Fruit flies increase attention to their frontal visual field during fast forward optic flow

Abstract

Fruit flies must compensate for the limited light gathered by the tiny facets of their eyes, and image motion during flight lowers light catch even further. Motion blur is especially problematic in fast regions of the visual field, perpendicular to forward motion, but flow fields also contain slower regions, less affected by blur. To test whether fruit flies shift their attention to predictably slower regions of a flow field, we placed flies in an arena simulating forward flight and measured responses to turning cues in different visual areas. We find that during fast forward flight, fruit flies respond more strongly to turning cues presented directly in front, and less strongly to cues presented to the sides, supporting the hypothesis that flying fruit flies shift visual attention to slower moving regions less affected by motion blur.

1. Background

The fruit fly, Drosophila melanogaster, is a passively unstable flier [1] and relies on vision to actively stabilize flight [2–5]. Flying flies continuously collect information about self-motion to adjust heading and counter unintended motion, such as perturbations owing to wind [6,7]. The utility of vision depends on the information present in a scene, but is further limited by signal noise produced by dim light and fast motion (figure 1a) [8].

Figure 1.

(a) Top-down views of a fruit fly moving straight through a 10 cm corridor, with high contrast patterns along the walls, at dusk light levels. Leftmost figure shows the mean image displacements (Δ image) at azimuths around a fly that is laterally perturbed from its course by 2 cm. Regions orthogonal to the direction of travel, especially the dorsal and ventral regions where image displacement (α) is greater than interommatidial angle (Δϕ), show the strongest effect to perturbations. Rightmost figures show the average signal-to-noise ratio (SNR) at different azimuths and how these ratios change in response to increased forward speed (arrow length). Demarcations indicate angles with SNR below 5 (the Rose criterion) and 1, indicating lower SNR where perturbation signals were highest. (b) Computer-generated stimuli projected onto walls of the arena. (c) View inside the arena, with a fly immobilized in the centre, below an infrared light beam casting shadows onto wing beat sensor below. The tails on the dots are not rendered, but shown here to indicate motion. (d) View of one stimulus from slightly behind the fly. Blue arrows indicate lateral motion of dots located inside the annulus (leftward in this example). Yellow arrows indicate the forward motion of dots outside the annulus. (e) Top–down cross-sectional view of the fly and stimulus in figure 1d to show that annulus position is described by its middle polar angle (θ) measured from the anteroposterior (AP) axis.. (f) Sample trial data showing ΔWBA during a lateral left and right turn stimulus.

Rotation and translation produce characteristic patterns of image motion over the entire visual field [9,10]. For example, an insect moving forward sees a focus of expansion in the direction of travel, motion away from this focus throughout the visual field and a focus of contraction behind [9,11]. Flow fields can be described by vectors representing image velocities in each part of a scene [11,12]. During forward translation, for example, the lowest image speeds are directly in front of and behind the insect (near the focuses of expansion and contraction) and the highest are perpendicular to the direction of travel (although image speeds are further reduced by object distance).

Images moving quickly over the retina have reduced time for light absorption [13]. Because photon absorption follows Poisson probability distributions, in which variance equals the mean [8,14], noise (standard deviation) increases only by the square root of light level [13,14]. Fewer absorptions produce lower absolute noise, but higher relative to the signal [8,15], implying that speed degrades image quality in a manner similar to dim light, in which signal-to-noise ratio (SNR) is a useful measure of image information [10,11].

A variety of flying insects have been shown to use specific regions of their visual field for different tasks, such as locusts regulating ground speed [16], honeybees controlling head turning [17] and hawkmoths maintaining flight stability [18]. Since regions of optic flow differ in speed, animals might shift the visual areas for course correction to predictably slower regions. Specifically, a forward-translating fly, by shifting attention from the periphery towards the focus of expansion, would then stabilize flight with visual regions of improved SNR (figure 1a, right). To determine if fruit flies use this strategy, we monitored the corrective steering responses of tethered Drosophila in a visual flight arena while simulating forward flow and presenting lateral perturbation cues at different angles.

2. Methods

(a). Subjects and preparation

Female Drosophila melanogaster were raised on a standard food medium, maintained on a 16 h : 8 h light:dark cycle and collected 4–6 days after eclosion. We cold-anesthetized flies and tethered them to a 0.1 mm tungsten rod by the dorsal prothorax, then allowed at least an hour at room temperature for recovery. We then suspended them in the centre of a virtual arena (figure 1b) for testing. Each of the 65 flies participated only once under each experimental condition.

(b). Steering responses

Drosophila respond to visual stimuli by adjusting the relative amplitudes of their left and right wing beats [6]. We measured steering responses with an infrared light emitter above the fly, and sensor pair below (figure 1c). A wing beat analyzer recorded the wing shadows as voltages, giving estimates of each wing beat amplitude (WBA) based on sensor occlusion. The right WBA subtracted from the left gives ΔWBA, which correlates with steering yaw torque, positive for right turns, negative for left, and near zero when a fly is not turning [19,20].

(c). Visual stimuli

The virtual arena was a perspex cube with 200 mm edges, and the rear face open for access. A projector and mirrors displayed unique images simultaneously on each of the 5 faces, covering 5/6 of the entire visual field (figure 1b,c), and corrected for the variable distances to the corners and sides from the perspective of the fly tethered in the centre. Room lights were dimmed to increase the projection contrast. We displayed 24 open-loop 1.5 s trials, with 2.5 s rest between each. During rests, flies controlled the angle of a vertical stripe in closed-loop, which active flies tend to fixate in front. Stripe fixation motivates flies to continue participation, and ensures they enter each experiment in an active tracking behavioural state. Ability to fixate a stripe was the only criterion for including flies in experiments.

During trials, we displayed a field of dots in a uniform random distribution, averaging 4.7 per steradian. We divided dots into regions either inside or outside of an annulus (ring), designated by two variable angles measured from the flies' anteroposterior (head-to-thorax) axis (figure 1d,e). Annuli were centred forward and always covered 29% of the visual area, meaning annuli near the frontal field were thicker, while those near the periphery were thinner (figure 1e; electronic supplementary material, figure S1 for further details). Dots in an annulus moved left or right at 0.76π rad s⁻¹, simulating visual sideslip, to induce directed steering responses [21] (figure 1f). By contrast, dots outside the annulus (the remaining 71% of the arena), were either static (no forward flow (FF)) or flowed from front to back (FF) to simulate forward motion at 0.76π rad s⁻¹ when perpendicular to the fly, and provided no left or right steering cues. Dots moving out of their region during a trial became invisible to avoid overlap.

We assigned the annulus to one of four positions per trial (with middle angles of 41, 59, 75, 90 degrees from the anteroposterior axis), and combined each forward flow with each of the four sideslip annulus positions and two sideslip directions (left and right steering) for a total of 16 trials. The ΔWBA responses were sign-corrected such that following annular dots was considered positive, and averaged between 1000 and 1500 ms after the onset of a presentation, then pooled for each condition (unpooled data shown in electronic supplementary material, figures S2 and S3). Paired t-tests then compared mean responses across experimental conditions.

3. Results

The lateral dot flow, even restricted to the small visual regions of the annuli, induces steering responses that correspond to its motion (figure 2a). With no forward flow, flies respond to lateral cues originating forward or to the side, slightly favouring lateral cues. But in the presence of forward flow, flies make significantly stronger corrective responses to the most anterior annulus (41°, T = −2.47, p = 0.016), and significantly weaker responses to the lateral annulus (90°, T = 3.42, p = 0.001) (figure 2b). Responses at the other intermediate annuli presentations (59° and 75°) show no significant differences between forward flow conditions (T = 0.75, p = 0.456 and T = 0.32, p = 0.751, respectively).

Figure 2.

Steering responses depend on visual angle of the cue. (a) The mean ΔWBA for each experimental condition with standard error shaded (n = 65). (b) ΔWBA at each presentation position and forward flow condition. Asterisks indicate paired t-test significance of p < 0.05 (single) and p < 0.005 (double). (c) Representation of the four tested annulus positions, from most anterior to most lateral. Double arrows indicate ‘left or rightward' motion since both were pooled.

4. Discussion

During forward flight, regions perpendicular to forward typically have the greatest image speeds and suffer from the greatest motion blur. Our results suggest flies shift attention forward, towards the focus of expansion, consistent with a strategy to counteract the effects of motion blur, by both increasing responses to forward stimuli and decreasing responses to lateral stimuli.

Fruit flies typically execute long bouts of forward flight punctuated by rapid turns [22]. Similar attention strategies may occur for flow in other directions, or may be limited to only the common case of forward flight. The forward shift also suggests some value in using peripheral visual regions when motion blur is minimal, as flies might otherwise rely only on forward regions regardless of flow speed. This may reflect the higher visual displacement during sideslip in the peripheral regions as an insect passes near environmental features (figure 1a, left). Previous work has shown that flies spatially blur in peripheral regions during forward flow, consistent with neural summation that improves SNR [23], and that flies shift attention to different regions when responding to translational and rotational flow, which produce different image speeds on the retina [24].

The increased attention to the innermost annulus may also have significance for landing behaviour. Flies approaching a landing surface use flow field velocity [25] and this annulus encompassed the 45° visual region where maximal flow speeds occur. Flight speed is not the only factor that can reduce the visual SNR. Diminished light intensity, eye size, or image contrast might similarly make regions near the focus of expansion more useful for flight stabilization. Small flies face many challenges in executing competent flight, and attentional shifts may be one mechanism helping to maximize their information uptake.

Data accessibility

Data available from the Dryad Digital Repository: http://dx.doi.org/10.5061/dryad.4p3k567

Authors' contributions

N.P. and J.T. designed the experiment and wrote the manuscript. N.P. collected data. Both authors agree to be held accountable for the content therein and approve the final version of the manuscript.

Competing interests

We declare no competing interests.

Funding

This work was supported by the National Science Foundation, grant number IOS-1750833 to J.T.