction between the chasing and the optomotor systems. The 'efference copy' which represents the Optomotor following of a moving environment does not occur in normal behavioral situations, as fl ies usually move in stationary environments. Optomotor responses are rather assumed to compensate for unintended rotations which may occur as a consequence of internal asymmetries of the animal (Strauss and Heisenberg, 1990;Collett et al., 1993;Hengstenberg, 1993;Kern and Egelhaaf, 2000). Slowly varying rotational disturbances are compensated most effi ciently (Egelhaaf, 1987(Egelhaaf, , 1989Warzecha and Egelhaaf, 1996) in accordance with the view that optomotor course stabilization is destined to mainly balance the consequences of internal asymmetries of the animal. Hence, under normal conditions optomotor responses are supportive to ensure by feedback control a straight gaze during the intersaccadic interval.
During chasing, the situation is different because the animal is confronted with a moving target that is pursued very effi ciently by a smooth control system keeping the target largely fi xated in the frontal region of the visual fi eld. In addition, during chasing saccades only appear to be generated as catch-up saccades when the target changes its direction too rapidly to be fi xated by smooth following movements (Boeddeker and Egelhaaf, 2005). Hence, in contrast to cruising fl ight, during chasing the translational optic fl ow may be superimposed by a pronounced rotational component which makes it hard to extract distance information. How can the pursuing fl y cope with this problem? This may indeed not be a problem at all from the perspective of the chasing fl y -as long as it just follows its moving target and as long as the target does not collide with an obstacle. In other words, if the target -in normal behavior the leading fl y -moves on an unobstructed track, the chasing fl y can follow the target without jeopardizing its safety as long as it stays on the target's track. Indeed, the fl ight trajectories of the target fl y and the pursuing fl y appear to be quite similar, although the similarities have not yet been quantifi ed (Wagner, 1986b;Boeddeker and Egelhaaf, 2005).

ACKNOWLEDGMENTS
We are grateful to Dr. Norbert Boeddeker and Dr. Jens P. Lindemann for critically reviewing the paper and to the Deutsche Forschungsgemeinschaft (DFG) for fi nancial support. expected image motion during a turn is subtracted from the actual signal. Only if the turn occurs correctly, no optomotor response is evoked, but if it is subject to additional input, the optomotor system will still detect this and respond accordingly. Hence, if the chasing fl y does not only rotate, but also translates and, in particular, if the outside environment moves globally as in part of our experiments, it is hardly possible that the nervous system is able to predict the visual consequences of self-motion and compensates for it via an efference copy. This is because the animal has no a priori knowledge of the 3D-layout of the environment that affects the retinal image motion during translatory self-motion not to speak of external global image displacements. Thus, if the efference copy scheme would apply to the chasing of male blowfl ies, the error angle between the target and the chasing fl y should be affected consistently by optomotor stimulation -in contrast to our experimental results.
From these considerations it appears to be most likely that the chasing commands control a gating system that reduces the gain of the optomotor system during chasing or even suppresses it completely, probably in a largely frequency independent way ( Figure  1C). In this regard the chasing system is similar to the phonotactic system of the cricket which has also been concluded, despite earlier evidence in favor of an additive scheme Weber et al., 1987;Böhm et al., 1991) to rely on a suppression of optomotor following during pursuit (Webb et al., 2004). During chasing behavior of blowfl ies not only optomotor following is suppressed, but also the saccadic fl ight and gaze style which is characteristic of cruising fl ight and is characterized by the alternation of rapid saccadic turns and straight fl ight segments with virtually no rotation between saccades.

FUNCTIONAL CONSEQUENCES: NO SPATIAL INFORMATION IS AVAILABLE DURING CHASING
If the interpretation were correct, that the saccadic fl ight strategy of blowfl ies is a means to facilitate the processing of spatial information during the translatory phase in the intersaccadic interval Karmeier et al., 2006;Lindemann et al., 2007;review: Egelhaaf, 2009) it would be hard for the animal to obtain spatial information during chasing maneuvers as well as during optomotor following.