Blowfly flight characteristics are shaped by environmental features and controlled by optic flow information

A distinguishing behavioral feature of many insects including flies is their saccadic flight style. It is characterized by periods where body orientation stays relatively constant and by brief saccadic turns where the animal reaches high rotation velocities (e.g. Boeddeker et al., 2005; Boeddeker et al., 2010; Braun et al., 2010; Braun et al., 2012; Collett and Land, 1975; Geurten et al., 2010; Mronz and Lehmann, 2008; Ribak and Swallow, 2007; Schilstra and van Hateren, 1999; Tammero and Dickinson, 2002a; Wagner, 1985; Wagner, 1986; Zeil, 1983; Zeil, 1986). Body saccades are accompanied by head saccades. Saccades of body and head differ in various aspects (Boeddeker et al., 2010; van Hateren and Schilstra, 1999), which can be essential for the nervous system to extract spatial information from the retinal image displacements (Kern et al., 2006). Saccadic turns are difficult to infer from the flight trajectories of large insects, such as blowflies or bees, which, probably as a consequence of inertia, change their overall direction smoothly (Boeddeker et al., 2005; Schilstra and van Hateren, 1999). In contrast, in the much smaller Drosophila, saccadic turns have been inferred on the basis of the flight trajectories, which look much more jerky than those of large insects (e.g. Tammero and Dickinson, 2002a; Mronz and Lehmann, 2008).

All changes in flight behavior directly affect the visual input of the animal, as movements of the animal lead to characteristic retinal image displacements. This optic flow (OF) yields cues about the animal's self-motion as well as the three-dimensional structure of the environment, and is a source of information for course control (for reviews, see Collett, 2002; Gibson, 1979; Koenderink, 1986; Lappe, 2000). Flight speed, for instance, depends on environmental clearance, as has been shown for bees (e.g. Baird et al., 2010; Portelli et al., 2011; Srinivasan et al., 1996). Yet it is not entirely clear how saccade amplitudes depend on the spatial layout of the environment during free flight. Saccade amplitude is highly variable (e.g. Boeddeker et al., 2010; Mronz and Lehmann, 2008; Schilstra and van Hateren, 1999; Tammero and Dickinson, 2002a; van Hateren and Schilstra, 1999), and depends to some degree on previously learnt information (see Hesselberg and Lehmann, 2009; Menzel, 2009) and the internal state of the animal (e.g. Budick and Dickinson, 2006; Cardé and Willis, 2008; McArthur and Dickman, 2011). For both freely flying and tethered Drosophila, saccades were concluded to be directed away from the eye experiencing image expansion, for example during obstacle approach (Bender and Dickinson, 2006; Budick et al., 2007; Reiser and Dickinson, 2010; Tammero and Dickinson, 2002a; Tammero and Dickinson, 2002b; Tammero et al., 2004). In contrast, collision avoidance of blowflies could be successfully accounted for by a behavioral model (‘CyberFly’) in which the initiation and direction of saccades is controlled through the output of a pair of wide-field neurons, each excited by front-to-back motion in one visual hemifield (Lindemann et al., 2008).

In this study we challenge, by targeted modifications of the width and layout of the flight arenas, three OF-based mechanisms to predict the direction of saccades and compare these predictions with the blowflies' behavior. Two mechanisms use expansion OF fields according to studies in Drosophila (Tammero and Dickinson, 2002a; Reiser and Dickinson, 2010) and one mechanism is based on intersaccadic OF strength.

Not only saccade frequency and amplitude vary with the layout of the flight arena. Rather, we find that blowflies adjust their translation velocity depending on environmental clearance in a manner similar to that of honeybees (e.g. Baird et al., 2010; Portelli et al., 2011; Srinivasan et al., 1996). From the intersaccadic OF fields during intervals, we suggest eye regions that are potentially involved in the control of translation velocity.

We recorded the behavior of 2- to 7-day-old female blowflies of the genus Lucilia taken from our laboratory colony.

We used two different types of flight tunnels, one with obstacles (‘obstacle tunnel’) and three straight tunnels of different width. All tunnels were covered with a Perspex lid and were illuminated from below by eighteen 50 W halogen lamps (luminance in the middle of the arena was 250 cd m^–2). Temperature was kept between 27 and 33°C.

The walls and floor of this tunnel (230×40×20.5 cm, length × width × height; Fig. 1A) were covered with a red-and-white random square pattern (pixel size: 4×4 mm). This pattern exhibits good contrast to the fly and facilitates the detection of the animal on the images. Two Perspex obstacles (0.6×26×20.5 cm) were installed in the tunnel. The first obstacle was attached to the left wall perpendicularly to the tunnel's long axis 108 cm from the entrance, the second one was attached to the right wall 128 cm from the entrance. The first obstacle was covered with the same red-and-white random pattern lining the tunnel. For recording purposes, the pattern was printed on transparent plastic film sheets. The second obstacle carried a red random square pattern printed on standard white paper (pixel size: 17×17 mm).

The basic tunnel (230×36×20.5 cm) had adjustable sidewalls starting 0.8 m from the entrance (Fig. 1B). These allowed us to narrow the tunnel from 36 to 18 or 9 cm. Movable flaps mounted at the beginning and the end of the sidewalls served as a guide railing. The walls and floor of the tunnel were covered with a red random square pattern (pixel size: 4×4 mm).

The flies' trajectories in the obstacle tunnel were filmed by two orthogonally arranged high-speed cameras, one above the tunnel and the other at its entrance (500 frames s^–1 sampling rate, 1024×1024 pixels; MotionPro, Redlake, San Diego, CA, USA). The filmed area was 40×40 cm at the tunnel top and encompassed the two obstacles. In the straight tunnels, both cameras were used to film the flies from above and, thus, to enlarge the recording area (70×40 cm). In preliminary experiments we used a third camera, running at 50 frames s^–1, that was located at the entrance of the arena to judge changes in flight altitude, and found that they were small compared with the changes in the flies' lateral position.

Flies were tested individually. If they did not fly within 2 min, they were removed from the arena and tested again later. In the narrow straight tunnel (9 cm), some flies were reluctant to fly or did not fly at all, limiting the total number of tested flies and flights per fly. With the exception of the narrow tunnel, flies stopped flying within the recording area only rarely. Only trajectories in the direction away from the entrance hole were stored for later analyses. We analyzed only complete passages without wall, bottom or roof contact of the fly. Such contact occurred only rarely, apart from the narrow tunnel condition.

The recorded video sequences were automatically analyzed with software developed in our workgroup (Lindemann et al., 2003). For each image, the coordinates of the fly's center of mass were detected and transformed into millimeter coordinates, using a Perspex calibration cube with markers on it. The top camera(s) (straight and obstacle tunnels) yielded the fly's x- and y-coordinates, the back camera (obstacle tunnel only) the x- and z-coordinates. By combining the top and back views, the time course of the fly's altitude in the obstacle tunnel was reconstructed (see Boeddeker et al., 2003). These data were used to confirm that the flight altitude remained relatively constant.

The limited spatial resolution of our video footage did not allow us to determine head yaw angles, but did allow the determination of body orientation (convention: 0 deg=straight ahead along the longitudinal body axis; leftward orientations are assigned a positive angle, rightward ones a negative angle). All further data analysis was carried out using MATLAB (version 7.x, The MathWorks, Natick, MA, USA). x- and y-coordinates as well as body orientation data were convolved with a Gaussian window (width 30 ms, sigma 6 ms) to reduce digitization jitter before calculation of velocity. If not stated otherwise, a turn was considered a saccade if the rotation velocity about the vertical body axis exceeded 400 deg s^–1. Saccade duration was determined as its temporal width at half-maximum peak velocity. Saccade amplitude was approximated by the change in yaw angle between two crossings of a 200 deg s^–1 velocity threshold. The flight distance was calculated as the sum of the flight sections traversed between two successive points of the flight trajectory projected into the x–y plane.

The OF was calculated by an algorithm that determines – for every viewing direction (‘sampling points’) – the distance between the fly and the respective part of the flight arena (bottom, wall, obstacle, etc.) seen in this viewing direction (see Foley et al., 1996) and takes into account the next position change of the fly along the trajectory (Koenderink and van Doorn, 1987). The sample points were evenly spaced; their angular distance was either 2 or 5 deg in azimuth and elevation as is specified below. The temporal resolution was 2 ms. OF calculations are independent of contrast, illumination and texture of the walls.

We calculated the intersaccadic OF for different parts of the visual field (see Results) for intersaccadic intervals (ISIs) of at least 30 ms. In the context of the control of translational velocity, OF analysis was based on the 30 ms interval prior to a saccade, i.e. before the yaw velocity exceeded 200 deg s^–1 the next time. OF analysis in the context of saccade generation was limited to a 10 ms time window, ranging from 30 to 20 ms before the next saccade. All OF calculations are based on pure translatory intersaccadic movement, because intersaccadic head yaw shows virtually no rotational changes of yaw orientation during the ISI (Kern et al., 2006; van Hateren and Schilstra, 1999). Residual rotations were eliminated within the 10 or 30 ms interval by setting the yaw angle during the movement from point A (t=0 ms) to point B (t=2 ms) to the mean value of the two yaw angles at A and B. Note that body yaw is not much different from head yaw orientation for most of the intersaccadic time (see van Hateren and Schilstra, 1999). The pitch angle – not resolved in the present study – was fixed at 20 deg (head tilted upwards). The roll angle was set to 0 deg. These values correspond to the mean head angles during semi-free flights (Kern et al., 2006; van Hateren and Schilstra, 1999).

In addition to the flights filmed in this study, for OF analysis we used 10 head trajectories (each 3.45 s in length) recorded in a cubic arena (side length 40 cm) by van Hateren and Schilstra (van Hateren and Schilstra, 1999). This was done to test proposed mechanisms for the control of saccades and translational velocity under a wider range of conditions. To facilitate comparison of the performance of the mechanisms across environments, we eliminated rotation velocities of the head by setting the yaw, the pitch and the roll angle to their mean values individually for each 2 ms interval of the ISIs. Therefore, only translational movements occur in all ISIs used in OF analysis. ISIs were defined as in our previous study (Kern et al., 2005).

From the OF in a 10 ms intersaccadic window (see above) we predicted the direction of the next saccade using three alternative methods. (1) Azimuth location of the focus of expansion (FOE): the FOE is defined as the location in the flow field from which all neighboring vectors point away. (2) Difference of expansion-congruency of OF in the left and right eye: while approaching an object, its retinal image expands. Expansion manifests itself in the OF as a radial pattern of velocity vectors around the focus of outflow. We calculated a measure of congruency of intersaccadic OF fields from such radial patterns separately for both eyes. The direction of the next saccade is predicted from the difference of the expansion-congruency measure obtained for the left and right eye. (3) Difference of OF strength: this measure is determined from the difference in OF strength in two selected areas, located symmetrically in the left and right visual field (for details see Results and the Fig. 12 legend).

To assess the visual field areas that might be relevant for controlling translation velocity, we compared intersaccadic OF strength in various parts of the visual field and determined the mean length of OF vectors within specific regions of the visual field 30 ms before the next saccade (details see legend of Fig. 5).

The flight trajectories of blowflies were seldom straight; they meandered (Fig. 2Bi) both in the flight tunnel with obstacles and in the straight tunnels, i.e. without any obvious need to turn (Fig. 2Bii,iii). Turns were accompanied by rapid changes in the longitudinal body axis of up to 60 deg in less than 40 ms with velocities of up to 3000 deg s^–1 (Fig. 2Ai,iii). These saccades were separated by periods of much smaller turning (Fig. 2Aii). In the straight flight tunnels, saccade direction switched in a rather regular fashion. The saccadic turns were not immediately reflected in the flight trajectories (see also Schilstra and van Hateren, 1999), which showed only much smoother changes in direction that were not necessarily concurrent with the saccades (Fig. 2Ai). In our flight arenas, translational velocities ranged between 0.5 and 2 m s^–1 and varied during the flights at a slower time scale than the rotational velocities (compare Fig. 2Aiv,Biv with 2Aiii,Biii).

Most flies used a relatively narrow S-shaped flight corridor when negotiating the obstacle arena (Fig. 3A) and most of the trajectories ran close to the tunnel midline in the straight tunnels (Fig. 3B–D). Flies flew most frequently in central regions of the flight tunnels. This behavior is similar to the centering response in bees (Dyhr and Higgins, 2010; Kirchner and Srinivasan, 1989; Serres et al., 2008; Srinivasan et al., 1991), which is thought to be controlled by the OF in the lateral visual field (Srinivasan et al., 1991).

In the straight tunnels, the trajectories covered slightly less than 50% of the tunnel width (Fig. 3B–D). The frequency distributions of presence of the flies within axial strips of each flight tunnel normalized to the tunnel width are rather similar for all tunnels, i.e. the minimum distance to the walls seems to be controlled not in absolute but in relative terms. Single trajectories usually covered a smaller area of the tunnel than the entire population of trajectories (Fig. 3E, insets 3B–E). This finding implicates that single flies do not always center perfectly in the tunnels. They thus tolerate to some extent imbalances of OF on the eyes. Flight trajectories of honeybees are also shifted towards either sidewall of the tunnel if their goal at the end of the tunnel is off-center (Serres et al., 2008), and flies with one eye covered tend to fly closer to the wall on the side of the open eye (R.K. and M.E., unpublished data).

Does the fly's translational velocity depend on the structure of the environment? To assess to what extent the fly's translational (two-dimensional) velocity depends on the environment, we modified the surroundings by changing the width and spatial layout of the flight arenas. In straight tunnels, translational velocity increased with increasing width (Fig. 4A). Mean translational velocities for nine out of 11 individual flies were larger in the wide than in the medium tunnel (Wilcoxon signed rank test, P=0.0186; Fig. 4B). Likewise, for the five flies that could be tested in all three straight tunnels, the mean translational velocities in the narrow tunnel were lowest (data not shown). Individual flies seem to have a preferred speed range in a given flight tunnel that could differ by more than a factor of two between individual flies (Fig. 4B). The mean preferred translational velocities varied to a larger extent between individual flies than within a given fly (Fig. 4B).

In the obstacle tunnel (maximum width 40 cm), translational velocity changed not only from trajectory to trajectory but also at different locations along a trajectory (Fig. 4C). On average, it fell between the mean velocities in the narrow (width 9 cm) and the medium tunnels (width 18 cm; Fig. 4A). At first sight, this result appears to be in conflict with our finding that in the straight tunnels translational velocity increases with tunnel width. However, the constrictions in the flight corridor resulting from the obstacles (Fig. 4A) might lead to a decrease in flight velocity. In addition, fast turns, which were more frequent in the obstacle than in the straight tunnels, led to reductions in forward velocity (see below). Flight speed thus is expected to be smaller in the obstacle tunnel than in a straight tunnel of equal width.

Flight speeds in the straight tunnels increased with tunnel width. Thus, flight speed might be regulated by a mechanism that keeps the intersaccadic OF constant, as the translational OF depends on the distance to environmental structures. This has been concluded for bees (see Baird et al., 2010; Srinivasan et al., 1996). However, it is not clear a priori which eye regions are used for speed control. Therefore, we identified those eye regions in which the OF varied least (i.e. smallest standard deviation) during the last 30 ms of the ISI across a large number of different conditions (Fig. 5A–E). Fifteen regions of partly differing angular horizontal extent were located symmetrically with respect to the frontal midline. These always had a vertical extent of 40 deg and were analyzed for five different elevations (Fig. 5A–E). The uppermost region was centered about the horizon (Fig. 5A); the lowest was centered 40 deg below the horizon (Fig. 5E). In addition to our flight tunnel data, we also included data from a previous study (van Hateren and Schilstra, 1999) because here the flight trajectories were more variable with respect to flight direction and height.

OF strengths, i.e. angular retinal velocities, varied least (in Fig. 5, blue shading) across the five different flight arenas for intermediate retinal elevations (Fig. 5B,C). Relatively large differences (yellow shading) between OF strength in the different arenas occurred independent of azimuthal size and location for regions that covered more ventral areas of the visual field (Fig. 5E). The biggest variation (red shading) across OF strength was found for eye regions close to the horizon that include the lateral eye (Fig. 5A). For example, the OF strength amounts to 532±166 deg s^–1 in the narrow tunnel, 417±127 deg s^–1 in the medium tunnel and 246±99 deg s^–1 in the wide tunnel if the equatorial lateral visual field between azimuthal values of 60 and 90 deg is considered. This large difference suggests that lateral eye regions are not used for velocity control. However, the average intersaccadic OF strength within a bilateral pair of 30 deg wide frontolateral regions at an elevation of 20 deg below the horizon (one of three regions colored dark blue in Fig. 5C) was close to 100 deg s^–1 across flight arenas (Fig. 5F). The retinal velocities were only moderately larger (up to 140 deg s^–1) for the other two frontolateral regions shown in dark blue in Fig. 5C.

Hence, a translational velocity controller that adjusts the OF to a pre-set value is likely to use OF from frontolateral and moderately ventral eye regions. If flight altitude was 10 cm above ground, as judged from our behavioral data and assumed for our OF calculations, features on the ground considered for flight speed control would have a minimal distance of approximately 12 cm.

Turning direction and velocities were not distributed homogeneously along the flight paths (shown in Fig. 6A for the obstacle tunnel). Rather, left turns prevailed before and during passage of the first obstacle, whereas right turns dominated the area before clearing the second obstacle.

The amplitudes of saccades, i.e. turns with peak velocities exceeding 400 deg s^–1, were variable; most of them were smaller than 50 deg in all flight arenas (Fig. 6B,C). Only in the obstacle tunnel were saccades with larger amplitudes frequent (compare Fig. 6B and 6C); here the median saccade amplitude was largest (28 deg) as compared with 16, 17 and 23 deg in the narrow, medium and wide straight tunnels, respectively. Saccade amplitudes tended to be bigger when the fly approached the arena wall closely and, thus, with lower time-to-contact (Fig. 6D).

Saccade amplitudes were linearly related to saccade peak velocities (Fig. 6E), whereas saccade duration was rather constant for saccades larger than 20 deg (Fig. 6F). The under-representation of small saccades may be the consequence of the saccade detection threshold employed. These characteristics do not depend much on the environment, as they resemble those published by Schilstra and van Hateren (Schilstra and van Hateren, 1999) for body saccades in free-flying Calliphora in a geometrically different environment. Also, the saccades of freely flying Drosophila (e.g. Mronz and Lehmann, 2008; Tammero and Dickinson, 2002a) and those elicited by visual expansion in magnetically tethered Drosophila [fig. 3 in Bender and Dickinson (Bender and Dickinson, 2006)] share characteristics with those measured in freely flying calliphorids [present study and Schilstra and van Hateren (Schilstra and van Hateren, 1999)]. Nevertheless, there are species-specific as well as methodologically grounded differences [see Bender and Dickinson (Bender and Dickinson, 2006) for a thorough discussion of the differences between saccades of tethered and freely flying Drosophila].

The geometry of the environment has an influence on saccade frequency that, on average, is approximately twice as large in the obstacle tunnel as in the straight tunnels (Fig. 7). Saccade rate increased via shortening the ISIs, whereas saccade duration was kept rather constant (not shown). If the OF experienced by the fly modulates saccade frequency, we expect – in a given environment – an influence of flight velocity on saccade number. However, saccade frequency did not depend consistently on flight speed in our flight arenas (Fig. 7). We obtain qualitatively the same results with saccade thresholds lower than the 400 deg s^–1 used here.

Because in Drosophila the OF corresponding to an expanding object has been shown to trigger collision avoidance maneuvers, the likelihood and the size of saccades increases when the fly is heading towards a nearby obstacle (Tammero and Dickinson, 2002b). We therefore determined for all four flight arenas and the cubic arena (van Hateren and Schilstra, 1999) the ISI duration as a function of the time-to-contact to the arena wall 30 ms prior the next saccade (Fig. 8). The duration of an ISI is an indicator of how early the fly generates a saccade after a section of straight flight. The ISIs tended to become smaller as the time-to-contact to the arena wall decreased. This finding helps to understand the high mean saccade frequency in the obstacle tunnel, because here the time-to-contact, on average, is smaller than in the straight tunnels.

Saccades occurred everywhere along the flight corridor of the obstacle tunnel (Fig. 3A) without obvious differences for the different amplitude classes (Fig. 9A–D). Although a minimum of two saccades would suffice to fly around the obstacles, i.e. one to clear each obstacle, this ‘minimal strategy’ was not employed by blowflies in the present study. Most flights contained between four and eleven saccades (Fig. 9E). Even after traversing the obstacle tunnel several times, flies did not consistently use less saccades in later flights (flight numbers 6–9) as compared with their early (1–4) flights (Wilcoxon signed rank test, P=0.4453, N=10).

Are saccade amplitudes related to the total number of saccades used to clear the obstacles? In nearly 50% of flights we detected two saccades right to the edge of the first obstacle (rectangle in Fig. 6A). Flights with one or three saccades each occurred approximately 20% of the time; flights with four and five saccades were rare (Fig. 10A). The distributions of saccade amplitudes for single saccades, doublets or triplets overlapped substantially, with medians amounting to 31, 37 and 36 deg, respectively (Fig. 10B).

Saccadic turns seem to go along with changes in translational velocity (Fig. 2), and as the widest range of saccadic amplitudes was observed during flights in the obstacle arena, these data are used to quantify these changes. Similar results were obtained for the straight tunnels (data not shown).

For saccades exceeding 40 deg, translational velocity, in particular its forward component, was reduced considerably with a peak at approximately 50 ms after onset of the saccade (Fig. 11A,B). Thus, blowflies do not decelerate before the turn (as a car driver would), but only during and after it. In parallel, the sideways velocity component may increase considerably, implying that after saccades flies continue to move into their pre-saccadic flight direction for considerable portions of the ISI. Such sideways velocities were seen not only after large saccades, but also after small saccades. They increased with saccade amplitude (Fig. 11C). The appearance of sideways movement after saccades of low amplitude hints at a function of saccades in the context of distance information retrieval from OF in the straight tunnels as well as in the obstacle tunnel (see Discussion).

How might saccades be controlled by OF? In previous studies on flies, OF has been proposed to play a decisive role in determining the timing and direction of saccades (e.g. Dickinson, 2005; Lindemann et al., 2008; Mronz and Lehmann, 2008). Therefore, we determined three measures based on the intersaccadic OF that might be relevant in controlling saccade direction. Fig. 12 exemplifies flow fields for two time steps along a sample trajectory in the obstacle tunnel; Fig. 13 then shows the averaged data obtained for all environments.

This measure is related to the procedure employed by Tammero and Dickinson (Tammero and Dickinson, 2002a). Assuming centers of expansion at 46 deg to the left and right of the frontal midline, the horizontal OF was summed separately in four areas (Fig. 12A, top). The vertical extent of each area was 40 deg, centered about the horizon. The vertical OF was summed separately in four areas located as shown in Fig. 12A (bottom). Then, horizontal expansions were calculated as the difference of the summed signed (see inset top right) horizontal flow components (H, see inset) between areas 1 and 2 (left) as well as areas 3 and 4 (right). Vertical expansions were calculated as the difference of the summed signed vertical flow components (V, see inset) between areas 5 and 7 (left) as well as areas 6 and 8 (right). Horizontal and vertical expansions on either side were summed to yield the left and right total expansion. Saccades were predicted to be directed to the side corresponding to the smaller of these two numbers. In addition to expansion centers assumed at 46 deg to the left and right, we employed the same procedure for expansion centers at ±30 deg. Areas 1–4 were of equal size, i.e. we omitted the most lateral parts of the visual field. According to the rule sketched for the example flow field in Fig. 12A, the next saccade is predicted correctly to be leftwards, independent of the location of the assumed expansion centers. Saccade number two (Fig. 12B), however, is predicted to be leftwards for expansion centers at ±30 deg, but to be rightwards for expansion centers at ±46 deg. Taking all flight data together (Fig. 13B), the predictions are correct and are well above chance level for all environments, and even exceed 80% in two cases for the centers of image expansion at ±30 deg. In contrast, if the centers of expansion are located at ±46 deg, the percentage of correct predictions is close to chance level for three environments. Only for the cubic arena is the level of correct predictions higher. We conclude that the saccade direction can be well predicted based on the OF expansion congruency across environments, if the center of expansion is sufficiently frontal in the visual field.

OF strength was approximated by the lengths of the flow vectors, summed separately within six equally sized areas, covering in azimuth the complete visual field from 90 deg left to 90 deg right (Fig. 12B). The vertical extent of each area was 40 deg, centered about the horizon. Saccades were directed towards the side of the area experiencing the lower OF strength. From the example flow field shown in Fig. 12A, the next saccade is correctly predicted according to this rule to be directed leftwards, independent of the pair of areas compared with each other. Saccade number two (Fig. 12B), however, is predicted to be leftwards only if the frontolateral areas of the visual field are exploited. From the difference between total OF in either the lateral or frontal areas, the prediction is rightwards. Taking all flight data together, predictions based on the total intersaccadic OF in the lateral eye regions are poor for all but the cubic arena (green areas/markers in Fig. 13C). The OF in only the frontal eye regions (red areas/markers in Fig. 13C) results in a better performance, especially in the straight tunnels, where it even exceeds a level of 80% for the two narrow tunnels; however, only chance performance was reached for the obstacle tunnel. The best overall performance was achieved with correct predictions well above chance level for all tested environments when we took the intersaccadic OF in frontolateral areas into account (blue areas/markers in Fig. 13C). Note that we obtained qualitatively the same results if we took only the signed horizontal components of the OF vectors rather than the entire vector length into account.

When characterizing the turning responses elicited by expansion flow fields, especially in experiments on tethered flies, it is frequently assumed that the FOE coincides with the location of the most pronounced image expansion (e.g. Reiser and Dickinson, 2010). We therefore also attempted to predict saccade direction from the azimuthal FOE location. A saccade away from the side of the FOE in the preceding ISI was regarded as predicted correctly. Predictions are correct significantly above chance level only for the obstacle tunnel (Fig. 13A). For the cubic arena (van Hateren and Schilstra, 1999), prediction performance is close to chance level, whereas for the straight tunnels predictions based on the FOE bearing are significantly worse than chance level. Hence, the prediction of saccade direction based on FOE location does not account for the experimental data.

These results suggest that exploiting the expansion congruency of intersaccadic OF with a center of expansion at 30 deg or the overall intersaccadic OF across corresponding frontolateral regions of both eyes is a much more robust, though far from perfect, predictor of the direction of saccades than is the location of the FOE. On the whole, the expansion congruency of intersaccadic OF with a center of expansion at 30 deg is the best predictor of saccade direction under the variety of environmental conditions tested here. In any case, if the prediction mechanism also takes into account the lateral areas of the visual field, the performance decreases for most tested environments.

Blowflies modify the frequency, direction and amplitude of their saccadic turns as well as their translational flight velocity with respect to the shape of the environment. We provide evidence for the hypothesis that the control of both saccades and translational velocity rely to a large extent on the intersaccadic OF generated during self-motion of the animal. Our conclusions are based on data obtained in five environments differing in size and spatial layout. Rather than the lateral visual field, where the OF, at least during forward flight, tends to be strongest, OF in eye regions looking well in front of the fly is suggested to be used by both the saccade and the velocity controller.

In the following, we will discuss how selected flight parameters are related to the visual environment and how they might be controlled by the experienced OF.

Car drivers normally decelerate when obstacles restrict their path and blowflies also decelerate when obstacles come closer. This phenomenon has also been described for bees flying in flight tunnels of changing width (Baird et al., 2010; Portelli et al., 2011; Srinivasan et al., 1996). Interestingly, no differences in average flight velocity were found in Drosophila cruising in cylindrical flight arenas of considerably different diameters [1.0 m (Tammero and Dickinson, 2002a); 0.14 m (Mronz and Lehmann, 2008)], whereas David (David, 1982) found an increase of flight speed with increasing width of horizontal wind tunnels. This apparent discrepancy might be the consequence of largely different spatial constraints in tunnel-like versus cylindrical flight arenas.

The most plausible sensory cue controlling flight velocity is the OF generated during flight [tethered flight (e.g. Götz, 1968; Theobald et al., 2010); free flight (Baird et al., 2005; Baird et al., 2006; Baird et al., 2010; David, 1979; David, 1982; Dyhr and Higgins, 2010; Farina et al., 1995; Fry et al., 2009; Frye and Dickinson, 2007; Kern and Varjú, 1998; Preiss, 1993; Serres et al., 2008; Straw et al., 2010)]. It appears that by adjusting their flight speed, insects keep the OF on their eyes at a ‘preset’ total strength (Srinivasan et al., 1996). Accordingly, insects decelerate when the translational OF increases, for instance, while passing a narrow gap or flying in a narrow tunnel, as has been observed in bees (Baird et al., 2010; Portelli et al., 2011; Srinivasan et al., 1991; Srinivasan et al., 1996), moths (Verspui and Gray, 2009) and blowflies (present study).

Our results indicate that not all parts of the visual field of the flies contribute to the input of the velocity controller. This suggestion was based on the assumption that those eye regions across which the overall OF is kept largely constant in different environments may be decisive in controlling the animal's translation velocity. We were able to determine the OF across different eye regions on the basis of behavioral data obtained in five flight environments, which differed largely in size and spatial layout. Although this selection of environments inevitably comprises only a small range of potential environments in which blowflies show flight behavior, the spatial constraints set by the tested environment go along with a wide range of systematic environment-dependent variations of flight parameters, such as flight velocity.

Whereas the OF in the most lateral and ventral parts of the visual field depends considerably on the environmental context, in the moderately ventral (from the horizon to 40 deg beneath) and frontolateral visual field the OF is kept largely constant irrespective of the flight arena. Therefore, we suggest that the latter eye region is most relevant for velocity control of blowflies. Frontolaterally located eye regions have also been demonstrated to play a major role in velocity control in bumblebees (Baird et al., 2010). Recently, Portelli et al. (Portelli et al., 2011) concluded that honeybees also adapt their flight velocity to keep the retinal velocity relatively constant in the frontolateral visual field, independent of the spatial layout of the environment. In line with our results, viewing directions of 45 deg in the bee's visual field yielded a less variable OF than viewing directions of 90 deg in flight tunnels varying in width along their horizontal and/or vertical extent. These findings do not contradict studies on honeybees (e.g. Baird et al., 2006) or on Drosophila (e.g. David, 1979; Fry et al., 2009), where flight speed could be influenced by the OF caused by floor and/or side wall structures. The patterns employed in these studies also stimulated those eye regions that we concluded were involved in speed control for blowflies, but also for bees (Baird et al., 2010; Portelli et al., 2011).

The OF has been used by us as a first approximation of the sensory input of the velocity controller, although the computations in the visual motion pathway do not lead to a true representation of velocity, but also depend on texture and contrast of the environment. Moreover, spatial integration of local motion measurements across the visual field is nonlinear (for reviews, see Egelhaaf, 2006; Egelhaaf and Borst, 1993). Hence, future analysis should test whether our conclusions concerning the eye regions involved in velocity control are still valid if the specific features of biological motion computation are taken into account. Moreover, similar to those free-flight studies mentioned above on bees and Drosophila, all our conclusions on velocity control of blowflies are, inevitably, only correlative because they are based on free-flight behavior in a range of environments. Further analyses, potentially on tethered flying animals, where the stimuli to which the animal is exposed can be varied more systematically than is possible in free flight, are relevant to further challenge these conclusions on velocity control.

Our experiments revealed that in blowflies, in addition to their direction, the amplitude and frequency of saccades are the most relevant parameters to adjust turning behavior in response to environmental conditions; in contrast, saccade duration stays rather constant. How saccades of Drosophila depend on the environment can be inferred from two studies based on cylindrical flight arenas with largely different diameters: 1.0 m (Tammero and Dickinson, 2002a) and 0.14 m (Mronz and Lehmann, 2008). Mean saccade amplitudes and peak velocities were found to be considerably larger in the small flight arena than in the large one, reflecting the necessity to execute very sharp turns under spatial constraints to avoid collisions with the arena walls. Our finding that saccade amplitude – and with it saccade peak velocity – increases with decreasing time-to-contact is in agreement with this observation. In contrast to our results on blowflies, the duration of Drosophila free-flight saccades was concluded to depend on the flight arena and to increase notably with saccade amplitude (Mronz and Lehmann, 2008). The reasons for this difference are not clear, but may be species specific.

There is consensus that OF plays a decisive role in controlling the direction and amplitude of saccades. However, it is still inconclusive which OF parameters may be decisive. The times at which saccades are elicited in Drosophila and their direction were concluded to depend on the characteristics of retinal image expansion (Tammero and Dickinson, 2002a). In tethered flying Drosophila, saccades could be elicited in most parts of the visual field by expansion stimuli, with reduced response probabilities in frontal and rear regions and a rather broad probability peak in the lateral visual field (Bender and Dickinson, 2006; Tammero and Dickinson, 2002b; Tammero et al., 2004). On this basis it was concluded that image expansion in the front triggers landing and lateral expansion primarily collision avoidance responses. Hence, collision avoidance is assumed to be most effective during flight with a strong sideways component (Tammero et al., 2004). This conclusion has recently been challenged by correlating the occurrence of saccades of free-flying Drosophila with predictions based on hypothetical expansion- and rotation-sensitive control systems (Mronz and Lehmann, 2008). A significantly higher correlation was found assuming a rotation-sensitive system. These results led to the suggestion that image expansion on the side helps to achieve centering, for instance, in flight tunnels.

Our results suggest that blowflies do not rely on the retinal location of the FOE for eliciting saccades. Rather, we can predict to some extent the direction of saccades from the difference of expansion congruency in OF as well as from the total amount of OF on either side of the visual field during the previous ISI. The spatial pooling of local motion information across large parts of the visual field, as is the basis of both OF measures, might be accomplished by individual or groups of motion-sensitive lobula plate tangential cells (LPTCs) (for reviews, see Borst et al., 2010; Egelhaaf, 2006). Indeed, the differential output of two model LPTCs, sensitive to horizontal image motion, located in both halves of the visual system of a behavioral blowfly model (CyberFly) is appropriate to some extent to control saccade generation with the goal of avoiding collisions with the walls of a flight arena (Lindemann et al., 2008). Also expansion-based mechanisms of saccade control do not necessarily require neurons that are sensitive to image expansion; rather, such a mechanism could rely on the signals of a combination of LPTCs with different preferred directions. Very recently, de Vries and Clandinin (de Vries and Clandinin, 2012) identified three cell types in the lobula complex of Drosophila that are selectively sensitive to looming stimuli and mimic objects on a direct collision course. Because unilaterally presented looming stimuli trigger saccadic turns in flying Drosophila (Tammero and Dickinson, 2002a; Tammero and Dickinson, 2002b), such neurons – if they exist in blowflies as well – might be involved in the saccadic steering behavior characterized in our study.

Our results suggest that the lateral parts of the visual field are not involved in determining saccade direction in blowflies. This feature might be related to the way blowflies fly: during ISIs they predominantly fly forward, with some sideways component after saccades that shifts the FOE slightly towards frontolateral locations (Fig. 13D). Because the sideways movements increase with saccade size and saccades, on average, are largest in the cubic arena, only in the latter environment FOE locations exceeding 50 deg occur, though only at a very low frequency (Fig. 13D). Because the FOE is located only rarely beyond 30 deg in the lateral visual field under all tested spatial conditions, the prediction of saccade direction on the basis of the expansion congruency of OF may be much better for FOEs assumed at ±30 deg rather than ±45 deg (Fig. 13B). In contrast, in Drosophila – which are able to hover and fly sideways (e.g. Ristroph et al., 2009) – lateral and even rear portions of the visual field have also been shown to be involved in saccade control. Therefore, a mechanism eliciting saccades in Drosophila that favors image expansion with an FOE at more lateral retinal positions than in Calliphora is plausible from a functional point of view, as has been concluded on the basis of free-flight studies (Tammero and Dickinson, 2002a), but also in tethered flight (Bender and Dickinson, 2006; Tammero and Dickinson, 2002b; Tammero et al., 2004).

Predicting saccade direction correctly with a probability of approximately 80% is far from being perfect. However, the mechanisms proposed here are still very simplistic and do not take into account the well-established non-linearities of the motion detection system (for reviews, see Borst et al., 2010; Egelhaaf, 2009). Moreover, saccade control might be influenced not only by visual input but also by other factors not controlled for during our experiments. In fruit flies, frequent ‘spontaneous’ saccades not triggered by visual cues (Bender and Dickinson, 2006) have been interpreted as part of a systematic search strategy, for instance, to localize attractive odors (Reynolds and Frye, 2007). Based on this finding, it is possible that a fraction of the saccades observed in our tunnels is actively generated to probe the three-dimensional layout of the environment based on the saccade-accompanying sideways movements. Depth can be judged best based on the OF generated by translational movements performed perpendicular to the direction of interest. Hence, the sideways motion component after saccades may facilitate the extraction of depth information in the frontolateral visual field. Sideways motion plays an important role in the context of peering movements in locusts and praying mantids (Kral et al., 2000; Sobel, 1990). Accordingly, the lateral movements of bees often generated close to objects have been attributed to an active vision strategy that facilitates the judgment of distances to objects (see Boeddeker and Hemmi, 2010) and to solve navigation tasks (Dittmar et al., 2010; Braun et al., 2012).

Body saccades tend to occur with some regularity in both blowflies (see Schilstra and van Hateren, 1999) and Drosophila (Tammero and Dickinson, 2002a), which suggests that saccades are triggered by an internal clock. This idea is not supported by our experimental evidence, which showed that saccade frequency varied between the different environments (see also Tammero and Dickinson, 2002a; Dickinson, 2005). Moreover, a behavioral blowfly model generates regular saccades without explicit modeling an internal clock for saccade generation (Lindemann et al., 2008).

In blowflies, saccade rate in a given environment only slightly increases with translation velocity. This is similar in Drosophila (Mronz and Lehmann, 2008). However, the saccade rate of blowflies distinctly depends on the three-dimensional layout of the environment – being by far larger in the obstacle tunnel than in the straight tunnels. In the straight tunnels, the OF in frontolateral eye regions results from more distant structures, given that flies tend to be oriented along the tunnels' long axis. Hence, at a given flight velocity, frontolateral OF is weaker in the straight tunnels than in the obstacle tunnel, which may account for the lower saccade rate. Flies fly faster and need to change flight direction to avoid colliding with obstacles less frequently in open terrain. Accordingly, saccade rate is, on average, lower and saccade amplitudes are smaller the longer the estimated time-to-contact to an obstacle. Neurons, signaling time-to-contact either of an approaching object or during the animal's approach towards surfaces, have been described in various animals, including frogs (Nakagawa and Kang, 2010), gerbils (Shankar and Ellard, 2000), pigeons (Frost and Sun, 2004), hawk moths (Wicklein and Strausfeld, 2000), mantids (Yamawaki and Toh, 2009), locusts (Rind and Simmons, 1997; Gabbiani et al., 1999) and Drosophila (de Vries and Clandinin, 2012; Fotowat et al., 2009). In blowflies, such neurons have not yet been characterized.

The sideways movement components after saccades probably have their origin in inertial forces because the sideways velocity after saccades increases with saccade amplitude. Nonetheless, flies may actively control sideways movement, although Drosophila and hoverflies appear to do this to a much larger extent than blowflies (Schilstra and van Hateren, 1999; Ristroph et al., 2009; Sugiura and Dickinson, 2009; Braun et al., 2010; Geurten et al., 2010). Partial drift compensation presumably results from the fact that blowfly saccades are a combination of yaw, roll and pitch movements, leading to a banking behavior like that of airplanes during sharp turns (Schilstra and van Hateren, 1999; van Hateren and Schilstra, 1999; Braun et al., 2010). The sideways motion components of Calliphora just after saccades have been suggested to have functional significance. Previous studies have shown that motion-sensitive LPTCs probably involved in flight control directly encode information about intersaccadic sideways motion of the animal, and thus indirectly about the spatial layout of the environment (Karmeier et al., 2006; Kern et al., 2005; Kern et al., 2006; van Hateren et al., 2005). In addition, a saccadic controller as part of a model of blowfly visual course control (CyberFly) with sideways drift was shown to be slightly superior to a controller without sideways drift, because it is thought to facilitate the CyberFly to extract distance information in the frontolateral visual field (Lindemann et al., 2008).

The temporal relationship between saccades and a corresponding decrease in flight velocity differs for Drosophila and blowflies. Drosophila's peak saccade velocity tends to coincide with the drop in flight velocity (Mronz and Lehmann, 2008; Tammero and Dickinson, 2002a). The difference may be due to the fact that Drosophila saccades were not resolved on the basis of body orientation, but as a change in flight direction. As a consequence of inertia, these changes may occur slightly later than changes in body orientation (Schilstra and van Hateren, 1999). Trajectory-based saccade characterization is likely to lead to considerably larger saccade durations in Drosophila [exceeding 100 ms (Mronz and Lehmann, 2008; Tammero and Dickinson, 2002a)] than described for blowflies (approximately 30 ms; Fig. 6F, Fig. 11) (Schilstra and van Hateren, 1999). If body yaw is determined and used for saccade analysis – as in the present paper – the saccade duration of Drosophila drops down to the 30 ms range (Bergou et al., 2010; Fry et al., 2003). Head saccades in blowflies are even shorter than body saccades (Schilstra and van Hateren, 1999; van Hateren and Schilstra, 1999), indicating that the head can turn even more rapidly than the body, which prolongs the ISIs that we here suggest to be used by blowflies for controlling saccades and translation velocity.

A portion of the experiments was conducted by undergraduate students and by Frederique Oddos. We gratefully acknowledge the assistance of Grit Schwerdtfeger. Claudius Strub wrote the software to calculate optic flow.