Visual control of host pursuit in the parasitoid fly Exorista japonica

Some insect predators visually pursue and catch prey (e.g. Gilbert, 1997; Olberg et al., 2000). The tachinid fly Exorista japonica, which is a parasitoid of many kinds of lepidopterous larvae, pursues a crawling larval host on foot. The pursuit of a suitable larval host in E. japonica is presumed to be visually controlled because this fly will pursue a moving rubber tube (C. Tanaka, personal communication). After pursuing a host, the female fly lays her eggs directly on the host cuticle. The first-instar larvae that emerge penetrate the host integument after a few days of incubation (Nakamura, 1994).

E. japonica tends to oviposit on the head and thoracic segments of its larval host because the host attempts to remove eggs from its abdominal segments, but not from its head and thoracic segments (Nakamura, 1997). Hence, adaptive behaviour for the fly should be to pursue and approach the host’s head.

To investigate the control mechanism for this task, we presented E. japonica with larvae of the common armyworm Mythimna separata and videotaped the pursuit behaviour. Frame-by-frame analysis clarified how pursuit behaviour is controlled by visual stimuli such as the visual angle, angular position and angular velocity of the host.

Tachinid flies, Exorista japonica Townsend, and common armyworms, Mythimna separata, were obtained from our stock cultures. The flies were reared according to the methods of Nakamura (1996) and Tanaka et al. (1999). The worms were reared on an artificial diet (Silk Mate, Nihon Nosan Kogyo, Japan) according to Kanda (1991). All rearing was performed at 25±2°C, 40–80 % relative humidity and on a 16 h:8 h L:D photoperiod.

Female flies that had no experience of oviposition were used. The fly was presented with the worm in a circular glass arena (8 cm in diameter, 9 cm in height). A circle of white paper was laid on the bottom of the arena. Pursuit of the host by 14 flies was videotaped (Sony; Digital Handycam DCR-TRV10) at a speed of 30 frames s^–1 from a dorsal view in sunlight and pursuit by 17 flies was videotaped under fluorescent lamp illumination (60 Hz). For each fly, one pursuit was recorded. Recording was performed at 22–25°C.

Video images of the pursuit were digitized by computer (Sony; PCV-R52). For each frame, the positions of the head and tail of both the fly and the larval host were measured automatically. Their positions were superimposed on the video image, and some data were corrected manually from the video image. Corrected data were filtered by smoothing. In the smoothing procedure, data for a given frame were calculated by averaging the data for three successive frames: the previous, current and following frames. From the x,y coordinates of positions, the angle of the fly’s longitudinal body axis relative to externally based coordinates (Φ) and the angular positions of the head (θ_H) and tail (θ_T) of the host relative to Φ were calculated (Fig. 1A). The centre, a position half-way between the head and tail, was calculated for the fly and the host. Measurement and transformation of data were performed by a program written in the laboratory with Microsoft Visual Basic 6.0.

When the flies settled on the bottom of the arena, a host larva was introduced. Most flies reacted to the appearance of the larva and turned towards it (Fig. 2). However, some flies took off and landed on the wall or top of the arena without approaching the larva. When these individuals detected the presence of the larva, they took off again and landed on the ground near the larva. After orienting towards the larva, the fly pursued the crawling larva on foot, continuing its pursuit until it had oviposited a few times. The data presented here were derived only from pursuits of crawling larvae, although other behaviours such as examination, oviposition and escape (see Discussion) were observed. The total duration of pursuit analyzed was 1903 s (57 098 frames).

Although the fly rolls its head with respect to its body during head cleaning, no detectable yaw movements of the head were observed during pursuit. The angular measurement of the fly’s longitudinal body axis was therefore assumed to represent a valid measure of the direction of the head. We could not assess this with greater precision because of the limitations of the video system in which the fly was too small to allow accurate measurement of head direction.

The flies pursued their host discontinuously with repeated stop-and-run movements. During a run, their movements consisted of rotation and translation (Fig. 1B). Translation was resolved into forward and sideways components because the fly was able to move sideways, keeping the direction of its body axis constant, like a hoverfly’s flight (Collett and Land, 1975).

When a distant fly first detected the host, it oriented towards and approached the host (Fig. 3A). During pursuit, the fly seemed to keep the direction of its longitudinal body axis (Φ) between the absolute angles of the positions of the host’s head (Φ+θ_H) and tail (Φ+θ_T) (Fig. 3B). After the approach, the fly tracked the target with sideways rather than rotational body movements, while keeping its distance to the target constant (Fig. 4).

The ‘stop’ and the ‘run’ phases were determined from the plots of translation and rotation as a function of time. The ‘stop’ phase was defined as the period during which there was little translation or rotation (see Fig. 4). The ‘run’ phase was defined as the period between the ‘stop’ phases.

The mean individual durations of the ‘stop’ and the ‘run’ phases in sunlight were 0.346±0.678 and 0.149±0.066 s (means ± s.d.), respectively (N=1560). Variation in the duration of the ‘stop’ period was larger than that of the ‘run’ period (Fig. 5). When the larva was not crawling, the fly tended to remain still, suggesting that the larva’s motion affected the ‘stop’ duration of the fly. The ‘stop’ duration seemed to be inversely related to the retinal velocities of both the head (Δθ_H) and tail (Δθ_T) of the larva during the ‘stop’ period (Fig. 6). The reciprocal of ‘stop’ duration was significantly related to the retinal velocities of both the head (Δθ_H; N=1560, t=6.382, P<0.0001) and tail (Δθ_T; N=1560, t=6.759, P<0.0001), although these relationships were weak (r²=0.025 for head; r²=0.028 for tail).

The total amount of translation during a run was significantly related to the duration of the run (N=1560, t=53.377, P<0.0001; Fig. 7A). The fly accelerated during first 50 ms of a run, then walked at a constant speed of approximately 30 mm s^–1. The total amount of forward translation during a run was usually between –3 and 5 mm, whereas sideways translation usually ranged from –5 to 5 mm (N=1650; Fig. 8A). Although translation occurred in all directions, it was most commonly directed at 30–75° to the forward direction (Fig. 8B).

The total amount of rotation during a run was related to run duration (N=1560, t=10.275, P<0.0001; Fig. 7B). However, in contrast to translation, the regression was weak. Therefore, run duration seemed to depend mainly on translation distance.

During a run, the fly seemed to orient towards the larva. The correlation coefficient between the total amount of rotation during a run and the angular position of the host head was greatest; those for the centre or tail were lower (Fig. 9A). This suggests that the fly orients towards the host’s head. The highest correlation coefficient between the fly’s rotation and the larval host’s head position was found at a delay of one frame (33 ms). Thus, the latency of the run was estimated to be 30 ms. It is difficult to estimate this variable more precisely because the host was crawling so slowly that there was little difference in its angular position between successive frames while the fly had stopped. However, subsequent analyses do not depend critically on the accuracy of the latency estimate. The regression between the fly’s rotation and the larva’s head position 33 ms (one frame) earlier was significant (N=1650, t=26.058, P<0.0001; Fig. 9B). The rotation gain (regression coefficient) was 0.136.

When the larva’s head was moving towards the fly’s midline, the rotation gain was 0.100 (N=244, t=6.543, P<0.0001; Fig. 10A), whereas when the larva’s head was moving away, the gain was 0.141 (N=1406, t=25.464, P<0.0001; Fig. 10B), suggesting that the sign of the host’s visual angular velocity affects the gain of the fly’s rotation. However, the partial correlation coefficient between the fly’s rotation and the visual angular velocity of the host’s head (when the angular position of the host’s head was held constant) was weak (–0.031). There is no conclusive proof that the visual angular velocity of the host affects the rotation gain.

The scatterplot for rotation as a function of the host’s angular position (Fig. 9B) suggests that rotation gain is positive at error angles (the host’s angular position) between –90 and –30° and between 30 and 90°, but that gain may be zero between –30 and 30°. When the data were divided into two groups depending on whether the sign of host’s head position was plus (right) or minus (left), the regression coefficient between rotation and host’s head position became larger for each group (0.212 for right; 0.165 for left). Both regression lines crossed the abscissa at an error angle of approximately ±15°, suggesting the possibility that the fly orients only to error angles larger than 15°.

To eliminate measurement errors, the data were used to analyse translation only when the amount of translation was greater than 2 mm, because small translations can also be caused by rotation. During a run, the fly approached the host’s head as shown in Fig. 3A. The total amount of both forward (N=753, t=–10.320, P<0.0001; Fig. 11A) and sideways (N=753, t=30.899, P<0.0001; Fig. 11B) translation during a run was significantly related to the angular position of the host’s head 33 ms before the start of the run. The amount of forward translation was large when the host’s head was near the fly’s midline, while sideways translation was large when the host’s head was distant from the fly’s midline. This relationship assists the fly to move towards the host’s head (N=753, t=42.228, P<0.0001; Fig. 12).

The visual angle subtended by the target can be used to estimate distance if target size is known. Thus, it is probable that the visual angle subtended by the host affects the amount of translation performed by the fly. The total amount of forward translation during a run was negatively related to the visual angle subtended by the host (N=753, t=–14.464, P<0.0001; Fig. 13A). When this angle was small, the fly moved forward a greater distance. In contrast, the total amount of sideways translation during a run was positively related to the visual angle subtended by the host (N=753, t=7.214, P<0.0001; Fig. 13B).

When illuminated by a fluorescent lamp, the fly seemed to pursue the host normally. Most flies readily pursued the host; illumination by the fluorescent lamp did not seem to affect the frequency of pursuit. There were no large differences between the results obtained for flies illuminated by sunlight and those for flies illuminated by a fluorescent lamp (Table 1), suggesting that artificial illumination has little effect on the pursuit behaviour of this fly.

Female E. japonica use multimodal cues for seeking a host and oviposition. First, the fly is attracted by odours emitted from plants damaged by its host (Kainoh et al., 1999). After arriving at these damaged plants, the fly finds a host visually and approaches it (Tanaka et al., 1999). Once the fly has approached within approximately 5 mm of a host, it begins ‘examination’ behaviour consisting of facing and touching the host with its front tarsi (Nakamura, 1997). During examination, the fly checks the texture and curvature of the host; it has been shown to prefer a cylindrical shape rather than a flat board or a cube (Tanaka et al., 1999). Finally, the fly extends its ovipositor and attaches an egg to the host cuticle.

The present study analyzed only the visually controlled stages of host pursuit. Our analysis indicates a correlation between the fly’s movements during a run and the visual stimuli provided by the host, suggesting that pursuit of the host is controlled mainly by visual cues. However, it is probable that the motivational state of pursuit behaviour is affected by stimuli from other modalities such as the odour of the host and its faeces. It is therefore important also to examine the effects of these odours on the pursuit behaviour of the fly.

The following observations were made: (i) rotation during a run is positively correlated with the angular position of the host’s head; (ii) the direction of translation during a run depends on the angular position of the host’s head; and (iii) forward translation is negatively correlated with the visual angle subtended by the host.

Many behavioural studies have reported the visual control of pursuit in insects (Land and Collett, 1974; Collett and Land, 1975; Zeil, 1983; Zhang et al., 1990; Gilbert, 1997). The control of rotation and translation in E. japonica appears to be similar to that in the hoverfly Syritta pipiens (Collett and Land, 1975), and the control of stop-and-run movements is similar to that in the tiger beetle Cicindela repanda (Gilbert, 1997).

During pursuit, changes in angular orientation of the hoverfly depend on the position of the target image. Targets outside the fovea are fixated by a rapid body saccade with a gain of 0.9. Although the rotation gain of E. japonica was much less than this, it is able to pursue the host sufficiently well because of the low velocity of the host’s crawling movements. In addition, the fly can move in any direction with respect to its longitudinal axis; the low rotation gain does not restrain the direction of pursuit.

The relationship between the rotation and target angle (angular position of target) suggests the possibility that E. japonica orients only to target angles greater than 15°. It is also possible that the fly tracks target angles of less than 15° with its head and those greater than 15° with body saccades. Although the fly’s head could rotate during pursuit, we did not detect any such yaw movements of the head in the present study because the size of the fly in the video images was too small to measure the head angle. Further experiments using tethered flies are required to examine these possibilities.

The sign of the host’s angular velocity seems to affect the rotation gain of E. japonica. This suggested that the fly might utilize information about target motion to predict the future position of the target. However, partial correlation analysis did not support this possibility. Although predictive tracking has been reported in some insects, such as the praying mantis (Rossel, 1980) and the hoverfly (Collett and Land, 1978), there is no conclusive proof that E. japonica does this.

The hoverfly performs sideways movements during pursuit flight. Normally, the sideways velocity of the hoverfly is not controlled relative to the target’s position. However, when the target moves slowly, sideways velocity depends on target position. As Collett and Land (1975) point out, sideways tracking will only be as efficient as rotational tracking when the distance between the pursuing fly and the target is small. This may explain why E. japonica shows sideways movements after approaching within a certain distance of the host (Fig. 4). In addition, sideways movements enable the fly to keep its body axis perpendicular to the host, facilitating subsequent examination and oviposition behaviour. Another possible function of sideways movements is motion camouflage (Srinivasan and Davey, 1995). However, in the present case, it is not critically important for the fly to conceal its motion, because its host has a weak visual system (see below).

Although E. japonica seems to respond to the motion of the larval host’s head during pursuit, it is unlikely that the fly can discriminate the head from the tail by pattern recognition. When the fly is presented with a moving rubber tube, the fly pursues it as if the leading edge of tube were the host’s head. Thus, the fly may walk and turn towards the leading edge of a variety of moving objects.

E. japonica, like the hoverfly, seemed to use the visual angle subtended by the target to estimate distance and to control forward translation. Although we found this relationship for the horizontal angle subtended by the host, the vertical angle may also be used to estimate distance. When the fly pursues the host at close range, it may also use binocular vision for distance estimation. However, this possibility cannot be considered further at present because there is little information available regarding binocular vision in E. japonica.

In E. japonica, the visual angle subtended by the target seems to be used to control not only forward translation but also sideways movement. If the size and speed of the target are constant, the fly must move faster the closer it is to the target. Thus, the fly’s tendency to move sideways faster with larger angles subtended by the target will assist successful tracking.

The duration of the fly’s run depends on the amount of translation during the run, while the duration of the ‘stop’ interval seems to be inversely related to the angular velocity of the host, as has also been reported for the tiger beetle (Gilbert, 1997). However, we could find no clear correlation between stop duration and visual stimuli such as angular position, angular velocity and the visual angle of the host. The duration of the stop and run phases may therefore be controlled by some unknown factor. The visual stimuli received from a freely moving host are complex and varied; to analyze host pursuit in detail, we are developing a simplified system in which a host model is moved by a motor. The present finding that fluorescent illumination does not seem to affect pursuit behaviour suggests that future experiments can be carried out in the laboratory.

There are several possible functions of the stop-and-go running patterns shown by E. japonica. Miller (1979) and Gilbert (1997) suggest that stop-and-go running patterns may serve a sensory function: each pause allows sensory information to be gathered and analyzed more effectively than during a run. In the case of tiger beetles, the image of the prey is degraded during high-velocity running because of the structured background, and the beetle has difficulty in detecting the prey reliably. During the stop period, it is easier for the beetle to detect a moving prey against a stationary background.

Similar reasoning may account for the stop-and-go running pattern seen in E. japonica. Although the stop period provides tiger beetle prey with more opportunity to escape, E. japonica runs little risk of losing its host because of the host’s low velocity. Thus, stop-and-go running enables the fly to pursue its host precisely with little risk of losing the host.

It is also possible that it is easier for the fly during a stop to detect a possible attack from the host. M. separata larvae sometimes sway their head and thoracic segments and try to bite the fly when touched by it (Nakamura, 1997). The fly’s escape during a stop interval was sometimes observed in the present study; stop-and-go running may therefore be safer for the fly than continuous running.

It is unlikely that the fly is evading detection by stop-and-go running. Most of its hosts, lepidopterous larvae, have six stemmata. Because each stemma has a retinula of only seven photoreceptor cells, the resolution of each stemma is weak. In addition, the receptive fields of the six stemmata are widely separated (Ichikawa and Tateda, 1982), suggesting that the visual system of these lepidopterous larvae is unlikely to be able to detect small objects against a structured background.

The authors are grateful to Dr I. A. Meinertzhagen for reading the manuscript and providing valuable comments and to Dr J. Okada for useful advice on image analysis.