Visual cues eliciting the feeding reaction of a planktivorous fish swimming in a current

High sensitivity to motion across the visual field is characteristic of many visual systems. For example, in many vertebrates, motion detection cells in the brain can have directional sensitivity and maximal responses to stimuli moving at different speeds (Barlow and Hill, 1963; Cronly-Dillon,1964; Jacobson and Gaze,1964; Guthrie and Banks,1974; Kawasaki and Aoki,1983). In addition, the ability to detect a lower contrast stimulus (i.e. the modulation sensitivity), usually occurs at intermediate presentation frequencies or velocities(Kelly, 1972; Sekuler et al., 1978; Sekuler et al., 1990). The detection of predator or prey by fishes, although it may involve characteristics such as shape, spatial pattern and chromatic aspects,certainly responds dramatically to targets in motion. A central part of any visual stimulus must be the contrast of the target compared with the radiance of the background space light. Because zooplankton are small and usually semi-transparent, in general, they appear as low-contrast objects under viewing conditions that minimize the effects of scatter and reflection of downwelling light into the direction of view. However, near the surface where the downwelling irradiance is high relative to the horizontal irradiance,semi-transparent plankters will scatter downwelling light from their external and internal refractive interfaces and appear as bright targets, as long as the viewing distance is relatively short and the downwelling irradiance is not diffuse (i.e. the plankters are irradiated by the solar beam) (E. R. Loew and W.N.McF., unpublished). With depth, or as the veiling brightness increases with an increase in viewing distance, this phenomenon becomes less obvious. Thus, the contrast of a semi-transparent object when viewed horizontally depends on the veiling brightness, the viewing distance, the depth, the characteristics of the downwelling irradiance (diffuse vs beam) and the differences in refractive index between any intra-organismal compartments(Videen and Ngo, 1998).

Planktivorous fishes that feed in currents provide an example where the motion between a fish and its zooplankter prey are consistent and definable(Hobson, 1972; Hamner et al., 1988). By feeding zooplanktivourous fishes in flumes, Kiflawi and Genin(1997) demonstrated that peak ingestion can occur at intermediate current speeds and McFarland and Levin(2002) showed that individuals will cease feeding at higher speeds. Four phenomena may chronologically occur before a fish reacts to a particle: (a) detection, (b) recognition (c)decision (i.e. threshold of triggering mechanism at the sensory system level)and (d) motor response. Behaviourally, only the feeding reaction can be observed (i.e. the motor response). The timing of the feeding reaction should be, however, largely dependent on the timing of the detection. In particular,the distance at which a feeding reaction is triggered is limited by the maximum detection distance.

In investigating predator–prey interactions, Dill(1974) measured the escape responses of prey and modelled the rate of change in apparent size of a predator (i.e. the loom) as it approached the prey. For fish feeding on plankters in a current, angular velocity, apparent size and loom were described mathematically for particles that approached head-on and at various distances offset from the position of the fish in the current(McFarland and Levin 2002). In that study, the temporal-resolution of a fish's reaction to an incoming plankter was limited by the 2D spatial resolution and by the low temporal resolution of the system used to capture time-sequence images for analysis(McFarland and Levin 2002). In this paper, using a high-speed video system, we examine the reaction of fish under similar conditions and evaluate, in three dimensions, what attribute(s)of a plankter appear to influence the feeding reaction of fish. We test three alternative hypotheses: (1) The feeding reaction of planktivourous fish may be triggered by a fixed apparent size (i.e. angular size) of the approaching particle. (2) Fish may react to plankton once it reaches a given threshold angular velocity as it is carried passively by the current. (3) The mechanism triggering a fish's reaction may be the loom (i.e. the rate of change of the angular size) of the prey while approaching the fish. The experiments were carried out using three different current speeds and two treatments, which provided plankton with different properties of contrast (i.e. semi-transparent and darkened Artemia) in order to test the effect of contrast on detection mechanisms.

Shiner perch (Cymatogaster aggregata Gibbons) were collected using a beach seine at Jackson Beach, San Juan Island Washington, USA, during August of 2000 (total length 6.7±0.08 cm; mean±s.e.m., N=8), and April 2002 (total length 8.8±0.17 cm; N=8). Fish were held at the Friday Harbor Laboratories in flow-through seawater tanks at 12°C for one week before use. Fish were exposed to a 12h:12h light:dark cycle, provided through fluorescent illumination from above. Integrated irradiance was 5.1×10¹⁷ photons cm^–2 s^–1 over 300–750 nm (OCEAN OPTICS S1000 spectroradiometer; Dunedin, FL, USA).

Experiments were carried out in a flume tank as described in McFarland and Levin (2002) with a working section 50 cm long, 12 cm wide, and a water depth of 17 cm(Fig. 1). The experimental tank was illuminated from above with a daylight fluorescent lamp (two 20 W lights,model F20T12 GE). Integrated irradiance was 8.2×10¹⁷ photons cm^–2 s^–1 over 300–750 nm, measured at the water surface level of the experimental tank (OCEAN OPTICS S1000 spectroradiometer).

Water flow was generated by a two-bladed propeller and current speed controlled by varying the voltage via a variac to the AC/DC motor(Dayton AC/DC model 2MO338 motor; Lincolnshire, IL, USA). A catch screen was placed 10 cmdownstream from the working section to isolate fish in the flume. To assure repeatability of current speed for different trials, the voltage output of the variac was prescribed through the use of a digital voltmeter. Three layers of collimators (5 cm wide) were positioned upstream to produce a homogeneous flow through the working section. Viewed through the collimators,the upstream aperture of the collimator tubes provided a dark background to the fish. The background radiance from the collimator, as measured with an OCEAN OPTICS S1000 spectroradiometer between 300 and 750 nm, was only 3.1%when compared with the radiance from a white Teflon reflective surface and was similar to the radiance from a black surface (4.3%). Flow speed was calibrated from dye injected in the water upstream from the collimators and filmed at 125 images s^–1 using a high-speed video camera (Red Lake, model 1000 S series motionscope; San Diego, CA, USA). The current flow as seen by discrete dye travel in the horizontal dimension was laminar over the working section, with minor turbulence only observed just upstream from the catch screen. To avoid wall effects, only trials in which a fish initiated a reaction towards the prey more than 2 cm from the walls, bottom and water surface, were analysed.

Fish movements were recorded at 125 images s^–1 with the video camera placed perpendicular to the flume. We capture 3D-movements of each fish by positioning a 30 cm×60 cm mirror at 45° above the experimental tank (Fig. 1). Therefore, lateral and top images of the fish were produced at one time, the bottom half of the camera filming the fish laterally to determine height in the water column and position along the length of the flow tank, while the upper half of the camera filmed through the mirror and viewed the fish from above against the floor of the tank, allowing positions across the width of the flow tank to be determined. Frame-by-frame analysis of the horizontal and vertical positions of each fish at the time of first departure from a steady swimming mode and at the moment of engulfment of a plankter (Artemianauplii) was accomplished though use of the software Scion Image (NIH image analysis). The 3D-coordinate position (x, y, z) of each fish in the flume was established against two ruled grids, one placed on the backwall of the flume and another on the bottom of the flume. For horizontal positions of the fish (x, y), the midpoint between the eyes of each fish was used as reference, and for the vertical position (z) the centre of the fish's observable (right) eye.

Artemia nauplii were used as in Kiflafi and Genin (1997), since they have a shape and size similar to the prey of shiner pearch, calanoid copepods, and because they lack escape behaviour(Coughlin and Strickler, 1990; Trager et al., 1994). Individual fish did not require conditioning to swim in the flume and, even though naïve, fed on Artemia when introduced [also reported for natural zooplankton by McFarland and Levin(2002)]. As suggested by Webb(1982) for visual reaction in fish, the size of an Artemia (i.e. particle diameter) was considered as the mean between its length and width (mean size 0.053±0.002 cm; N of subsample =10).

Two sets of experiments differing in prey type were carried out. Perch collected in August of 2000 were used in the first set of trials, where naturally semi-transparent Artemia nauplii were released into the flume-tank (Fig. 2). In a second set of experiments, shiner perch collected in April 2002 were presented with darkened Artemia (Fig. 2). A stone for grinding a Sumi ink stick with water was used to make liquid black ink. Darkening was obtained by adding a suspension of black ink (0.037 g in 1 l of water) into the Artemia's tank overnight. The Sumi ink produced a darkening of the Artemia gut.

In both experiments, fish were acclimated in the flume tank at zero current speed for 40 min before each experiment. Current speed was then increased slowly over a 2 min period to the chosen trial speed. The order of trial speeds was chosen at random among the six possible combinations of orders using slow (0.52 body length s^–1 for semi-transparent Artemia and 0.66 body length s^–1 for darkened Artemia), intermediate (1.88 body length s^–1 for semi-transparent Artemia and 1.60 body length s^–1for darkened Artemia), and fast flow speeds (3.12 body length s^–1 for semi-transparent Artemia and 3.01 body length s^–1 for darkened Artemia). These speeds are within the range in which C. aggregata swims using pectoral fin locomotion, i.e. they are lower than the gait transition speeds (from pectoral fin to caudal fin locomotion), observed in this species and size(Mussi et al., 2002). Ten minutes after a trial speed was reached, live darkened or natural(semi-transparent) Artemia nauplii were introduced from behind the collimator baffles and in the approximate centre of the section (i.e. midway between the walls and in midwater). We assumed that, after recirculating in the flow tank (i.e. Artemia took a full trip around the flow tank before being preyed upon), the nauplii were distributed randomly over the tank cross section. The numbers of trials for the natural (semi-transparent) Artemia experiment were 15, 15 and 19 for slow, intermediate and fast current speeds, respectively, and 10 trials at each current speed were performed using darkened Artemia. Each trial involved one single fish. For each individual in both darkened and natural conditions, up to four prey captures were analysed at each current speed.

Fish first reacted to an approaching particle by a noticeable change in position at the time T_R(Fig. 3). Because a latent period must occur between when the reaction is triggered by the particle and the actual visible reaction of the fish(Dill, 1974; Domenici, 2002), we have defined the theoretical time at which the particle triggers fishes' `inner reaction' as T<sub>L</sub> (the reaction at the sensory system level). The time of the fish's first `visible reaction' was defined as T_R (the reaction at the motor level), and the time at which the fish captured the particle was defined as T_C. The hypothesized delay between T_L and T_R would be due to the conductive times of the neurosensory and neuromuscular systems. While we were not able to determine T_L experimentally, its definition is needed in order to facilitate further theoretical considerations.

Neither darkened nor semi-transparent Artemia nauplii were visible on the video images. However, the jaw protrusion of the fish that corresponded to the engulfing of an Artemia was visible. Previous video recordings from close up at all speeds indicated that the position of the tip of the mouth at maximum mouth protrusion was a good estimator of the position of an Artemia at the time of capture along the tank axis. The vertical position of the particle was therefore estimated as the midpoint between the two edges of the mouth tip at maximum protrusion, while the position of the particle across the width of the tank was estimated by using the top image of the tip of the mouth at the time of maximum protrusion, in accordance with previous observations. Therefore, the time T_R and the position of the mouth at T_C when a fish captured a particle (easily detectable by jaw protrusion), were used to estimate the position of the nauplius at T_R. This assumes that nauplii moved passively within the current (T_R to T_C in Fig. 3). The vertical distance between a fish and a food particle at T_R (Z_vertical) was estimated from video side views. The video top view provided estimates of Y_forward (the distance between the fish and the prey approaching along the current axis), and X′_lateral(the lateral distance between the fish and the prey along the axis perpendicular to the current, in the horizontal plane).

The geometry in Fig. 3 was reconstructed for each capture event, and all analyses were done using the plane D_totalX_lateralY_forward, where D_total is the estimated total distance between the fish's eyes and the food at T_R, and X_lateralis the overall lateral distance between the fish's eyes and the axis of motion of the prey along the current. D_total is at an angle alpha(α) from the fish's axis. Left and right responses were pooled as if they were all responses to particles approaching on the right side of the fish.

The angular size threshold (β_R, i.e. β at T_R) of the prey was calculated as the angle subtended by the plankton onto the fish's eye at the time of reaction(Fig. 3). The apparent loom threshold (λ_R, i.e. loom at T_R) was calculated as in Dill's (1974)equation, modified for the case of approaching objects not necessarily in line with the fish's body axis, following McFarland and Levin(2002).

The effect of current speed and plankton type (natural vsdarkened) was tested using two-way ANOVAs (analysis of variance) for each feeding reaction variable (i.e. D_total, X_lateral, Y_forward, Z_vertical, α, angular size, angular velocity, loom). To account for multiple simultaneous two-way ANOVAs, the level of significance was adjusted within columns using the sequential Bonferroni technique(Rice, 1989).

If the time at which a particle first triggers a fish's reaction (i.e. the threshold at the sensory system level at T_L), is caused by a given angular size (β_L), angular velocity(ω_L) or loom (λ_L) of an approaching particle, independent of its speed, then the `apparent' angular sizeβ _R, angular velocity ω_R or loomλ _R (i.e. at T_R), should vary with current speed as the approaching object covers a longer distance between T_L and T_R (assuming a constant latency) when travelling at higher speeds. The potential effect of current speed on the threshold for the fish's response can be visualised by plotting how angular size, loom, and angular velocity vary as a function of Y_forward (Fig. 4). Since current speed for each trial is constant, decreasing values of Y_forward also correspond to time. The angular size increases continuously as the particle travels along Y_forward (Fig. 4A) and its rate of increase is higher the smaller the lateral distance X_lateral is. Therefore, assuming the theoretical angular size β_L to be constant at different flow speeds, the experimental angular size β_R should increase with current speed.

Loom increases as the particle travels along the current up to a certain(relatively small) value of Y_forward, and then decreases again (except if X_lateral=0 cm). For example, for an X_lateral value of 2, loom increases up to Y_forward values around 1.5 cm(Fig. 4B). Therefore, if we assume λ_L to be constant at different current speeds, thenλ _R should increase with current speed up to Y_forward values of about 1.5 cm, as the distance covered by the particle along Y_forward would increase with current speed. Loom decreases only when very close to the fish. The angular velocityω, conversely, increases continuously as the particle travels along Y_forward, although this increase has different rates depending on X_lateral (Fig. 4C). In particular, if we assume ω_L to be constant at different speeds, then ω_R should increase with current speeds (as Y_forward gets shorter). Therefore, we expect that the angular variable(s) implicated in the mechanisms triggering the feeding reaction should increase with speed.

For those angular variables that are found to increase with speed, we estimated latency as follows. To find a theoretical constant value of latency for all current speeds, an index of homogeneity (I_H)between each average `angular variable' (i.e. angular size, angular velocity or loom) calculated at T<sub>L</sub> (i.e. assuming a given latency)at different current speeds was calculated using (maximum `angular variable'– minimum `angular variable') / minimum `angular variable'. Each angular variable at T_L was calculated as the theoretical value that the fish would experience if, for each sequence, the particle had triggered a reaction at a position with X_lateral at T_L being the same as at T_R (since a straight course was assumed), while Y_forward was considered longer at T_L than at T_R by a distance that was calculated as the current speed multiplied by the theoretical latency.

The alternative hypotheses for the three mechanisms triggering feeding reaction were further tested as follows:

1. The frequency distributions of X_lateral were compared between darkened and semi-transparent Artemia experiments, since it can be assumed that low values of X_lateral may imply that the feeding reaction is triggered by an angular size or a loom mechanism,while high X_lateral values suggest a mechanism based on angular velocity. This is because, at any Y_forward,angular size and loom should be highest for particles coming head on, while the angular velocity should be highest at low values of X_lateral only for very small values of Y_forward (not observed here; see theoretical values in Fig. 4). Therefore, if an angular size or a loom threshold are used, the first particles to achieve threshold will be those with the smallest X_lateral. Conversely, if angular velocity is used, the first particles to reach threshold should always be off line from the fish's axis, i.e. at X_lateral>0.

2. The potential roles of angular size, angular velocity and loom in the feeding reaction of the semi-transparent and darkened Artemia,respectively, was tested by comparing the Y_forward and X_lateral positions of the particles at T_R and the theoretical values of these positions based on the average experimental angular size, angular velocity and loom at each current speed.

Darkening (dark vs semi-transparent Artemia) had an effect on most variables measured (Table 1 and 2). In particular, the reaction distance D_total was significantly greater for the semi-transparent Artemia (Table 1 and 2, Fig. 5). The results from a two-way ANOVA (adjusted using the sequential Bonferroni method; Rice, 1989) revealed that current speed had no significant effect on all reaction distance parameters(i.e. D_total, X_lateral, Y_forward, Z_vertical), while darkening did (Table 2). No significant interaction effect of current speed and darkening was found. Neither current speed nor darkening had any effect on reaction angle α(P>0.5 and P>0.05, respectively), and the interaction effect was not significant according to the sequential Bonferroni technique(see Table 2).

Current speed and darkening had no effect on angular size, and no interaction effect between darkening and current speed was found(Table 2 and Fig. 6). Conversely, both angular velocity and loom were affected by current speed (P<0.001 in both cases), as well as darkening (P=0.0106 and P<0.001, respectively), and the interaction of these two factors was significant for loom (P<0.005)(Table 2 and Fig. 6).

Angular variables involved in triggering the feeding reaction to plankton are expected to increase with current speeds, except for loom at very small distances (see above and Fig. 4). While loom may theoretically decrease at small distance from the fish, loom increases with decreasing Y_forward when the Y_forward and X_lateral ranges of all observed speeds are considered (only intermediate speed is shown as an example, in Fig. 4). The results show that with semi-transparent Artemia, β_Rdoes not vary significantly with speed(Table 2 and Fig. 6), ω_Rincreased with current speed, while λ_R was lowest at the intermediate speeds, and higher at the low and the high current speeds. Therefore, an index of homogeneity I_H was calculated only for angular velocity. I_H was highest when a theoretical latency of 385 ms was considered. That is, if we assume a latency of 385 ms,the values for the angular velocity when the fish's `inner reaction' occurs(ω_L) are relatively constant (i.e. 0.26 rad s^–1, 0.28 rad s^–1 and 0.26 rad s^–1 for slow, intermediate and fast speeds,respectively).

When darkened Artemia were used, the angular sizeβ _R is relatively constant at different current speeds, while both ω_R and λ_R increased with current speed in darkened Artemia (Fig. 6). Therefore, I_H was calculated for both loom and angular velocity. However, a maximum I_H was found only for loom, and it resulted in a latency of 230 ms. Using this theoretical latency, λ_L is relatively constant (i.e. 0.013 rad s^–1, 0.015 rad s^–1 and 0.013 rad s^–1 for slow, intermediate and fast speeds, respectively). Conversely, I_H increased indefinitely when increasing values of latency were applied to ω_L.

The frequency distributions of X_lateral of the semi-transparent vs darkened treatments were compared using a Chi square test. Four bins (0–1, 1–2, 2–3, >3 cm) were used for the comparison (Fig. 7). The distributions of the two treatments are significantly different(Chi² 22.69; 3 d.f.; P<0.0001). The highest peak in the distribution for the darkened treatment was in the bin X_lateral=0–1 cm, i.e. for particles almost in line with the fish.

The comparison between the Y_forward and X_lateral positions of the particles at T_R and the theoretical values of these positions based on the average experimental angular size, angular velocity and loom at each current speed is shown in Fig. 8. For the semi-transparent Artemia experiments, the theoretical values of Y_forward and X_lateral at T_R based on the average angular velocity are closer to the experimental values than those based on the average loom or angular size for the fast current speed (ANOVA; P<0.01; Tukey post test, P<0.01 and P<0.05 for the comparisons between loom and ω_R, andβ _R and ω_R, respectively), although differences were not significant for the intermediate and slow speeds(Fig. 8). The theoretical values of Y_forward and X_lateral based on the average loom for intermediate and fast speeds are closer to the experimental values than those based on angular velocity (intermediate and fast; ANOVA; P<0.05; Tukey Post-test, P<0.05 in both cases), although there were no significant differences when values based on averages β_R and ω_R were compared. Differences were not significant in any case for the slow speed(Fig. 8).

Planktivorous fishes typically feed on transient zooplankton delivered by water currents (Stevenson,1972; Hobson and Chess, 1976, 1978; Thresher, 1983; Hobson, 1991). When a plankter is transported by the current towards a fish's visual field, angular size,loom and angular velocity change over time(McFarland and Levine, 2002). These parameters could act independently or together as attentive cues to alert a fish to the presence of particles and to trigger a sensory–motor reaction (O'Brien and Slade,1976; Dunbrack and Dill,1984; Dill et al.,1997; Raviv,2000).

The geometry of the particle position and, in particular, its lateral distance, at which the feeding reaction occurs can give us some insights on the main visual mechanism involved. A relatively distant food particle travelling towards the fish at a given lateral distance(X_lateral) from the nasal–caudal axis has a relatively low angular velocity, which is increasing very slowly. As the particle approaches, a sharp increase in angular velocity takes place, and it reaches a peak (Fig. 4)whenY_forward=0 and subsequently decreases as the particle travels along its rectilinear path. By contrast, a straight-on particle(X_lateral=0) has zero angular velocity throughout its approach towards the fish. Conversely, loom and angular size are maximised when lateral distance is minimal. This scenario suggests that preference for particles coming at a lateral distance X_lateral=0 implies angular size or loom as the mechanisms triggering the feeding reaction. Conversely, if particles coming at small lateral distance are not attacked,then angular velocity may be the mechanism involved in the feeding reaction.

It is reasonable to suppose that fish would capture most particles at low X_lateral values for minimizing energy expenditure, if detection mechanisms were not limiting factors. However, in spite of possible energetic advantages of capturing prey in line with fish's nasal–caudal axis, very few particles were taken in when semi-transparent Artemiatravelled at small X_lateral values (i.e. <1 cm; Fig. 7). In fact, most particles produced a reaction when they approached off-axis with peak frequency values of X_lateral at 2–3 cm, and with values of X_lateral as high as 6 cm(Fig. 7). These results suggested that a threshold angular velocity ω_R might alert a fish to approaching semi-transparent Artemia, and promote the feeding reaction to food at greater reaction distances(Fig. 5).

By contrast, with darkened Artemia, the peak response frequency occurred when prey approached closest to a fish's nasal–caudal axis. The larger reaction to darkened prey at low X_lateral(Fig. 7), suggested that loom or angular size rather than angular velocity might have triggered a fish's reaction to darkened prey (Fig. 8). However, further explanations, are needed to account for the decline in frequency at X_lateral values >3 cm in semi-transparent Artemia (Fig. 7). This decline may be related to the energetic cost of feeding at larger X_lateral and to the fact that peak values ofω _R decline as X_lateral increases(Fig. 4C).

The effect of current speed on each angular variable suggests that angular velocity is the mechanism triggering the response to semi-transparent Artemia, while loom is the mechanism triggering the reaction to feed on darkened plankton. This is because angular velocity for the semi-transparent Artemia and loom for the darkened Artemiawere the only variables that both (1) increased with current speed as expected if a relatively constant latency and a constant angular threshold are assumed(2) allowed calculation of a reasonable latency. These results are in line with our comparison between the experimental Y_forward and X_lateral and the expected Y_forward and X_lateral based on averages angular size, angular velocity or loom. Such a comparison suggests that angular velocity may be the variable that triggers the feeding reaction to semi-transparent Artemia,although differences are significant only for high current speed. Conversely,the fitting of the experimental data with the model for the darkened Artemia treatment is not significantly different between angular size and loom. This is because the shapes of the curve for angular size and loom are very similar (Fig. 8). However, as discussed above (Materials and methods section), if angular size was the mechanism triggering the feeding reaction, one would expect the position of the particle at the apparent reaction time(T_R) to be different for different current speeds. Let us consider a realistic example in which the sensory threshold occurs at a given angular size corresponding to a distance of 10 cm at T_L. Let us consider a given latency of 200 ms. In this case, at T_R, high-speed particles should be closer to the fish, at approximately 5 cm distance from the fish vs 9 cm distance for slow speed particles. This assumes the same time interval (200 ms) between when the sensory threshold is reached (i.e. T_L) and the motor response (i.e. T_R) in the fish at all current speeds. Conversely, if a loom mechanism is considered, higher current speed would cause both (1) higher rate of change of the looming angle and (2) a longer distance travelled between T_L and T_R. Higher loom in high speed would cause reaching the sensory threshold from a larger distance. As a realistic example, let us consider that sensory threshold (i.e. at T_L) of loom may be reached at 10 cm for high speed particles vs 6 cm for slow speed particles. However, this difference may be in large part cancelled out, at least in our experimental conditions, by the longer distance travelled by the high speed particles. Following our example, during a given latency period of 200 ms, higher speed would allow particles to travel approximately 5 cm vs 1 cm travelled at slow speed. This would cause slow and fast particles to be at a similar distance (i.e. 5 cm according to our example) from the fish at the time of the fish's response. Therefore, loom, but not angular size, can explain particles travelling at different speeds to be in similar positions(Fig. 8D–F) relative to the fish at the time of reaction (T_R). We conclude that loom is the most likely variable involved in triggering the feeding reaction of fish preying on darkened Artemia.

The reaction models based on average angular velocity ω_Ror loom λ_R fit better at high, rather than at low, current speed for darkened and semi-transparent Artemia, respectively(Fig. 8). An explanation could be that variability in the time for reaction may decrease with current speed(i.e. when a fish needs to minimize latency to successfully capture the particle). Previous work on escape response in zebra danio (Brachidanio rerio) showed that predator speed had no effect on the apparent loom threshold, while it had an effect on the reaction distance and, therefore,latencies were considered negligible (Dill,1974). However, latencies of escape responses are likely to be shorter than those of feeding reactions, as escape responses are triggered by the Mauthner cells, a pair of giant neurons in the hindbrain of most fish species (Eaton and Hackett, 1984). Batty(1989) reported escape response latency to visual stimuli to be about 40 ms. The reaction to an approaching prey (e.g. plankton), may have a longer latency of the order of 100–300 ms, because more time is needed to process a prey's movement and trajectory,and then to direct an attack. This range corresponds approximately at the values found for maximizing I_H (i.e. assuming that fish respond to a given value of angular velocity or loom at all current speeds).

At higher current speeds (e.g. 40 cm s^–1) than those used here, C. aggregata swam in a flume but did not attempt to capture plankton (McFarland and Levin,2002). This may have been due to sensory–motor constraints,which influenced the fish's reaction time preventing successful capture when plankton approached at high speeds. Furthermore, feeding may become energetically costly at very high current speeds, even if fish may be capable of capturing the prey.

Transparency in the aquatic environment represents a significant adaptive cryptic mechanism (Johnsen 2001a,b, 2002). However, visual adaptations, such as UV and polarization vision, have evolved among planktivores to increase apparent visual contrast(Lythgoe and Hemmings, 1967; Lythgoe, 1979; Loew et al., 1993; McFarland and Loew, 1994; Cronin et al., 1994; Shashar et al., 1998). Foraging success in planktivores is also strongly dependent on the light,turbidity, background space light, and visibility of pigmented parts, such as the eyes or the gut of the prey (Vinyard,1980; McFall-Ngai,1990; Thetmeyer and Kils,1995; Johnsen and Widder,1998; Tsuda et al.,1998; Utne-Palm,1999; Utne-Palm and Stiansen,2002). Under the same background light conditions, fish predation was shown to be higher when transparent prey with larger eyes were used vs prey with small eyes (Zaret 1972), and reaction distances for transparent prey were significantly lower than those for pigmented prey(Thetmeyer and Kils, 1995; Tsuda et al., 1998; Utne-Palm 1999). However, it was shown by video imaging in the near field that transparent zooplankton irradiated from above when viewed horizontally against a darker background appeared as bright targets due to light scattering off their bodies(Loew and McFarland, 2000). Some planktivorous fish search for prey outside Snell's window (i.e. a cone angle of about 97.2° through which the terrestrial hemisphere can be seen), because prey are seen against a dark background(Lythgoe, 1979).

In our study, feeding reaction to zooplankton appeared to occur at longer distances for semi-transparent Artemia than for a darkened one. This result would be in line with Loew and McFarland's(2000) observations, and is in agreement with the general observation that changes in prey contrast may cause differences in reaction distances, although most observations are on pigmented prey against a light background (e.g. Utne-Palm, 1999). In our study, the dark background of the collimator tubes would enhance the visibility of semi-transparent plankton; therefore, semi-transparent Artemia should be visible at a greater distance due to their higher light scattering (diffuse reflectance) against the dark background. Because light was weakly reflected from a dark Artemia's body there would be less contrast between the target and the background in darkened Artemia (Fig. 2 and E. R. Loew and W.N.McF., unpublished).

The higher visibility of the natural (semi-transparent) plankton at longer distances, however, does not explain why shiner perch did not react to semi-transparent particles coming straight on. At small X_lateral, fish could theoretically react to a threshold angular velocity, although such a threshold would be reached at very small values of Y_forward(Fig. 4) (i.e. when it may be too late for the fish to react in time to catch the prey). Alternatively, fish could have used loom as a mechanism triggering their feeding reaction. A threshold loom would be reached at relatively high values of Y_forward for the case of small X_lateral (Fig. 4). If we assume that the apparent size of semi-transparent and darkened particles was the same, then a Y_forward of about 6 cm should trigger a reaction when X_lateral=0, as in the semi-transparent Artemia at intermediate speed. Because relatively few fish reacted to natural plankton at small X_lateralvalues, it is possible that the apparent size of the darkened plankton was greater than that of the natural plankton, although there we have no evidence that darkening can cause a change in apparent size. If semi-transparent Artemia had a relatively small apparent size, for any value of X_lateral and Y_forward, loom for a darkened Artemia would be greater than for a semi-transparent Artemia. In this case, it would be possible that when using semi-transparent Artemia, a threshold loom may have occurred when the prey was too close to the fish to react in time for prey capture. Alternatively, it may be possible that shiner perch do not have enough behavioural flexibility in order to shift from a reaction based on angular velocity mechanism to one based on loom within a single feeding session, and it may be `locked' in using the most favourable mechanism of detection for any given contrast conditions. However, we are not aware of any supporting evidence from other sources, and therefore this suggestion remains highly speculative.

Under the conditions of our experiments, we may be dealing with both positive and negative contrast situations. For the natural prey irradiated from above viewed against the darker background there may be positive contrast due to internal scatter. However, for the darkened prey the contrast may be negative. Since on- and off-channels use separate pathways that may differ in properties, it is possible that some of the observed differences in feeding reaction mechanisms may be due to the way the visual system handles positive and negative contrast (Wheeler,1982).

We have examined a fish's response to food in a flume tank where turbulence in the water flow was minimized, and where the detection of plankton was constrained by the dimensions of the flume. Therefore, this study was simplified compared with foraging in natural water currents. In nature,turbulence (Aksnes and Giske,1993; Kiorboe and Saiz,1995; McFarland and Levin,2002), as well as schooling can affect foraging behaviour in planktivores (Ryer and Olla,1991; O'Driscoll,1998; Lachlan et al.,1998; Foster et al.,2001). Nevertheless, individual responses in a flume to approaching zooplankton allowed us to standardize feeding reaction parameters and to define a fish's reaction to plankton.

In our study, fish reaction to approaching zooplankton was assumed to be independent of the particle shape or its intrinsic motion. In order to standardize prey size and behaviour, we used only Artemia nauplii. Artemia is a non-evasive prey since no significant escape motion has been recorded (Coughlin and Strickler,1990) and, generally, planktivores preferentially select non-evasive prey items, even when the evasive prey are larger(Drenner et al., 1978; Vinyard 1980, 1982). Although Artemia is not a natural prey of shiner perch, perch showed similar behavioural patterns when feeding on natural plankton(McFarland and Levin, 2002). In addition, by staining Artemia with black ink, we provide an artificial set-up that may have affected the behaviour of Artemia or its physiology (e.g. any chemical cues it may provide). However, as suggested by its lack of significant evasive behaviour(Coughlin and Strickler, 1990), Artemia was most likely just passively carried by the current and therefore its behaviour should not have affected our results.

The dark background we used does not correspond to the most common environment where plankton is found. However, certain environmental conditions, i.e. twilight, as well as in certain conditions of turbidity,cloudiness and depth, may also provide a relatively dark background. Nevertheless, conditions of visibility created in laboratory set-ups such as ours are quite different from natural conditions, and it would be interesting to investigate the feeding reactions of planktivorous fishes to a variety of plankton types in natural conditions.

Our results showed that at high current speeds, angular velocity was the likely cue for triggering the feeding reaction when semi-transparent Artemia were used. Loom appeared to be the main mechanism when prey were darkened. It is possible that both mechanisms act in concert during some of the feeding phases (detection, pursuit, capture), although one mechanism might eventually prevail in detecting certain prey characteristics (e.g. motion type, prey contrast).

As our major conclusion, we suggest that prey contrast dramatically affects detection and the subsequent feeding reaction. When prey contrast was high due to scattering (semi-transparent prey), feeding reactions occurred at greater distances and at increasing X_lateral values where angular velocity tends to be higher. When prey contrast was low (darkened treatment),reaction distances diminished and feeding reactions occurred for prey approaching at smaller lateral distance (X_lateral). This study showed that visual behaviour of planktivores is highly flexible because fish appear to be able to utilize different visual properties of the approaching object for their reaction to prey. Although angular velocity was the main cue for triggering feeding reactions to semi-transparent plankton at high current speeds, it is still possible that during the tracking of a plankter after a prey's detection, loom may have a role in determining the time of engulfment. This and other issues will be explored in a further paper dealing with the pursuit and capture of plankton.

We thank Friday Harbor Laboratory, its director Dennis Willows and all the staff, for the space and assistance provided. We thank also Simon Levin,Samuel Ramsden, Lyle Britt, Mike Burrows, Bruno Pernet and Ellis Loew for their valuable advice. The useful comments of two anonymous referees are greatly acknowledged. This paper is dedicated to Mac.