Edge detection depends on achromatic channel in Drosophila melanogaster

Resulting from the contrast between neighboring areas, edges are one of the most important visual cues for object segmentation and recognition (Marr and Hildreth, 1980; Biederman, 1987; Laws et al., 2003). Invertebrates to vertebrates, such as flies, honeybees, beetles, goldfish and budgerigars, showed a strong tendency to orient toward an edge (Varju, 1976; Lehrer et al., 1990; Osorio et al., 1990; Bhagavatula et al., 2009; Kim and Jung, 2010). The potential cues to define an edge include brightness, color, texture and relative motion (Mullen, 1985; Livingstone and Hubel, 1987; Regan et al., 1992; De Weerd et al., 1994; Greene and Brown, 1995; Reppas et al., 1997; Leventhal et al., 1998).

In insects, it has been shown that honeybees can detect boundaries that originate in brightness contrast. Behavioral analysis in honeybees suggested that edge detection is a color-independent process, which originates in the green-sensitive photoreceptor (Lehrer et al., 1990). Further experiments showed that honeybees could also use texture or the relative motion of different textures to define a boundary (Srinivasan et al., 1990; Lehrer and Srinivasan, 1993), and the combination of cues that define an edge could not enhance edge detection (Lehrer and Srinivasan, 1993). The honeybee's visual system processes motion/edge information in one channel and color information in another. In Drosophila it has been reported that flies tend to walk or fly toward simple objects such as a vertical black bar on white background (Reichardt and Wenking, 1969; Horn and Wehner, 1975; Reichardt and Poggio, 1976; Bülthoff et al., 1982; Neuser et al., 2008). Flies could also use vertical edges to control flight speed and judge the distance to objects (Frye and Dickinson, 2007).

In each Drosophila compound eye, there are about 800 ommatidia, each containing eight photoreceptor cells (R1–R8; Clandinin and Zipursky, 2002). All the ommatidia contain the major photoreceptor cells R1–R6, which express blue and UV-sensitive rhodopsin 1 (Rh1), and photoreceptor cells R7/R8, which express different combinations of rhodopsins, such as Rh3/Rh5 or Rh4/Rh6. Approximately 70% of the ommatidia are named the ‘yellow’ type, and they express UV-sensitive Rh4 in photoreceptor cell R7, and green-sensitive Rh6 in photoreceptor cell R8. The other 30% of the ommatidia, named the ‘pale’ type, express UV-sensitive Rh3 in photoreceptor R7, and blue-sensitive Rh5 in photoreceptor cell R8 (Cook and Desplan, 2001). Axons of R1–R6 photoreceptor cells project to the lamina, the first neuropile in the optic lobe, while axons of photoreceptor cells R7 and R8 project directly to the medulla, the second neuropile in the optic lobe (Hardie, 1985). The variety of rhodopsin in the retina and complex neural circuitry in the optic lobe provide the possible physiological basis for color vision in Drosophila (Gao et al., 2008; Morante and Desplan, 2008).

The R1–R6 photoreceptor channel has been shown to participate in phototaxis, tropotaxis and motion detection (Heisenberg and Buchner, 1977; Rister et al., 2007; Yamaguchi et al., 2008). However, the R7/R8 photoreceptor channel is thought to be crucial for color vision and the detection of e-vector orientation but not motion detection (Heisenberg and Wolf, 1984; Hardie, 1985; Yamaguchi et al., 2008). Previous experiments showed that R1–R6 were necessary for fixation and optomotor behavior in both flying and walking flies, and R7 played a secondary role (Heisenberg and Buchner, 1977; Morton and Cosens, 1978; Mimura, 1982; Coombe, 1984; Strauss et al., 2001). Lamina pathways L1, L2 and L3, which are downstream of R1–R6, are also suggested to be necessary in orientation behavior (Rister et al., 2007). All these experiments showed that the R1–R6 channel is important in orientation behavior. However, there has been no definitive data about the function of R7/R8 channel in orientation behavior towards edges, making it necessary to selectively switch off the whole color channel to fully separate orientation behavior towards edges from color vision. In the present study, using a new color light-emitting diode (LED)-based Buridan's paradigm in which edges were produced by adjustable colors, we investigated the relationship between these photoreceptor channels and edge detection in the fly. We found that motion channel (R1–R6) is necessary and sufficient for edge detection, and that the color channel (R7/R8) is dispensable for edge detection.

The flies used in all experiments were raised on a standard medium of corn meal and molasses (Guo et al., 1996) in a 12 h:12 h light:dark cycle at 25°C and 60% humidity. Three- to five-day-old flies were tested at least 24 h after their wings had been shortened to one-third of their normal length under CO₂ anesthesia. The following flies were used in the experiments: wild-type Canton S (WTCS), wild-type Berlin (WTB), ninaE¹⁷ (Hanai et al., 2008), sev^LY3;rh5²;rh6¹ (Yamaguchi et al., 2008), ninaE¹⁷,P[Rh1>3] and w;;ninaE¹⁷ e[s],P[Rh1>4] (Feiler et al., 1992), yw;;ninaE¹⁷,P[Rh1>5,y⁺] (Chou et al., 1996), and yw;;ninaE¹⁷,P[Rh1>6,y⁺] (Salcedo et al., 1999), ninaE¹⁷ e[s],P[Rh1>4], ninaE¹⁷,P[Rh1>5,y⁺] and ninaE¹⁷,P[Rh1>6,y⁺].

We used Buridan's paradigm (Bülthoff et al., 1982) with modifications. During the experiment, a wing-cut fly was allowed to walk for 3 min on a circular platform, 86 mm in diameter, surrounded by a water-filled moat to prevent its escape (Fig. 1A). The moat was surrounded by a panoramic LED display, 290 mm in diameter and 345 mm in height. The LED display was a cylinder of evenly distributed 128 (row) × 32 (column) LED units and was computer-controlled with LED Studio software (Shenzhen Sinorad Medical Electronics, Shenzhen, China). The refresh rate of the LED panels was 400 Hz. A camera (WV-BP330, Panasonic System Networks, Suzhou, China) directly above the arena was connected to a computer to record the fly's walking track at a rate of 12 frames s⁻¹, and position coordinates of the fly in each frame were calculated using Limelight software (Coulbourn Instruments, Whitehall, PA, USA). Male and female flies were used alternately.

Each LED unit on the display included three diodes that separately emitted red, green and blue light. The absolute irradiance of the blue and green LEDs in all the visual stimulation was measured with the Spectroradiometer USB2000+ (Ocean Optics, Dunedin, FL, USA). The peak of the blue LED's spectrum curve was at 466 nm, while the peak of the green LED was at 519 nm (Fig. 1B). We used the red/green/blue (RGB) value to describe the brightness and chromatic properties of a visual stimulus. In the RGB color model, brightness ranging from completely dark to fully bright was divided into 256 grades, with 0 as the darkest and 255 as the brightest. RGB[0,0,255] (absolute irradiance value: 2.9483×10¹³ quanta cm⁻² s⁻¹) represents the brightest blue with the red and green diodes off; RGB[0,255,0] (absolute irradiance value: 3.9557 ×10¹³ quanta cm⁻² s⁻¹) represents the brightest green with the red and blue diodes off (Fig. 1B). In some cases, we simply use B[m] and G[n] to represent blue color of RGB[0,0,m] and green color of RGB[0,n,0] for convenience.

A circle composed of color dots, as seen in Fig. 1C, is used to represent the horizontal view of the visual stimulus on the LED cylinder. Different angles are marked around the circle (Fig. 1C). The orientation of a fly towards a certain angular position (θ in Fig. 1C) on the LED cylinder is indicated by the projection of a vector formed by two consecutive position points (red dot and green dot in Fig. 1C) of the fly on the platform (Fig. 1C). The error angle α represents the smallest angle between a fly's orientation and an edge. The range of error angle is –102 to 102 deg (Strauss and Pichler, 1998). The probability distribution of a group of flies' orientation over 3 min is schematically shown in a polar map (Fig. 1D). The fly's orientation towards an edge is quantified by the performance index of orientation behavior towards edges (PIEO). When two edges are present at 0 and 180 deg on the circle by setting 0–180 deg of the circle as one color and 180–360 deg of the circle as the other color, the circle is divided into four equal sections of 45–135, 135–225, 225–315 and 315–0–45 deg. The 135–225 and 315–0–45 deg sections correspond to an edge, the other two sections to the absence of an edge. The probabilities of a fly's orientation in sections with or without an edge are defined as P_edge and P_edgeless. The PIEO is then defined as (P_edge–P_edgeless)/(P_edge+P_edgeless) (Fig. 1E), and the performance index of color preference (PICP) is used to measure the fly's preference for different colors. The probabilities of the fly's orientation in 0–180 and 180–360 deg sections are defined as P_0–180 and P_180–360. PICP is then defined as (P_0–180–P_180–360)/(P_0–180+P_180–360) (Fig. 1F). The mean walking speed is defined as total walking distance divided by its time. Turning angle is the change of a fly's orientation between two frames.

The Lilliefors test was used to estimate whether data followed a normal distribution. We used the t-test to determine the significance of differences among groups. The t-test was also used to compare the PIEO (PICP) value of a group to zero, which is the theoretical random value.

We adopted a revised Buridan's paradigm, in which two edges could form between the blue half and green half of an arena; hence edges are constituted of both color contrast and luminance contrast. Luminance of the two colors was adjusted to the highest level in the RGB mode (see Materials and methods). Individual WTB or WTCS flies with clipped wings were allowed to walk freely for 180 s on the platform surrounded by the LED arena. If the arena was made of a homogenous mixture of blue and green (RGB[0,255,255]), the flies showed even distribution of orientation (Fig. 2A,B for WTB; Fig. 2I,J for WTCS) and the direction preference for edge represented by PIEO was not significantly different from zero (theoretical value in random distribution, Fig. 2E for WTB; Fig. 2M for WTCS). In contrast, in the presence of a blue/green edge, i.e. one half of the arena was blue and the other half green, the flies showed a higher probability of walking towards the blue/green edges (Fig. 2C,D for WTB; Fig. 2K,L for WTCS). PIEOs were significantly higher than zero (Fig. 2E for WTB; Fig. 2M for WTCS). The error angle distributions of WTB and WTCS in the presence of a homogenous mixture of blue and green are flat, while the distributions in the presence of a blue/green edge show a peak near 0 deg, but the peak is smaller in WTCS (Fig. 2F for WTB; Fig. 2N for WTCS). Yet the distributions of turning frequency of WTB in the presence of a homogenous mixture of blue and green and in the presence of a blue/green edge are similar (Fig. 2G), while the turning frequency distribution of WTCS is less steep in the presence of a blue/green edge as in a homogenous mixture of blue and green (Fig. 2O). The mean walking speed of WTB is higher in the presence of a blue/green edge than in a homogeneous mixture of blue and green (Fig. 2H), while the mean walking speed of WTCS is decreased (Fig. 2P). Such edge-orientation ability was further confirmed by changing the number of edges. As shown in Fig. 2R,S more edges in the arena led to preferences for corresponding directions. We gradually changed the luminance from maximum to zero in the LED cylinder to form a visual stimulus with only one obvious edge. It is interesting that when only one edge was present, the flies showed preferences not only for the edge, but also for the direction opposing the edge (Fig. 2Q). This so-called anti-fixation behavior has been reported in previous experiments with Buridan's paradigm (Reichardt and Poggio, 1976). In addition to orientation behavior towards edges, we also noticed that flies demonstrated a preference for the blue area over the green area, as indicated by PICP in Fig. 2E,M.

Taken together, these results show that a blue/green demarcation is able to generate edge-orientation behavior in the adult fly.

As the blue/green edge constituted a difference not only in color but also in brightness, we wondered whether both or only one of these factors contributes to the fly's orientation behavior towards edges. So we tried to separate the contribution of brightness contrast and color contrast in our visual stimulus by adjusting the color or luminance that creates the edges. We first examined the orientation behavior of wild-type flies in the presence of blue/green edges with various blue/green intensity ratios, which was achieved by linearly changing the luminance of the blue light and fixing the luminance of the green light at G[255] (Fig. 3A). We found that there was a blue/green intensity ratio, the so-called point of equal luminance (POE), at which WTB did not show significant orientation behavior towards edges. The POE corresponded to a blue/green ratio of B[226]/G[255] (Fig. 3B). Furthermore, we noticed that WTB flies showed a preference for the blue area not only at the POE but also at other blue color intensities, such as B[220], B[232] and B[255] (Fig. 3C). To test the equiluminant stimuli at a range of mean luminances, we fixed the luminance of the green light at lower intensities of G[232] and G[197] and linearly changed the luminance of the blue light. We also found corresponding POE (Fig. 3D–I). However, when we fixed the luminance of the blue light at intensities of B[197] and linearly changed the luminance of the green light, the POE was not as clear as in previous cases (Fig. 3J,K). It is worth noting that according to the unitary irradiance spectrum, the blue/green irradiance ratio at the POE in most cases is about 0.4 (Fig. 3B,E).

Because at POE the blue/green contrast still existed, the loss of orientation behavior towards edges suggests that orientation behavior towards edges is independent of color contrast but dependent on luminance contrast. We then tested whether flies could use only luminance contrast-based edge to control their orientation. So we tried two types of luminance contrast-based edges: G[255]/G[197] in green and B[255]/B[197] in blue. In both cases, WTB flies showed obvious edge-orientation behavior (Fig. 3M). The PIEO and PICP values of WTB in the blue luminance contrast were not significantly different from the results in the green luminance contrast, as shown in Fig. 3N. Finally we wanted to know whether the color contrast contributes to orientation behavior towards edges in the presence of luminance contrast, so we tried to compare the orientation behavior towards edges under the same luminance contrast with and without color contrast. We found that the POE of blue/green ratio is B[226]/G[255] (see Fig. 3B), which means that flies' perception for B[226] and G[255] is equal luminance. Then we tried two types of edges: B[226]/B[255] and G[255]/B[255]. Since B[226] and G[255] are equal luminance, two types of edges of B[226]/B[255] and G[255]/B[255] should present a similar luminance contrast, but their color contrasts are different (Fig. 3O). We found that there was no significant difference between the orientation behavior towards edges under these two conditions (Fig. 3P), indicating that color does not enhance orientation at perceptually approximate luminance contrasts. These results suggest that flies could use only luminance contrast-based edges but not color contrast-based edges to guide their walking orientation.

As orientation behavior towards edges is luminance dependent and color independent, we wanted to establish the underlying physiological mechanism. For this, we used an Rh1 mutant, ninaE¹⁷, in which the monochromatic R1–R6 photoreceptor cells are dysfunctional, but the chromatic R7/R8 photoreceptor cells are normal. Thus this mutant is normal in color sensing but abnormal in luminance sensation and motion detection (Yamaguchi et al., 2008). Another mutant line we used was sev^LY3;rh5²;rh6¹, in which the chromatic photoreceptor cells R7 are absent and the working rhodopsins of Rh5 and Rh6 in photoreceptor cells R8 are abnormal. Thus this mutant is normal in luminance sensation but abnormal in color sensation. As shown in Fig. 4A,C, ninaE¹⁷ appeared to be defective in orientation behavior towards edges while sev^LY3;rh5²;rh6¹ was not significantly different from the wild-type control in orientation behavior towards edges (Fig. 4A,C,D). These results demonstrate that R1–R6 photoreceptor cells, but not R7/R8, are necessary for orientation behavior towards edges. Besides, both WTCS and sev^LY3;rh5²;rh6¹ showed a preference for the blue area, and yet ninaE¹⁷ showed a preference for the green area.

To establish whether R1–R6 photoreceptor cells are sufficient for orientation behavior towards edges, we ectopically expressed different rhodopsins in the R1–R6 photoreceptor cells in the Rh1 mutant background flies. We found that over-expression of Rh4, Rh5 or Rh6 could efficiently restore the edge-orientation defect in the ninaE¹⁷ mutant, but expression of Rh3 failed to restore the edge-orientation defect (Fig. 4A,E). Expression of Rh3 actually resulted in an edge-avoiding effect, which was an unexpected result (Fig. 4A). An over-expression of Rh5 could also restore the color preference in the ninaE¹⁷ mutant to the WTCS level. The peak of error angle distribution of sev^LY3;rh5²;rh6¹ in the presence of blue/green edges was near 0 deg, while that of the distributions of ninaE¹⁷ and ninaE¹⁷,P[Rh1>3] was near 100 and –100 deg (Fig. 4J). Yet the distributions of turning frequency of ninaE¹⁷, sev^LY3;rh5²;rh6¹ and ninaE¹⁷,P[Rh1>3] in the presence of blue/green edges were similar (Fig. 4K). The mean walking speed of these three lines were also not significantly different (Fig. 4L). Ectopic expression of Rh4 or Rh5 in R1–R6 photoreceptor cells could fully restore the defect of w;;ninaE¹⁷ to the WTCS level, but expression of Rh6 failed to restore the edge-orientation defect in yw;;ninaE¹⁷. Such ectopic expression of Rh4 or Rh5 switched the preference of w;;ninaE¹⁷ from green to blue, but the expression of Rh6 resulted in loss of color preference between blue and green (Fig. 4B). Taken together, these results show that R1–R6 photoreceptor cells are both sufficient and necessary for orientation behavior towards edges in Drosophila.

According to above results, the POE should be decided by the spectral sensitivity of photoreceptors R1–R6. To verify whether ectopic expression of other rhodopsins in R1–R6 might shift the POE, we examined the orientation behavior of all the rescue flies in the presence of blue/green edges with various blue/green intensity ratios (Fig. 5). We found that POE was drastically shifted with ectopic expression of rhodopsin in R1–R6, e.g. replacing Rh1 with Rh5 compared with WT flies (Fig. 3B), less blue light was required to balance the green intensity (Fig. 5G,H), giving a blue/green ratio of 0.169 compared with 0.426 for WT flies.

We have shown that edge-orientation behavior in Drosophila can be elicited by luminance contrast but not by color contrast. The monochromatic channel, outer photoreceptor cells R1–R6, contribute to edge detection, though the color channel (inner photoreceptor cells R7 and R8) does not. There are several lines of evidence for this conclusion. First, wild-type flies can orient towards achromatic edges but not to edges defined by equiluminant color contrast. Second, color contrast did not enhance orientation behavior towards edges at a salient luminance contrast. Third, flies with mutated R1–R6 photoreceptor cells could not perform edge detection. Fourth, genetic removal of the function of photoreceptor cells R7 and R8 did not diminish the orientation index to edges. Finally, flies with the Rh1 in R1–R6 photoreceptor cells changed to Rh4, Rh5 or Rh6 could perform orientation behavior towards edges, whereas flies with the Rh1 changed to Rh3 showed edge aversion. Furthermore, such ectopic expression of different rhodopsins in outer photoreceptors altered the POE.

Equal absolute irradiance green and blue will be perceived by the human eye as brighter green (Rovamo et al., 1996). Similarly a wild-type fly may perceive blue and green as having a different brightness. In flies, we do not even know all the luminosity functions, so we could not certify whether two stimuli had exactly the same luminance contrast. However, according to the results of POE experiments in Fig. 3B, we speculate that the stimuli used in Fig. 3O (B[226]/B[255] and G[255]/B[255]) should present a similar luminance contrast.

At the point of POE (Fig. 3B), the relative photon catch between Rh4 and Rh6 receptors for the blue field is about 0.051, while for the green field it is close to zero, which means that there is a significant color contrast for R7 and R8. Moreover, previous research found a type of medulla neuron contacting the R8 photoreceptors in different ommatidia, which could be the potential neural substructure for comparing the light signal between Rh5 and Rh6 (Morante and Desplan, 2008). These two points suggest that there may well be effective color contrast between the blue and green stimuli at the point of POE.

Expressing Rh3 in R1–R6 photoreceptor cells did not restore the edge-orientation defect in Rh1 mutant background flies, but gave rise to an inverted edge preference. This indicated that replacing Rh1 with Rh3 could not support the orientation behavior towards edges in flies, but such flies may be able to detect edges. Previous electrophysiological experiments reported that flies with Rh1 changed to Rh3 in R1–R6 photoreceptor cells showed similar but slight slower response to UV flash than wild-type flies (Feiler et al., 1992; Liu et al., 2008), which also suggests that the substitution of Rh1 with Rh3 may cause other unpredicted effects. Expression of Rh6 in R1–R6 did not rescue the orientation behavior towards edges in yw,ninaE¹⁷ but rescued the orientation behavior towards edges in ninaE¹⁷. This suggests the difference in eye color may also cause other unpredicted effects.

Previous research showed that ocelli may function in phototaxis and edge fixation for walking blowflies (Wehrhahn, 1984). Behavioral results in fruit fly showed that ocelli are not necessary in phototaxis behavior but may function in modulating compound eye sensitivity (Hu and Stark, 1980). In our experiments, flies with mutation in R1–R6 but with normally functioning ocelli showed no obvious orientation behavior toward edges. This result indicates that ocelli do not participate in orientation behavior toward edges, which is consistent with previous reports.

Both a previous publication (Strauss et al., 2001) and our work have shown that orientation of ninaE¹⁷ mutant is abnormal, although in the study of Strauss et al., ninaE¹⁷ showed residual orientation in Buridan's paradigm. The difference between our results and previous results may mainly be caused by the obviously different experimental conditions. In our experiments, the visual stimulus was two edges formed between the blue half and green half of the LED arena. Luminance of the two colors was adjusted to the highest level (RGB[0,255,0]/RGB[0,0,255]) in the RGB mode. ninaE¹⁷ was tested under the visual condition with both color contrast and luminance contrast. In our experimental conditions with color stimulation, ninaE¹⁷ did not show edge orientation behavior but showed weak preference for the green area. In addition, genetic background is clearly important for this behavior, given that WTB showed an obviously higher orientation score than that of WTCS in Fig. 2.

There are two main causative factors in color preference – chromatic difference and luminance difference. Wild-type flies showed a color preference for the blue area at POE with relative high mean luminance, which could be caused by chromatic difference (Fig. 3C,F). However, flies did not show color preference at POE with low mean luminance (Fig. 3I). Flies also showed a similar level of color preference for the blue area at other blue/green brightness ratios as at POE (Fig. 3C,F); this could be caused by a chromatic difference in the blue and green area or residual luminance contrast, which could not be excluded in our experiments. In our experiments, ninaE¹⁷ showed a preference for the green area and sev^LY3;rh5²;rh6¹ preferred the blue area (Fig. 4B). This is consistent with a previous study by Yamaguchi et al. (Yamaguchi et al., 2010).

Our work implies that edge detection is a color-blind process in the fruit fly, although the achromatic channel in Drosophila is blue and UV sensitive (Kirschfeld and Franceschini, 1977; Britt et al., 1993). Previous research in honeybees implied that edge detection is also a color-independent process (Lehrer et al., 1990; Srinivasan, 2011). The reason for achromatic processing of edges in these two species may have originated in the natural environment, in which achromatic processing of edges is sufficient for the insect to survive.

In goldfish, the edge-triggering response is an intrinsic factor of optomotor response, implying that edge detection has an internal relation with motion detection (Kim and Jung, 2010). In our experiments, orientation behavior towards edges originates from the monochromatic channel (outer photoreceptor cells R1–R6), which is also the motion channel. This also indicates that orientation behavior towards edges is related to the optomotor response.

Further research about the neural circuitry underlying edge detection in Drosophila could benefit from knowledge about the fly's visual system and powerful genetic tools in Drosophila, which will contribute to our understanding of how edge detection and parallel processing of visual information are executed in other animals.

We thank Dr Charles Zuker (University of California, San Diego), Dr Steven Britt (University of Colorado, Denver) and Dr Martin Heisenberg (University of Würzburg, Würzburg) for the flies. We also thank the Drosophila Genetic Resource Center at Kyoto Institute of Technology (DGRC). We thank Angelique Paulk for helping to calculate the relative photon catches of different receptors.

 
• LED

  light-emitting diode

 
• PICP

  performance index of color preference

 
• PIEO

  performance index of orientation behavior towards edges

 
• POE

  point of equal luminance

 
• RGB

  red, green, blue

 
• Rh1

  rhodopsin 1

 
• Rh3

  rhodopsin 3

 
• Rh4

  rhodopsin 4

 
• Rh5

  rhodopsin 5

 
• Rh6

  rhodopsin 6

 
• WTB

  wild-type Berlin

 
• WTCS

  wild-type Canton S