Learned Discrimination of Pattern Orientation in Walking Flies

Blowflies are highly acrobatic insects endowed with complex visual systems. The energy needs of such acrobatics are fueled by frequent visits to flowers, tree sap and the occasional picnic. The visual processes necessary for recognition of shapes have not been well studied in Diptera; however, data from a range of insects and the plants that they pollinate suggest that several floral features, including contour orientation, shape, pattern and symmetry, are critical during foraging (Lehrer et al., 1995; Giurfa et al., 1996; Dafni et al., 1997; Kelber, 1997).

The ability to learn a visual pattern and discriminate it from another is thought to be linked to natural behaviors such as landmark orientation and flower recognition. Colonial insects of the order Hymenoptera, that recognize routes between the nest and food sources, have long been considered unique among the insects with regard to their pattern discrimination capabilities. Despite this notion, a number of non-hymenopteran insects can perform pattern discrimination tasks, including cockroaches (Mizunami et al., 1998), fruitflies (Dill et al., 1993), butterflies (Allard and Papaj, 1996) and locusts (Wallace, 1958). In fact, a diversity of learning capacities exist within the Hymenoptera, and these abilities may vary in a species-specific manner (Wäckers and Lewis, 1999). Flies are also excellent learners; they can discriminate both shapes (Dill et al., 1993) and colors (Fukushi, 1985; Fukushi, 1989; Fukushi, 1990; Fukushi, 1994; Troje, 1993), suggesting that Diptera indeed have complex visual behavior. The complexity of fly visual systems is also clear from anatomical (e.g. Strausfeld, 1970; Strausfeld, 1976; Strausfeld and Lee, 1991) and physiological (e.g. Egelhaaf et al., 1988; Douglass and Strausfeld, 1996; Krapp and Hengstenberg, 1996) investigations.

Much progress has been made on revealing the relative contributions of specific visual features to the perception of form as demonstrated in learning studies conducted on hymenopteran insects. Behavioral studies of pattern discrimination have revealed that bees can discriminate the orientation of patterns presented in the vertical plane (for reviews, see Wehner, 1981; Srinivasan, 1994). Recent studies on discrimination of pattern orientation have expanded the view of hymenopteran form vision to include the mechanisms of abstracting particular aspects of a pattern, such as orientation (Srinivasan et al., 1993; Giger and Srinivasan, 1995), and generalizing to novel patterns that contain the same orientation information (van Hateren et al., 1990).

Like vertebrates (pigeons, Jitsumori and Onkubo, 1996; fish, Volkmann, 1975; cats, Vanduffel, 1997; primates, Orban and Vogels, 1998), insects are capable of discriminating differences in spatial structure in their visual environment, albeit with a lower resolution. The ability to make discriminations of fine spatial structure may be a significant adaptation for most visual organisms.

The aim of the present study was to determine whether blowflies, Phaenicia sericata, can make discriminations of pattern orientation similar to those demonstrated previously for honeybees. In addition, the conditions under which blowflies will preferentially and repeatedly search for food rewards were developed. The innate searching behavior, which can be elicited in food-deprived flies (Dethier, 1957; Dethier, 1976), served as the means to bias the flies’ behavior such that they searched for food during the training and testing phases of the experiments. Flies do not eat on the wing. Instead, they land on surfaces of interest and explore the area using tarsal contact receptors that are sensitive to water, salt and sucrose (Minnich, 1929; Dethier, 1954). After locating a drop of food, a fly will explore the occurrence of additional food through tarsal contact. This is true whether a fly is feeding on a flower, tree branch or picnic table. Thus, in natural conditions, flies locate food while walking.

This paradigm allows comparisons between the discriminatory capabilities of blowflies and those of the better studied honeybees. The degree to which flies are capable of visual discrimination and the significance of this with respect to the underlying neural arrangements are considered here.

Experiments were performed using Phaenicia sericata (Meigen) collected locally and maintained as a laboratory colony in the Arizona Research Laboratories Division of Neurobiology. Flies were reared under a 13 h:11 h light:dark cycle.

The behavioral assay included a learning period in which individual flies were trained to associate one visual pattern with a food reward in an arena in which the animals were free to walk. The assay was a modification of techniques used by Fukushi (Fukushi, 1985). Searching behavior, characterized by an intense search for food after a food-deprived fly had imbibed a small drop of sucrose solution (Dethier, 1957; Dethier, 1976), was elicited in food-deprived animals to bias their motivational state and to ensure that all animals were in a similar motivational state at the time of training. The searching behavior led to the discovery of a second sucrose drop that was paired with the positive (rewarded) visual cue. Repeated training resulted in an association between the positive visual cue and the sucrose reward. The animal’s ability to discriminate the positive visual cue from the negative (unrewarded) cue was subsequently tested in the absence of the paired sucrose reward.

Adult flies, male and female, were collected following emergence from the pupal case and anesthetized on ice. The animals’ wings were then clipped to prevent flight during experiments, and their thoraces and/or abdomens were labeled (colored Liquid Paper, The Gillette Company, Boston, MA, USA) so that animals were individually identifiable. Prior to control trials, animals were maintained collectively in a ‘holding arena’ consisting of a round, white bucket (diameter 28 cm and 37 cm tall) in which were food (dry sucrose and powdered milk), water and many colorful, textured, contrasting patterns and objects to provide a visually rich environment.

All control, training and testing trials were performed in a behavioral arena. The arena was illuminated from above with a 100 W light bulb. For experiment 1, flies were trained in a white cardboard Y-maze of the following dimensions: each arm of the Y-maze was 7 cm long by 6 cm wide, and the base of the Y-maze was 2 cm long by 6 cm wide; all sides of the Y-maze were 7 cm tall. As a consequence of the Y-maze design, flies were forced to make a choice between the two visual cues at the base of the ‘Y’. Thus, analysis of data collected in experiment 1 was limited to the first visit per trial (see section on scoring and statistical tests). This, however, was validated as a sufficient means of analyzing animal choice by comparison with the first visit data and response proportion data in experiment 2. Experiments 2 and 3 were conducted in another white, round arena (28 cm in diameter and 20.5 cm tall; see Fig. 1). A glass plate was placed in the bottom of the arena and was cleaned with distilled water between all trials.

Prior to all training and testing with a given set of visual cues, animals were tested for innate preferences on the third day of adult life. Each control trial involved the release of an individual fly into the base of the Y-maze (experiment 1) or into the center of the arena (experiments 2 and 3) from a glass vial that was angled up on its side by a piece of wax and that pointed between the two visual patterns located on the opposite arena wall and the observation of which visual pattern the fly first visited during the 210 s (3.5 min) of the control trial. A visit was defined as any time a fly walked within 2 cm of a visual cue. In situations in which an individual did not visit each of the two visual cues exactly 50 % of the time during control trials, the fly was subsequently trained to the cue it had visited less frequently.

After control trials, flies were returned to the holding arena for 2–4 days. Throughout this period, a water source was available in the holding arena, but the animals were deprived of food. Food deprivation was necessary to trigger the flies to imbibe the initial drop of sucrose which subsequently biased their motivational state towards searching for additional food. Biasing the animals in this way ensured that they would locate the sucrose drop that was paired with the positive cue. In addition, the initial drop of sucrose was used to elicit searching behavior in the unrewarded testing trials, in which the number of visits to each visual cue was counted.

The period of starvation that an individual fly experienced depended on how much fat body it had stored as a larva and how much food it had ingested during the first 3 days of its adult life. For this reason, individual flies were tested for searching behavior (see Results for description) throughout the second to fourth days of the starvation period. A fly that did not initiate searching behavior upon release over a sucrose drop was returned to the holding arena and tested later, after further food deprivation, for the presence of searching behavior.

Animals were trained and tested on the same day. Individual animals were trained and tested separately in a Y-maze or an arena in which two visual cues were positioned on adjacent regions of the wall (Fig. 1).

The fly was released from an inverted vial at the base of the Y-maze or in the center of the arena over a 0.5 l drop of 0.1 mol l⁻¹ sucrose solution. A second 0.5 µl drop of 0.1 mol l⁻¹ sucrose, the reward, was placed 1 cm in front of the positive visual cue. No food stimulus was paired with the unrewarded, negative visual cue. The inverted vial was removed while the animal imbibed the first sucrose drop. The experiments presented here confirm that after imbibing the first drop of sucrose the animal began to search in increasing radial loops (Dethier, 1957; Fukushi, 1985) and eventually located and imbibed the sucrose reward that was associated with the positive cue. Prior to and after discovering the reward, the fly was free to explore all regions of the arena. At 180 s (3.0 min) after imbibing the sucrose reward, the animal was removed from the training arena. Between trials, the floor was cleaned with distilled water to remove any chemical cues that the fly might have released during the previous trial. The arena was also rotated by 60–120 ° to reposition the location of the sucrose within external coordinates, and the patterns were exchanged randomly (using a randomization schedule generated with a table of random numbers) from right to left and positioned 6–14 cm apart from each other. The arena was surrounded by white paper or cloth walls at all times. These precautions were taken to ensure that the positive visual cue was the only cue that reliably indicated the location of the sucrose reward. Fresh sucrose drops were then placed in the arena and the same fly was trained again, within 5–10 min of the previous training trial. This procedure was repeated until the animal exhibited the testing criteria (see below for definition for each experiment).

The training method was optimized so that other factors, such as satiation, did not confound the searching frequency over training time (see Results). The volume of sucrose per drop and the concentration of the sucrose were chosen such that flies did not become satiated during the training period, which usually required between 5 and 10 trials per animal.

Testing trials were the same as training trials except that the sucrose reward paired with the positive visual cue was not present. The animal was released over the first sucrose drop in the center of the arena and its behavior videotaped from above. Each fly was tested once.

All training and testing trials were filmed with a digital camera (Panasonic) and recorded with a video recorder. The walking paths of training and testing trials were digitized every 67 ms and analyzed with a motion analysis system (Peak Performance Technologies, Inc., Englewood, CO, USA). A visit was defined as any excursion to within 2 cm of a visual cue. The first visit of any given trial was scored as either to the positive or to the negative visual cue. For experiments 2 and 3, the total number of visits to each cue during a given trial was also scored and used to calculate a response proportion (RP). RPs for each training and testing trial were calculated using: RP=(v_p)/(v_p+v_n), where v_p represents the number of visits to the positive rewarded cue and v_n represents the number of visits to the negative cue. An RP of 1.0 represents a 100 % response to the positive cue, and an RP of 0 represents a 100 % response to the negative cue. Given that individual flies were trained to the cue that they visited less frequently during controls, the positive cue was defined as either of the two cues.

RPs were calculated only for training trials in which the animal successfully located the sucrose reward. These RPs were averaged across all training trials for each individual, and individual averages then contributed to the group average that was used for the statistical analysis of training trials. An RP was also calculated for each animal’s single testing trial in experiments 2 and 3. The averaged RP of a group of animals during either training or testing was analyzed for a significant departure from random choice using the Wilcoxon signed-ranks paired test (Zar, 1996). Chance was defined as RP=0.5 for this two-choice paradigm. The averaged RP of a group of animals during testing was analyzed for significant departure from the averaged control RP using the Wilcoxon signed-ranks paired test.

For control trials, a two-tailed test was used, and averages deemed significantly different from each other were interpreted as a preference for the visual cue with the greater average. For training and testing trials, a one-tailed test was used, and averages deemed greater than chance (0.5) or greater than the averaged control RP for that experiment were interpreted as preferences for the visual cue in question. In all experiments, individual flies were not necessarily trained the same number of times (see individual experiments below for explanation of testing criteria) and the number of visits made per trial varied; therefore, the number of visits per fly is not simply the total number of visits divided by the number of flies.

The non-parametric test (Wilcoxon signed-ranks paired test) was chosen for its robustness (i.e. fewer assumptions regarding normality and homogeneity of variance; Martin and Bateson, 1993). The ratio method of calculating a response proportion for each animal was used to reduce the effects of individual variation. The binomial test was applied to the first visit data of experiment 2 testing because the data were measured on a nominal scale and could not therefore be tested with the paired Wilcoxon test.

Each animal was tested 5–14 times in the control trials of experiment 1. Experiment 1 consisted of 12 flies making 137 visits during the control trials and five flies making 34 visits during training trials. The testing criteria were achieved when the fly visited the positive cue first during at least two training trials. While this was true for all animals trained, often an individual fly was trained more times, resulting in up to eight training trials prior to the unrewarded testing trial. Only the first visit of each trial was scored in experiment 1. Visual cues for experiment 1 were 5 cm high by 6 cm wide and consisted of a square-wave striped pattern with a period of 1.3 cm (individual stripes were 0.65 cm wide). From the release location, the entire cue subtended an angle of 32 ° vertically and 37 ° horizontally on the eye of the fly, corresponding to 17.7 % of the 180 ° of vertical visual field and 10.3 % of the 360 ° of horizontal visual field. Individual stripes could be resolved by the fly from the release location and each subtended an angle of 4.65 °.

Each animal was tested eight times in the control trials of experiment 2. Experiment 2 consisted of 25 flies making 200 visits (eight each) during controls and seven flies making a total of 331 visits during training trials. The same seven flies were each tested once in a testing trial; the number of visits during these tests totalled 56 for all seven flies. The testing criteria were achieved when the fly visited the positive cue first during at least two training trials. While this was true for all animals trained, often an individual fly was trained more times, resulting in up to eight training trials prior to the unrewarded testing trial. Each testing trial lasted 5 min, and the number of visits to each visual cue was counted during this test. Visual cues for experiment 2 were 14.1 cm high by 10.5 cm wide and consisted of stripes with a range of widths (0.3, 0.5, 0.8 and 0.9 cm). From the release location in the center of the arena, the entire cue subtended an angle of 45 ° vertically and 37 ° horizontally. Visual cues were of comparable size in all experiments with the exception that in experiment 1 the height of the cue was 13 ° less. Individual stripes of widths 0.5, 0.8 and 0.9 cm, subtending angles on the fly’s eye of 2.05 °, 3.27 ° and 3.68 °, respectively, could be resolved by the fly from the release location. Stripes of only 0.3 cm wide could be resolved by the fly at the release location depending upon which ommatidia viewed the stripe. Where facet size is largest (at the upper front of the eye), interommatidial angles are smallest and, therefore, resolution is highest (Land and Eckert, 1985). At the release location, a 0.3 cm stripe subtends an angle of 1.23 °, which is at the theoretical limit of resolution. This stripe width, however, is clearly resolvable by the fly at closer viewing distances where it subtends larger angles. Each visual cue contained all four stripe widths.

Each animal was tested 10 times in the control trials of experiment 3. Experiment 3 consisted of 21 flies making 210 visits (ten each) during controls and seven flies making a total of 494 visits during training trials. The same seven flies were each tested once in a testing trial; the number of visits during these tests totalled 86. For experiment 3, the testing criteria were modified; each fly was trained a minimum of five times and was not tested until the first visit was to the positive cue in two consecutive training trials. Each testing trial lasted 5 min, and the number of visits to each visual cue was counted during this test. The stripe widths and dimensions of the visual cues were the same as those used in experiment 2.

Two factors account for the lower number of flies in the training/testing trials compared with the number of flies in the control trials. First, control trials were performed on a single day and required only 4–5 h to complete a set of 10 flies. Training/testing was also performed on a single day for an individual fly. Two to four days into the starvation period, all flies were tested for the presence of searching behavior. As mentioned above, the presence of searching behavior was not predictable and could occur any time 2–6 days after the control trials. Each fly required between 3–6 h to train and test individually. Thus, a maximum of three flies could be trained per day. As a consequence, untrained flies would die of starvation overnight. Therefore, more flies were observed in control experiments than in training and testing trials. Second, some flies did not exhibit searching behavior and eventually died of starvation.

Results are presented as means ± standard deviation (S.D.) unless stated otherwise.

During training and testing trials, food-deprived flies carried out searching behavior (Dethier, 1957). An example of the looping pathways characteristic of this behavior is shown in Fig. 2A. Searching began in the center of the arena after the fly had imbibed the initial drop of sucrose, and it continued throughout the trial. In training trials, looping behavior increased after the discovery of the reward sucrose drop (arrow in Fig. 2A). Similar searching behavior was exhibited in an unrewarded testing trial (Fig. 2B). The fly began searching in the center of the arena, walked towards the positive visual cue, and continued searching in the vicinity of this despite the absence of a reward. Thus, having learned the association between the positive cue and the reward, the fly demonstrated a learned bias towards this cue, confirming its ability to discriminate the two visual cues in question.

Experiment 1 demonstrated the ability of flies to discriminate between horizontal and vertical gratings. This was evident both from the population’s innate preference for horizontal gratings in the presence of vertical gratings (Fig. 3A) and from the ability of individual flies to learn a preference for the vertical grating when it was rewarded as the positive cue in training trials (Fig. 3B). During the control experiment, the average percentage of visits to the horizontal grating (68±18 %) was significantly different from the average percentage of visits to the vertical grating (32±18 %; Wilcoxon signed-ranks paired test, two-tailed, P<0.01, N=12, t=6.5; Zar, 1996). During training trials, flies showed a learned preference for the vertical grating in the presence of the horizontal grating; the average percentage of visits to the vertical grating (76±16 %) was significantly higher than to the horizontal grating (24±16 %; Wilcoxon signed-ranks paired test, one-tailed, P<0.05, N=5, t=0). An example of a single testing trial is shown in Fig. 3C, in which a fly was tested in the absence of the reward that had previously been paired with the vertical grating. The fly began searching at the release location over the initial drop of sucrose, walked towards the visual cues, and chose the vertical pattern over the horizontal one. Thus, flies discriminated between horizontal and vertical gratings during both control and training trials.

To investigate further the ability of P. sericata to discriminate between patterns of different orientation, flies were tested to see whether they could learn to prefer one of two oblique gratings (experiment 2). These two patterns were at +45 ° and −45 ° defined relative to a 0 ° vertical. In control trials, the flies did not demonstrate a significant bias for either of the two cues (Fig. 4A); the average percentages for +45 ° and −45 ° (42±19 % and 58±19 %, respectively) were not significantly different (Wilcoxon signed-ranks paired test, two-tailed, 0.10<P<0.20, N=25, t=105.5). In subsequent training and testing trials, each fly was trained to the visual cue it had visited less frequently in the control trials; for animals that visited +45 ° and −45 ° equally during controls, assignment of the positive cue in subsequent training trials alternated from fly to fly between +45 ° and −45 °. The flies demonstrated a learned preference for the positive cues; the average percentage of first visits to the positive cues during training (68±20 %) was significantly greater than chance (Fig. 4B left; Wilcoxon signed-ranks paired test, one-tailed, P<0.025, N=7, t=2). During testing, the flies preferentially visited the positive cue first (Fig. 4B right; binomial test, one-tailed, P<0.005, N=7; Zar, 1996; the probability of choosing the positive cue was assumed to be 0.32, the average percentage of first visits to the positive cue during control trials). A second measure of the learned preference was also consistent with these results; when the total number of visits during a single trial was counted and calculated as a response proportion, the average response proportions during training and testing (0.74±0.08 and 0.81±0.25, respectively) were each separately significantly greater than the chance level of 0.5 (Fig. 4C; Wilcoxon signed-ranks paired test, one-tailed, P<0.01, N=7, t=0 for training and P<0.025, N=7, t=2 for testing). In addition, the averaged response proportion during testing (0.81±0.25) was significantly greater than the average response proportion during control trials (0.32±0.17; Fig. 4D; Wilcoxon signed-ranks paired test, one-tailed, P<0.025, N=7, t=1).

Preferences for the positive cue were acquired early in the training process and were maintained throughout the training period. Fig. 5A depicts the averaged response proportion as a function of the training trial number for experiment 2. The learned preference was present as early as the second training trial. Variation in the averages reflects the trial-and-error process that is present as each fly investigates both the visual cues during the training. The individual patterns of learning were not identical, each fly having its own pattern of first visits. Over the course of the learning period, individual flies exhibited fewer visits to the negative cue, often searching exclusively at the positive cue in the unrewarded test (Figs 5B, 2B). In Fig. 5B one fly is shown during the testing trial in which it searched exclusively at the −45 ° visual cue that had previously been rewarded. Another fly (Fig. 2B) searched similarly at the +45 ° visual cue, which served as the positive cue in this case. The searching pattern throughout a testing trial was often not restricted exclusively to the positive cue: it was common for a fly to search initially at the positive cue for the first few minutes of the trial followed by occasional excursions to the negative cue later on in the trial. Fig. 6 depicts the temporal sequence of one such testing trial from experiment 3 (see below). This example demonstrates that the animal searches in the proximity of the visual cues for the entire 5 min of the testing trial.

Fig. 7 elaborates on the training patterns of three individual flies; the upper graphs depict the sequence of first visits, and the lower graphs depict the number of visits to the positive cue and the total number of visits to both visual cues for each trial. The first fly visited the positive cue first in the initial three trials (Fig. 7A upper graph). Correspondingly, all visits during trials 1–3 were primarily to the positive cue (Fig. 7A lower graph). In trials 4–6, however, the first visits were to the negative cue and the total number of visits increased, reflecting the fact that the fly now visited both visual cues. During the subsequent training trials (trials 7, 8) and the testing trial, all visits were exclusively to the positive cue, reflecting a learned preference for the positive cue. The second fly exhibited a similar pattern of exploration. Early trials were exclusively to the positive cue (Fig. 7B), whereas visits to both visual cues occurred during the middle trials (Fig. 7B lower graph). Later trials (trial 5 and test) reflected the learned bias to the positive cue; first visits and most subsequent visits were to the positive cue. Unlike the first two flies shown in Fig. 7, all the first visits of the third fly were to the positive cue (Fig. 7C upper graph). Each training trial, however, contained visits to both visual cues (Fig. 7C lower graph). Thus, this fly exhibited a different pattern of searching, suggesting individual variation in the searching patterns and possibly in the learning strategies used by flies.

The overall searching behavior of individual flies did not decrease as a function of training time. That is, the total number of visits per trial did not progressively decrease with training (Fig. 7 lower graphs). Thus, the pattern of visits did not reflect a change in motivational state with respect to satiation. Rather, the pattern of visits reflected an individual’s tendencies to search (factors such as turning rate, grooming frequency, etc.) and its learned preference for the positive cue.

Experiment 3 investigated the ability of flies to discriminate small differences in orientation. Visual stimuli consisted of two identical gratings, one vertically oriented at 0 ° and one oriented at +5 °. During control trials of experiment 3, flies did not exhibit an innate preference for either of the two visual patterns. The averaged percentages to the cues (55±19 % and 45±19 %) were not statistically different (Fig. 8A; Wilcoxon signed-ranks paired test, two-tailed, 0.10<P<0.20, N=21, t=71). During training and testing, however, flies exhibited a learned preference for the rewarded pattern, whether it was the vertical or +5 ° grating (Fig. 8B). Again, individual flies were trained to the pattern they had visited less frequently during the 10-trial control experiment. Both the training and testing RP averages (0.75±0.07 and 0.64±0.14, respectively) were individually significantly greater than chance (Wilcoxon signed-ranks paired test, one-tailed, P<0.01, N=7, t=0 for training and P<0.05, N=7, t=2.5 for testing). In addition, during testing, the average response proportion (0.64±0.14) was significantly greater than the averaged RP during control trials (0.36±0.14; Fig. 8C; Wilcoxon signed-ranks paired test, one-tailed, P<0.025, N=7, t=1).

In contrast to experiment 2, the average testing response proportion in experiment 3 was not high; whereas in experiment 2 it was 0.81, it was only 0.64 in experiment 3. This lower response, although significantly greater than chance, could be used to argue for an absence of learning in experiment 3. One possibility could be that flies do not learn to discriminate the positive cue per se, but that they use a generalized learning rule such as: alternate visits to the two visual cues until food is discovered. Such a behavior would generate an average response proportion near chance. The flies, however, clearly did not behave in this manner. The fly depicted in Fig. 6 spent the first 3 min of the 5 min trial searching exclusively at the positive cue. It did not alternate visits between the two cues as it might have done if the probability of discovering food was equal at either visual cue. In addition, when the performances of individual flies were compared in control and testing trials, it was clear that learning of the positive cue did indeed occur as a result of training (Table 1; Fig. 8C).

For the first time, learned discrimination of visual pattern orientation in a dipteran species has been clearly demonstrated. In the behavioral paradigm used here, flies could discriminate 90 ° differences in orientation such as horizontal versus vertical (Fig. 3A,B) and +45 ° and −45 ° gratings (Fig. 4B–D). The underlying mechanisms for such discriminations are not necessarily identical (see below). In addition, flies could discriminate a difference as small as 5 ° (Fig. 8B,C). Comparable results have been obtained in a variety of behavioral experiments performed on honeybees, except that honeybees can only discriminate orientation differences of 30 ° or more (Chandra et al., 1998). In a dual-choice experiment, the honeybees had difficulty discriminating deviations of less than 25 ° (Chandra et al., 1998) and would presumably not discriminate a difference of 5 ° as observed in the present study for blowflies.

One difference in the training paradigms that might account for these differences in behavior is that, when blowflies were trained, both the positive and negative visual patterns were present during both the training and the testing trials. In contrast, honeybees were presented with the training orientation and the unrewarded orientation together only during the testing phase (Chandra et al., 1998). The simultaneous presentation of both orientations during the learning phase may be critical for discrimination of smaller differences in orientation. Another distinction between the honeybee and fly studies that may affect the difference in their orientation-discrimination capabilities is that the flies walked rather than flew during the discrimination process. Self-induced changes in the orientation of the visual surround are minimized during walking, where rotation of the head is reduced compared with during flight. This reduction in noise could enable better orientation discrimination in walking flies compared with flying honeybees.

From anatomical studies, it appears that the optical requirements of a system capable of making such discriminations are met in flies of both sexes. In the region of highest resolution, blowflies have interommatidial angles ranging between 1.02 and 1.28 ° (Calliphora erythrocephala, males 1.07 °, females 1.28 °; Lucilia cuprina, males 1.02 °, females 1.13 °; Land and Eckert, 1985). The area of highest resolution in L. cuprina, a close relative of P. sericata, lies near the equator in the frontal visual field in both sexes (Land and Eckert, 1985). The small interommatidial angles of blowfly eyes, therefore, endow the fly with a visual resolution sufficiently high to resolve the differences present in the visual patterns used here. During the discrimination behavior, flies approached the positive pattern with the fronto-lateral eye regions leading (see Campbell, 2001). These eye regions encompass areas of high resolution as well as ommatidia with posteriorly decreasing interommatidial angles. The fact that flies approach the visual cue with the fronto-lateral eye regions leading may also be indicative of fixation behavior. This behavior can be characterized by either a straight or a curved path during which intermittent whole-body saccades are performed (Horn and Fischer, 1978; Osorio et al., 1990). As indicated in Fig. 6, flies often made whole-body saccades as they approached the visual cue. In flies, as in vertebrates, fixation behavior can serve to bring the object of interest into a region of high acuity.

While the ability to discriminate pattern orientation has been demonstrated for ants (Wehner et al., 1972) and honeybees (van Hateren et al., 1990; Srinivasan et al., 1993; Srinivasan et al., 1994; Giger and Srinivasan, 1995; Giger and Srinivasan, 1997a; Giger and Srinivasan, 1997b; Chandra et al., 1998), it has long been assumed that social hymenopterans are unique in having this ability, because of their status as foragers within a social colony. The demands of recognizing reliable nectar and pollen sources and returning to the hive have been argued as the adaptive basis for which honeybees would possess both a visual system capable of such orientation discrimination and well-developed learning mechanisms for repeated recognition and discrimination. Flies, as solitary organisms, have been assumed to rely less on an ability to learn spatial patterns, despite evidence that flies, including calliphorid blowflies, visit flowers, consume pollen and are likely to play a critical role in pollination of a variety of flowering plants (Kearns, 1992; Erhardt, 1993). The pattern with which individual flies forage and the contribution of the visual system to foraging behavior are, however, unknown.

In the control trials of experiment 1, flies showed a preference for horizontal over vertical gratings (Fig. 3A). The holding arena did not contain either horizontal or vertical gratings; the spontaneous bias in control trials was not, therefore, due to previous exposure to either pattern. This result was initially puzzling given that most other insect species studied have exhibited preferences for vertical over horizontal (Wehner, 1981). A possible reason why flies preferred horizontal to vertical gratings may be to do with the interaction between scototaxis (walking towards the darkest area of the visual field) and edge fixation (Osorio et al., 1990). Flies will spontaneously fixate an edge and walk towards it. When Drosophila melanogaster is presented with edges of varying heights and widths, fixation of the contrasting edge is modulated depending on the height and width of the stripe (Wehner, 1972). At wide widths and low heights, flies walk to the center of the horizontal stripe (presumably because of a scototaxic response) rather than fixate the edge (Wehner, 1972; Osorio et al., 1990). The visual cues in experiment 1, unlike those in experiments 2 and 3, subtended a vertical angle (32 °) of less than 40 °. Therefore, it is possible that the flies approached the attractive dark regions of the horizontal grating as part of a scototaxic response rather than the vertical edges of the vertical grating. This simple explanation would account for the spontaneous preference for horizontal over vertical. Learned preferences during subsequent training trials would, however, presumably be supported by more sophisticated visual processes, such as static orientation detection or directional motion cues (see below).

Spontaneous discrimination of pattern orientation in walking flies has been studied previously in flesh flies (Mimura, 1981; Mimura, 1987) and fruitflies (Mimura, 1982). These studies generally found a preference for star-shaped patterns over single bars at various orientations. Interestingly, Mimura (Mimura, 1981) found a spontaneous preference for a horizontal stripe over a vertical one. Partial covering of the eye revealed selective deficits in discrimination ability (Mimura, 1987); the anterior eye regions were deemed necessary for all but the detection of the horizontal stripe. Similar to the control trials in experiment 2 (Fig. 4A), Mimura (Mimura, 1981) did not find a spontaneous preference for either of two oblique stripes. A training paradigm such as that presented here is necessary for a complete dissection of visual discrimination capabilities in flies.

The learning and memory paradigm used here allowed clear discrimination of pattern orientation. Discrimination of pattern shape through template matching can be excluded from the possible mechanisms used by flies. Both the spatial structure of approaches to the visual cues and the range of body orientations used by individual flies demonstrated that P. sericata approaches the visual cue from a unique vantage point from trial to trial, so that template matching is not necessary for the recognition of pattern orientation (see Campbell, 2001). Rather, extraction of orientation information is likely to underlie the discriminations made by blowflies.

Several possible mechanisms could be employed in the discrimination of orientation. First, directional motion cues could signal differences in orientation. This is particularly evident in the case of horizontal versus vertical gratings. As a fly moves towards a horizontal grating, it will experience motion along the horizontal edges in the vertical directions, up and down (Fig. 9A; after Srinivasan, 1994). Conversely, the majority of motion cues generated as a fly walks towards a vertical grating will be in the horizontal plane, to the left and right. Neurons tuned to these cardinal directions of motion have been described in the optic lobes of flies (e.g. Eckert, 1980; Hausen, 1982; Hengstenberg, 1982; Douglass and Strausfeld, 1995). Thus, neurons capable of detecting the different directional motion cues generated as the fly moves towards vertical and horizontal gratings and, therefore, capable of discriminating between the two stimuli are present within the optic lobes. Directional motion cues, however, are less likely to underlie the discrimination of oblique orientations such as +45 ° and −45 °; as a fly walks directly towards either of these two visual cues, motion in all four cardinal directions (up, down, left and right) will be generated (Fig. 9B; after Srinivasan, 1994). A +45 ° bar will generate moving edges in the up-left and down-right directions, and a −45 ° bar will generate them in the up-right and down-left directions. The motion generated by each edge, however, consists of motion of equal magnitude in the four cardinal directions. The ability to discriminate the two orientations using directional-motion cues would, therefore, require complex connectivity among neurons that integrate the two directions with edge detection and assignment. In addition, the visual cues used in these experiments consisted not of only one oblique edge, but eight edges. Thus, a complex flow field was experienced during approach, consisting of many edges expanding across the retina. Directional motion cues are also not likely to underlie discrimination between +5 ° and 0 ° vertical; each cue would generate predominantly horizontally moving edges.

A second possible mechanism underlying orientation discrimination is the overall composition of the flow field. When a fly turns right in front of a grating, it will experience motion to the left originating from the vertical edges of the pattern. In the case of horizontal and vertical gratings (experiment 1), the amount of vertical edge providing motion is less for the horizontal grating than for the vertical grating. The direction, however, will be identical. The flow field generated by the vertical grating would be evenly distributed, whereas that of the horizontal grating would be broken into lines of motion (Fig. 9C). Theoretically, such information could be utilized to discriminate horizontal from vertical. This mechanism predicts that individual animals would consistently turn right or left when approaching the visual cues; the flies, however, do not perform stereotyped turns from trial to trial. Alternatively, knowledge of the direction of turning, coupled with the visual flow field experience, would allow for such a mechanism to underlie discrimination. This mechanism, however, is unlikely to be used to discriminate +45 ° from −45 ° and +5 ° from vertical gratings. In these cases, the amount of edge is identical, and the composition of the flow field over the retina will therefore be identical (evenly distributed) as the fly turns in one direction regardless of which of the two visual cues it faces.

In the case of the +45 ° and −45 ° gratings, the shape of the retinal image of the flow field projected on the eye as the fly turns in one direction would depend on the orientation (Fig. 9D). This feature could be detected by the central nervous system and underlie the discrimination. Neuronal correlates of this mechanism would require orientation-selective neurons that are motion-sensitive, but with motion-sensitivity restricted to the horizontal plane.

A fourth possibility is that a network of orientation-selective neurons computes subtle differences in orientation (Fig. 9E; after Srinivasan et al., 1994). Such a network has been proposed by Srinivasan and his colleagues to account for orientation-discrimination behavior in honeybees (Srinivasan et al., 1994; Chandra et al., 1998). Their model proposes a network of several orientation-sensitive channels each tuned to a range of unique orientations, separated by 120 ° with overlapping tuning curves (Srinivasan et al., 1994; Chandra et al., 1998). Behavioral experiments in which honeybees could not discriminate between two right-angled crosses rotated at different orientations suggested that the tuning curves of these hypothetical orientation-sensitive channels have a half-width of 90 ° (Srinivasan et al., 1994). The addition of a third channel to their model removed the ambiguity of discriminating between two orientations orthogonal to each other and, thus, a minimum of three channels is required (Chandra et al., 1998). This model elegantly explains honeybee orientation discrimination behavior and correlates well with intracellular recordings of optic lobe neurons in honeybees (Yang and Maddess, 1997). Similar neurons with large receptive fields and tuning half-widths of 90 ° have been described in the lobula complex of dragonflies (O’Carroll, 1993). Orientation-sensitive neurons with much smaller receptive fields have been described in blowflies and houseflies (McCann and Dill, 1969), although in all cases the morphology of such neurons remains unknown.

Of the four possible mechanisms, only the hypothetical orientation-sensitive channels can explain the discrimination behavior of blowflies in all three experiments. The model proposed for a system of orientation-selective neurons predicts that all differences in orientation should be discriminated (Srinivasan et al., 1994). Directional motion cues, however, are only sufficient for discriminating horizontal from vertical and +45 ° from −45 °. The flow field composition hypothesis is sufficient only to explain discrimination of horizontal from vertical, and the flow field shape hypothesis can explain only the discrimination of +45 ° from −45 ° and of +5 ° from vertical. Thus, either a variety of mechanisms are employed by blowflies to discriminate different orientations or a discrete network of orientation-selective neurons underlies all the discriminatory behaviors. The neural correlates of these large-receptive-field, broadly tuned channels have not yet been identified in flies.

All studies of orientation discrimination by honeybees are on flying bees. Flying towards visual cues affords additional types of visual inputs, such as optic flow as the insect pitches or rolls its body or elevates it relative to the ground. Presumably, such additional inputs could expand the possible types of mechanism available for the discrimination of orientation in honeybees. Walking and flying are also different in the temporal domain. In the experiments presented here, flies walked with an average speed of 2.75±0.80 cm s⁻¹ (N=7). Bees, however, fly at speeds of approximately 100 cm s⁻¹ when approaching stationary gratings (92 cm s⁻¹ calculated from the 0 Hz panel of Fig. 6 in Srinivasan and Lehrer, 1984, and 131 cm s⁻¹ calculated from the top left panel of Fig. 6 in Srinivasan and Lehrer, 1988). A flying insect will experience higher contrast frequencies as it flies quickly past a stationary grating than a walking insect would experience. The optimal contrast frequencies at which flies can discriminate orientation are not known, but orientation discrimination in honeybees is best at contrast frequencies between 0 and 36 Hz (Giger and Srinivasan, 1997b). Orientation discrimination, therefore, functions in the absence of, or at low contrast frequencies relative to, other visually guided behaviors in honeybees (Giger and Srinivasan, 1997b). Giger and Srinivasan (Giger and Srinivasan, 1997b) suggested that the orientation processing system in honeybees may be best suited for stationary images. Flies may have similar temporal constraints on orientation processing, and it is not surprising, therefore, that they can discriminate orientation while walking at speeds at which relatively low contrast frequencies would be encountered.

In summary, blowflies are capable of discriminating between patterns of different orientation in a learning paradigm in which they are free to explore. Under these conditions, they can discriminate very fine spatial differences. The significance of this behavior to natural foraging and the neuronal correlates of orientation-detection are unknown.

We would like to thank Drs Martina Wicklein, Mark Willis and Piotr Jablonksi for their invaluable input to this manuscript. We would also like to thank Dr Mark Willis for the use of his Peak motion analysis system, Mr Keith J. Rosen, a.k.a. the comma master, for helpful editorial comments on the manuscript, and Mr Robert Gomez for excellent technical assistance. We would also like to thank the two anonymous reviewers for their helpful suggestions. This work was supported by a grant from the NIH (RR08688) to N.J.S. and a National Research Service Award from the NIMH (MH11866) to H.R.C.