Multisensory integration for active mechanosensation in Drosophila flight

SUMMARY

To support robust behaviors in highly variable environments, animals rely on active sampling of their sensory surroundings. Here, we use tethered, flying Drosophila melanogaster and a multisensory behavioral apparatus simulating forward flight to determine how visual and mechanosensory information are integrated and control active movements of an important multimodal sensory organ, the antennae. We found that flies perform active antennal movements in response to varying airflow, and that the direction of these movements changes depending on the visual environment. Next, we found that antennal movements are amplified in the presence of visual motion, but only when the fly was flying. Through mechanical and optogenetic manipulation of mechanosensory input, we found that mechanosensory feedback is vital to antennal positioning at flight onset. Additionally, we observed unexpected changes in wingbeat frequency when the antenna was mechanically stabilized, suggesting that multiple antennal mechanosensors contribute to flight regulation. Finally, we show that integration of mechanosensory and visual cues for controlling antennal motion follows a “winner-takes-all” paradigm dependent on the stimulus frequency, mirroring visuo-mechanosensory guided behaviors in other species. Together, these results reveal novel behavioral gating of sensory information and expand our understanding of the efferent control of active sensing.

eTOC Blurb:

Mills et al. quantify how mechanosensory and visual inputs guide active mechanosensation in Drosophila. This study reveals that sensory modulation of active antennal movements is behaviorally gated, and multisensory integration follows a “winner-takes-all” paradigm dependent on stimulus frequency.

INTRODUCTION

An important function of nervous systems is the ability to quickly and efficiently integrate sensory information from multiple sensory modalities. Although a single sensory modality can be sufficient to guide motor actions in isolation (e.g. bacterial chemotaxis¹), complex animal behavior typically relies on several sensory modalities whose reliability varies depending on environmental conditions. Many behaviors rely on multisensory integration, such as host-seeking², the righting reflex (e.g. in hoverflies)³, and perception of self-motion⁴. To reconcile diverse sensory information from the environment, nervous systems implement different integrative mechanisms. In some cases, this integration is approximately the linear superposition of signals from distinct modalities^5,6 (but see Oie et al.⁷). However, in other instances, one sensory modality dominates (i.e. ‘winner-takes-all’) in its influence on motor commands^8,9.

Beyond integrating input from multiple sensory modalities, animals dynamically alter how they interact with and interpret their sensory environment through active movements¹⁰. These ‘active sensing’ behaviors, in which animals use muscles to position their sensors, directly modify how sensory information is acquired and can aid in effective sensation for guiding behavior¹¹. For example, rodents perform rhythmic, low-frequency movements of their whiskers to sense objects in their environment during goal-directed navigation. This active sampling of information (‘whisking’) influences incoming sensory input through sensory filtering¹² (‘gating’) and guides future sampling movements¹³. Active sensing behaviors are abundant across phyla^14–18, but insects in particular employ a remarkably conspicuous and diverse array of active antennal behaviors to sense their environment. For instance, cockroaches repeatedly touch and investigate objects of interest with their antennae¹⁹, and along with many other insects such as stick insects and beetles, will use active movements for obstacle avoidance^20,21. Many active olfactory behaviors have been described in insects, including active antennal movements for improving odor encoding in locusts²², contact-chemosensation for courtship^23–25, and hymenopteran odor trail navigation²⁶. Insect antennae can even be used to indicate relative body position, as Hadjitofi and Webb (2024) recently observed in the honeybee waggle dance²⁷.

Insect antennae are highly multimodal, and house many sensors supporting olfaction²⁸, hygrosensation and thermosensation^29,30, and mechanosensation^31–34. Thus, active antennal movements can influence multiple sensory modalities. These active antennal movements are guided by musculature located at antennal joints that enable directed acquisition of sensory information^20,35–40. This recurrent interaction between antennal motion and antennal sensation is critical for many insect behaviors, including tactile exploration^41,42, escape⁴³, oviposition⁴⁴, gravity perception^33,45–47, and flight^48–50. In addition to housing multiple senses, antennal movements are likely driven by multimodal information. Yet how different sensory inputs guide the movements of this dynamic sensor are not well understood.

In the fruit fly Drosophila melanogaster, antennal mechanosensation is crucial for flight control and navigation. A large population of mechanosensory stretch-receptive neurons, called Johnston’s Organ Neurons⁵¹ (JONs), detect passive motion of the antennae induced by both high-frequency air vibrations (e.g. courtship song^52,53) and static airflow^54,55. Flies use this mechanosensory information to guide several flight behaviors including orientation⁵⁶, long-term navigation⁵⁷, and groundspeed regulation⁵⁸. Flight is also heavily reliant on vision, which is crucial for stabilizing reflexes⁵⁹, rapid turns known as saccades^60–63, altitude control⁶⁴, obstacle avoidance⁶⁵, and groundspeed regulation^58,66. Further, evidence from previous studies demonstrated that antennal movements are regulated by both mechanosensory³⁹ and wide-field visual⁶⁷ sensory input. However, it is not fully understood how the brain integrates these two – often complementary – sensory modalities to regulate antennal positioning for important active sensing behaviors.

In this study, we used a custom multisensory behavioral apparatus to determine the role of mechanosensory and visual information in guiding antennal movements and forward flight regulation in tethered flying and nonflying fruit flies. We found that antennal position is actively regulated in response to varying airspeeds, regardless of whether the animal was flying or not. In addition, we discover that active antennal movements in the presence of visual stimuli increase the magnitude of airflow-dependent antennal deflections, indicating a potential increase in sensory gain. We also measured the influence of mechanosensory input on airflow-dependent antennal positioning and wingbeat kinematics by blocking antennal mechanosensory input and found that blocking antennal mechanosensation altered flight-dependent antennal positioning and wingbeat frequency. Additionally, we assessed the influence of progressive visual motion on active antennal movements and found that these responses are gated by flight state. Finally, we presented sinusoidal patterns of frontal airflow and progressive optic flow to gain a deeper understanding of how these multisensory stimuli are integrated by the antennal sensorimotor system and used this to build a model of the neural computations contributing to this integration.

RESULTS

Flies actively position their antennae in response to frontal airflow

To assess how mechanosensory and visual stimuli influence antennal movements and wingbeat kinematics during forward flight, we built a behavioral apparatus that delivers controlled frontal airflow and wide-field progressive optic flow along the anterior-posterior axis. We rigidly mounted flies to thin tungsten pins, fixed their heads in resting position, and placed them in the center of the behavioral apparatus. To record wingbeat information, we mounted a light-based wingbeat sensor posterior to the fly and recorded antennal movements at 60 frames per second using a camera positioned dorsal to the fly (Figure 1A, STAR Methods).

Figure 1. Multisensory behavioral apparatus and experimental design.

A) Schematic of the behavioral apparatus (not to scale). Fly is rigidly tethered to a pin and placed directly in the path of a frontal airflow tube. Visual displays are equidistant from the fly on the left and right side. Infrared light illuminates the fly from the front, and a wingbeat sensor detects light reflected by the wings from behind. A camera captures antennal movements from above the fly (dorsally). B) Schematic of the fly antenna. The pedicel is actuated by musculature located within the scape, enabling active movements. The arista is rigidly attached to the funiculus and is passively deflected by airflow. C) Image of the fly indicating points tracked on both the head and antennae, including a hair on the pedicel (blue), the arista (orange), and the base of four cephalic hairs (black). D) Schematic indicating measure of inter-antennal angle (‘iaa’) for the pedicel (blue) and arista (orange) antennal segments, and the head axis (midline). E) Block diagram depicting our experimental approach. The black lines and arrows denote the flow of signals, such as motor commands to the muscles or muscles acting on the wings. The bottom branch illustrates the visuomotor transformations that drive changes in wing kinematics (e.g., wingbeat frequency and amplitude) during tethered flight⁵⁸. Wing vibrations are detected by the antennae, illustrated by the feedback from the wings to the antennal mechanosensation block. Additionally, antennal movements may modulate wind sensing, as illustrated by the gray “active iaa” feedback arrow that goes through the mechanosensory block (illustrating its hypothesized indirect, modulatory effect rather than the direct flow of power or information). This paper characterizes the mechanosensory control of the antennal motor control loop, illustrated by the dashed branch (labeled by the ‘?’).

We tracked multiple points on several antenna segments (Figure 1B) to directly quantify antennal movements over time using DeepLabCut^68,69. To measure passive motion, we tracked the base and tip of the most distal segment, called the arista (Figure 1C and 1D, orange), which is a large, rigidly fixed hair-like appendage attached to the funiculus that transduces incoming airflow, amplifying passive motion^33,70,71. To measure active motion, we tracked the tip and base of a large hair (Figure 1C and 1D, blue) on the pedicel, the most proximal antennal segment that can move either passively or actively in Drosophila; this segment is actuated by four muscles located in the scape^38,39 (Figure 1B). We measured the angles of both the pedicel and arista relative to the midline of the head. Because both the visual and mechanosensory stimuli were bilaterally symmetric, we added the angles of the left and right antennae relative to the midline to compute the inter-antennal angle (‘iaa’, Figure 1D), a measure that has been extensively used in the past for gauging antennal position across insect species^36,49,72,73, for both the pedicel and arista.

A block diagram (Figure 1E) illustrates our experimental approach, the sensory stimuli we presented, and behavioral outputs we measured. Because the fly is tethered and the stimuli are presented in open-loop, changes to wingbeat kinematics do not drive reafferent visual feedback⁷⁴. However, antennal movements generate two forms of reafferent feedback: (1) the motor commands to the antenna drive changes to inter-antennal angle, potentially modulating wind sensing (Figure 1E, ‘active iaa’ feedback to mechanosensory block), and (2) changes in wingbeat kinematics can be detected by the antennae through high frequency vibrations^54,75 (Figure 1E, wing feedback to mechanosensory block); our study focuses on the first of these two forms of reafferent feedback.

To evaluate how flies actively position their antennae in varying airflow, we measured inter-antennal angles for both the pedicel and arista segments in response to wind speeds ranging from 0–300 cm/s. We presented all wind speeds in either ascending or descending order for 6 sec, for a total of 10 trials per fly (pseudo-randomized order, see STAR Methods). After a transition to a higher airflow rate, we typically observed a clear passive deflection (‘passive joint angle’, Figure S1A), followed by movements of the pedicel (Figure S1). To confirm that the active antennal movements we measured were not a product of passive displacement by wind, we performed the same experiment with freshly dead flies. In both dead flies and live flies in darkness (i.e., visual stimulus off), we observed an increase in the inter-antennal angle of the aristae with increasing wind speed (Figure 2A). Whereas the inter-antennal angle for the active, pedicellar segment reached a maximal increase of 2.7 deg at the highest wind speeds in dead flies (Figure 2B, light gray trace), live nonflying flies in darkness exhibited the opposite response, decreasing their active inter-antennal angle as wind speeds increased (Figure 2B, dark gray trace). This observation suggests that the active antennal segment was passively displaced at most ~3 deg apart at higher wind speeds. In live flies, however, active movements typically exceed the passive deflection that would be created by aerodynamic forces, and these movements vary as a function of wind speed. Indeed, as previously reported, flies tended to produce more active movements at the higher wind speeds compared to low (Figure S1E–G).

Figure 2. Airflow-dependent active positioning of the antennae is reversed in the presence of a static grating.

A) Example single-trial traces of the aristal (orange) and pedicellar (blue) antennal movements for a dead fly (left) and a nonflying, living fly (right) in darkness; gradient bar indicates wind speed increases every six sec. B) Average active antennal positioning over time as wind speed increases for dead (light grey) and live, nonflying flies (dark grey). All movements are baseline-subtracted relative to the average angle at 0 cm/s wind speed. C) Average pedicellar (active) and aristal segment inter-antennal angles, baseline-subtracted from 0 cm/s wind speed, for flies in darkness and with a static grating. D) Active inter-antennal angle across wind speeds for nonflying flies in darkness (purple) and with a static grating (green). Left: absolute inter-antennal angle at 0 cm/s wind speed (p=.34). Right: baseline-subtracted (i.e., relative to 0 cm/s wind speed) inter-antennal angle across wind speeds in both conditions. Flies in darkness moved their antennae towards the midline as wind speed increased (β=-.016, p<.001), while flies with a static grating moved theirs away from the midline (β=.045, p<.001). E) Similar to (D), for flying flies. As wind speed increased, flies in darkness positioned their antennae medially (β=-.009, p<.01), while those with a static grating positioned their antennae laterally (β=.035, p<.001). F) Average wingbeat frequency for both conditions. Flies with a static grating had a higher wingbeat frequency across wind speeds (all p<.05). G) Average relative wingbeat amplitude decreased as wind speed increased (see Figure S2). Wingbeat amplitude decreased more for flies with a static grating (β=-1.97e-3, p<.001) and was significantly lower at and above 100 cm/s wind speed (all p<.05). In subfigures B, C ,F, and G the shaded regions represent 95% CI. In subfigures C-G, the number of flies per wind speed can vary, so the range of flies is displayed on the figure. In subfigures D-G, statistical comparisons originate from model-derived estimated marginal means. See also Figures S1–S3, Tables S1, S5–S6.

Airflow-dependent active antennal positioning response reverses in the presence of a static grating

To determine how the visual environment influences wind-induced active movements, we presented flies with a static grating of black and white bars equidistant on their left and right (Figures 1A and 2C). In darkness, we observed that both nonflying and flying flies decreased their active inter-antennal angle as wind speeds increased (Figure 2Ci, blue traces). However, in the presence of a static grating, this trend reversed; active segment inter-antennal angle decreased slightly at the lowest tested airflow (50 cm/s) before increasing with higher wind speeds (Figure 2Cii, blue traces). Furthermore, this alteration of wind speed-dependent active positioning directly affected the positioning of the aristae; flies in the presence of a static grating exhibited greater aristal inter-antennal angle increases with higher airflow compared to flies in darkness (Figure 2C, orange traces).

In the absence of wind, we observed a small but statistically nonsignificant decrease in absolute active antennal position for flies in darkness compared to those presented with a static grating; this was true for both nonflying (Figure 2D, left; p=.34, t-test) and flying flies (Figure 2E, left; p=.70, t-test). Upon exposure to airflow, nonflying flies in darkness positioned their antennae medially (towards the midline; Figure 2D, right; p<.001, linear-mixed model). This trend reversed in flies presented with a static grating, which positioned their antennae laterally (away from the midline) with increasing wind speed (p<.001). Moreover, flies in the presence of a static grating positioned their antennae laterally with increased wind speed at approximately twice the rate (+1.45 deg per 50 cm/s wind speed) flies in darkness positioned their antennae medially (-0.80 deg per 50 cm/s). Post-hoc comparisons revealed significant differences in relative (i.e., baseline-subtracted, relative to the 0 cm/s condition) inter-antennal angle between the two visual conditions at all wind speeds above 0 cm/s (all p<.01, Holm-corrected). In flying flies, we observed a similar trend in antennal positioning as in nonflying flies: flies in darkness positioned their antennae medially as wind speeds increased (Figure 2E; p<.01, linear-mixed model) whereas flies with a static grating positioned their antennae laterally (p<.001). These differences in relative inter-antennal angle between the two visual conditions were statistically significant at and above 150 cm/s wind speed (all p<.01, Holm-corrected). Because deflections of the passive antennal joint (between the pedicel and arista) activate populations of sensory Johnston’s Organ Neurons (JONs), we compared the magnitude of this angle across windspeeds for the two visual conditions to determine how these movements may alter sensation (Figure S1). We found that in the presence of a static grating, flies exhibit active lateral antennal movements with increasing wind, while flies in the darkness exhibit active medial antennal movements (Figure 2D–E). These movements effectively tune the gain of the passive joint angle, increasing the deflection of the passive joint in the presence of a static grating relative to in darkness for both nonflying and flying flies (Figure S1H and S1I; all p<.001, Holm-corrected).

Airflow-dependent wingbeat dynamics are modulated by the visual environment

Previous studies demonstrate that active antennal movements occur during flight^{31,39,67,76–79}, so we next examined how wing dynamics changed in response to the same stimuli that elicit airflow-dependent antennal movements. Overall, we found that flying flies exhibited a variety of wing dynamics depending on both wind speed and the visual environment (Figures 2F and 2G). Flies in darkness decreased their wingbeat frequency as wind speed increased up to 200 cm/s (Figure 2F; p<.001, linear-mixed model). However, flies in the presence of a static visual grating exhibited a subtle increase in wingbeat frequency with higher wind speeds (p<.001 compared to flies in darkness, linear-mixed model). Additionally, flies presented with a static grating had a higher wingbeat frequency than those in darkness across all wind speeds (all p<.05, Holm-corrected). We also observed a decrease in relative (i.e., baseline-subtracted) wingbeat amplitude as wind speed increased in both visual conditions (in darkness; p=.006, linear-mixed model), although flies in the presence of a static grating exhibited a larger, more consistent negative trend (p<.001). There was a significant difference in relative wingbeat amplitude between visual conditions at and above 100 cm/s wind speed (all p<.05, Holm-corrected). We validated our measure of wingbeat amplitude in a separate experiment (Figure S2, STAR Methods). Further, we compared responses in flies presented with a static grating versus a featureless visual stimulus (a grey screen of the same luminosity as the static grating) and found that antennal movements were similar (Figure S3). Although wing dynamics could be affected by higher airspeeds due to drag, we do not observe consistent changes in wingbeat frequency or amplitude that correlate with these highest speeds across conditions (Figure 2F–G), so we conclude that these airflow-dependent changes are primarily initiated by sensory input to the wing motor system.

Relative active antennal positioning across wind speeds is flight-state invariant

Flying insects actively hold and maintain their antennae in a more medial position while flying^{31,73,76,79,80}. To determine whether airflow-dependent active antennal movements in Drosophila change during this steady-state flight positioning towards the midline, we directly compared the responses of nonflying and flying flies to varying wind speed both in darkness and the presence of a static grating (Figure 3). To quantify this movement, we measured the average position of the arista in nonflying and flying flies (Figure 3A and 3B). We found that the inter-antennal angle of the arista was 121.1 ± 4.8 deg and 97.6 ± 6.4 deg (mean ± SD) in nonflying and flying flies, respectively. Because we observed no significant differences in inter-antennal angle between visual conditions in both flight states without airflow (Figure 2D–E, left), we combined data from both visual conditions for this analysis.

Figure 3. Active frontal antennal positioning at the onset of flight is independent of airflow-dependent active movements.

A) Dorsal view of the fly’s head during nonflight (left) and flight (right), showing frontal antennal movement at flight onset. B) Schematic of frontal positioning from (A), with average aristae inter-antennal angle ± standard deviation for nonflying (left) and flying (right) flies (N=52 flies). C, i) Active antennal positioning across wind speeds for flies in darkness (see also Figure 2D–E, purple). Left shows absolute active inter-antennal angle for nonflying (yellow) and flying (red) flies (mean difference = 22.1 deg, p<.001). Right shows baseline-subtracted (i.e., relative to 0 cm/s wind speed) active inter-antennal angle across wind speeds. Wind speed-dependent antennal responses did not change in flight, despite medial positioning at flight onset (β=7.06e-3, p=.13). C, ii) Similar to (C, i), but flies are in the presence of a static grating (see also Figure 2D–E, green). Left shows absolute active inter-antennal angles (mean difference = 20.5 deg, p<.001). Wind speed-dependent antennal responses did not change in flight (β=-3.11e-3, p=.42). In subfigure C, the number of flies per wind speed varies, range of flies displayed on figure. Data displayed in subfigure C is from the same flies also presented in Figure 2D–E under different comparisons. Statistical comparisons across wind speeds originate from model-derived estimated marginal means. See also Table S2.

In the absence of wind, active (pedicellar) segment inter-antennal angle was significantly lower during flight in both darkness and with a static grating (Figure 3C, middle; both p<.001, paired t-tests). However, flight state did not affect relative active antennal positioning across wind speeds in darkness (Figure 3Ci, right; p=.13, linear-mixed model) or in the presence of a static grating (Figure 3Cii, right; p=.42). Furthermore, we observed no significant differences in relative inter-antennal angles between nonflying and flying flies across all wind speeds for both visual conditions (all p>.05, Holm-corrected). These findings suggest that airflow-dependent active movements of the antennae are not substantially altered during flight.

Silencing antennal sensory neurons shifts active antennal positioning more frontally at flight onset

Our findings thus far indicated that wind speed influences active positioning of the antennae and wingbeat dynamics. To better understand the sensory influence on antennal movements, we measured antennal positioning responses to airflow in flies with reduced sensory input from the antennae by expressing GtACR1⁸¹, an inhibitory channelrhodopsin, in all Johnston’s Organ Neurons (JONs; nan-GAL4>UAS-GtACR1)^82,83 (Figure 4A, yellow, ‘inactivated’). As an alternate method of abolishing JON activity, we glued the passive joint (pedicel-funiculus joint) of both antennae (Figure 4A, blue, ‘glued’); this manipulation prevents the deflection of the arista in response to wind, effectively silencing JON activity^75,84. To control for the presence of cyan light in our optogenetic inactivation experiments, we presented intact and glued flies with the same light stimulus used to invoke optogenetic silencing in the inactivated group.

Figure 4. Blocking antennal sensation disrupts active frontal positioning of the antennae.

A) Schematic of the D. melanogaster antenna highlighting the pedicel (active segment) and passively deflected arista. The arista is rigidly attached to the funiculus. In airflow, the funiculus and arista passively rotate (curved arrow), activating stretch receptive Johnston’s Organ Neurons (JONs). JON activity is silenced optogenetically (nan-GAL4>UAS-GtACR1, yellow) or mechanically by gluing the passive joint (blue). B) Absolute active segment inter-antennal angle in flies with intact JONs (grey), optogenetically inactivated JONs (yellow), and mechanically silenced JONs (blue) in no airflow. Optogenetic inactivation resulted in a significantly larger medial antennal movement (β=-9.26, p<.001) while gluing did not (β=-2.06, p=.40). C) Absolute active inter-antennal angle across all wind speeds in nonflying flies. No differences were found between conditions at all wind speeds (all p>.05). D) Same as (C), but in flying flies. Optogenetic inactivation resulted in significantly more medial antennal positioning across all wind speeds (all p<.01). Gluing the passive joint resulted in more medially positioned antennae above 200 cm/s wind speed (all p<.05). E) Wingbeat frequency varies across conditions and wind speeds. In the absence of wind, flies with inactivated JONs have a significantly lower wingbeat frequency (p<.001). Wingbeat frequency decreases with wind speed in flies with intact JONs (β=-.0318, p<.01) but increases with wind speed in flies with both inactivated JONs (β=.0478, p<.01) and glued JONs (β=.103, p<.001). Wingbeat frequency is significantly higher than control in flies with inactivated JONs at and above 100 cm/s (all p<.05) wind speed and in flies with glued JONs at all wind speeds (all p<.01). F) Wingbeat amplitude decreases with wind speed in all conditions. Glued passive joints result in a steeper decline (β=-6.61e-4, p=.012), but post-hoc comparisons revealed no differences across conditions (all p>.05). In subfigures E-F, the shaded regions represent 95% CI. In subfigures B-F, the number of flies per wind speed varies, so the range of flies plotted is displayed on the figure. In subfigures B-F, statistical comparisons originate from model-derived estimated marginal means. See also Figure S2 and Table S3.

We first investigated active movements across conditions in the absence of airflow to determine if JON silencing affects the active frontal antennal movement at flight onset. In the light control group (Canton-S>UAS-GtACR1) with no frontal airflow, we observed medial positioning of the antennae (-24.2 deg, p<.001, linear-mixed model) at flight onset (Figure 4B, grey, ‘intact), similar in magnitude to movements observed in wild-type flies (-22.1 deg and -20.5 deg, Figure 3C, middle). However, in flies with optogenetically inactivated JONs, we observed significantly greater medial positioning of the antennae (-34.4 deg, Figure 4B, yellow; p<.001, linear-mixed model) during flight. Likewise, we measured a similar increase in medial antennal positioning in flies with glued active passive joints (-31.1 deg, Figure 4B, blue), although this relationship was not significant when compared to intact flies (p=.40, linear-mixed model).

The above findings demonstrate that in the absence of wind, blocking antennal sensation alters the medial positioning of the antennae at the onset of flight. To understand how reduced mechanosensation affected antennal movements in the presence of airflow stimuli, we next investigated active antennal positioning in response to a range of wind speeds with silenced JONs. In nonflying flies, we observed an increase in active inter-antennal angle as wind speed increased in all conditions (Figure 4C; all p<.001, linear-mixed model). This wind speed-dependent trend had a significantly higher slope in flies with optogenetically (+1.64 deg per 50 cm/s wind speed) and mechanically (+1.42 deg per 50 cm/s) silenced JONs relative to flies with intact JONs (+0.44 deg per 50 cm/s). However, we did not observe any significant differences in absolute antennal position between flies with intact JONs and flies in either silencing condition at any single wind speed (all p>.05, Holm-corrected).

In flying flies, active inter-antennal angle also increased at higher wind speeds in all conditions (Figure 4D). However, flies with glued passive joints displayed slower lateral movement with increasing speed compared to flies with intact and inactivated JONs (p=.023, linear-mixed model). Notably, we observed further medial positioning of the antennae when JONs were silenced (Figure 4D), similar to the response in the absence of airflow (Figure 4B, blue). Flies exhibited this narrower inter antennal angle all wind speeds for flies with optogenetically inactivated JONs (all p<.01, Holm-corrected) and at higher wind speeds for flies with mechanically silenced JONs (all p<.05). Together, these results suggest that JON sensation contributes to the antennal set position during flight, independent of wind speed.

Complex effects of antennal mechanosensation manipulation on flight control

To determine how altered antennal mechanosensation affected flight control, we investigated wing dynamics across wind speeds in flies with silenced JONs. Unexpectedly, we found that flies with inactivated JONs demonstrate lower wingbeat frequency than control flies and flies with glued passive antennal joints in the absence of wind (Figure 4E, left; p<.001, Holm-corrected). As wind speed increased, however, flies with intact JONs decreased their wingbeat frequency, flies with inactivated JONs held their wingbeat frequency relatively constant, and flies with glued passive antennal joints rapidly increased their wingbeat frequency. Flies with inactivated JONs had significantly higher wingbeat frequency than flies with intact JONs at and above 100 cm/s wind speed (Figure 4E, yellow; p<.05, Holm-corrected), and flies with glued passive antennal joints had significantly higher wingbeat frequency at all wind speeds (Figure 4E, blue; all p<.01). In contrast, we observed a similar decreasing trend in relative wingbeat amplitude as wind speeds increased across all three experimental groups here (Figure 4F). Although we found a steeper decrease in wingbeat amplitude as wind speed increased in flies with glued antennal joints relative to flies with intact JONs (Figure 4F, blue; p=.012, linear-mixed model), we observed no differences between flies with intact JONs and flies with either silencing manipulation at any wind speed (all p>.05, Holm-corrected).

Flies actively position their antennae in response to varying optic flow speed during flight

To understand how progressive visual information affects antennal positioning in the absence of frontal airflow, we presented flies with a vertical bar pattern moving at rates of 0–50 cm/s with no airflow (Figure 5A, STAR Methods). In flying flies, we observed an active movement of the antennae away from the midline at a high (35 cm/s) optic flow speed compared to a low optic flow speed (5 cm/s), but in nonflying flies, antennal angle was approximately unchanged over this speed range (Figure 5B). When comparing no optic flow (0 cm/s) to the lowest optic flow speed we tested (5 cm/s), we saw a significant decrease in active inter-antennal angle in flying flies but not nonflying flies (Figure 5C; p<.05 for flying flies, paired t-test). To measure optic-flow-speed-dependent changes in antennal responses, we quantified movements of the antennae relative to the lowest optic flow speed (5 cm/s). Nonflying flies did not significantly change their antennal position at any optic speed except 35 cm/s (Figure 5D, left; p<.05 at 35 cm/s, Holm-corrected one-sample t-tests). However, in flying flies, we observed clear active movements of the antennae laterally with increasing optic flow (Figure 5D, right; p<.001, linear-mixed model), and baseline-relative antennal position was significantly different from zero at all optic flow speeds (all p<.05, Holm-corrected one-sample t-tests).

Figure 5. Flying flies increase inter-antennal angle with increasing optic flow rate.

A) Schematic of the optic flow experiment. The airflow tube was present with no airflow delivered. B) Average active (pedicellar) inter-antennal angle in response to progressive optic flow moving at 5 cm/s (light blue) and 35 cm/s (dark blue) for an example nonflying (top) and flying (bottom) fly for a single 8 sec trial. Shaded traces represent the average across 5 trials; transparent traces represent individual trials. C) Absolute active inter-antennal angle at 0 cm/s and 5 cm/s optic flow in nonflying flies (left, p=.22) and flying flies (right, p=.014). Average across all flies is represented by shaded traces, while individual flies are represented by faded traces. D) Average individual (grey) and cross-fly (black) active inter-antennal angle across all optic flow speeds (excluding zero), baseline-subtracted relative to the lowest speed (5 cm/s) for nonflying (left) and flying (right) flies. The active antennal response of flying flies is significantly larger than nonflying flies (β=.119, p<.001) and greater than zero at all non-baseline optic flow speeds (all p<.05). E) Wingbeat frequency across all optic flow speeds. F) Wingbeat amplitude increases as optic flow speed increases. In subfigures E-F, the shaded regions represent 95% CI. In subfigures C-F, the number of flies per wind speed varies, so the range of flies plotted is displayed on the figure. In subfigure D, statistical comparisons originate from model-derived estimated marginal means. See also Figure S2 and Table S4.

Across optic flow speeds, we did not observe substantial changes in wingbeat frequency (Figure 5E). However, we found that wingbeat amplitude grew slightly with increasing optic flow (Figure 5F). This contrasts with our observations in flies presented with only varying airflow, where we found that wingbeat amplitude consistently decreased as airflow increased (Figure 2G and 4F). These results suggest that steady-state differences in airflow and progressive optic flow have inverted effects on wingbeat amplitude in open-loop (tethered flight), with faster optic flow driving increases in wingbeat amplitude and faster airflow decreasing wingbeat amplitude.

Frequency-dependent responses to dynamically changing airflow and optic flow

To understand how mechanosensory and visual information are combined to influence active sensation by the antennae, we presented both wind and progressive optic flow stimuli synchronously (Figure S4, STAR Methods). We delivered wind and optic flow as sinusoidal stimuli at three frequencies (0.3, 1.3, and 2.3 Hz), oscillating around a set baseline value simulating forward flight. In natural settings, optic flow speed depends strongly on the geometry of the visual environment due to motion parallax (i.e., distant objects traverse the retina more slowly than objects closer to the observer), and thus there is no simple one-to-one correspondence between optic-flow speed and wind speed. Therefore, we measured steady state antennal responses to wind speed (Figure S6A) and optic flow (Figure S6B) to determine oscillatory ranges (100–200 cm/s for wind, 5–35 cm/s for visual) and baseline values (150 cm/s wind, 20 cm/s visual) for both oscillating stimuli. We presented one stimulus oscillating from its lowest to highest range in a sinusoidal pattern while holding the other modality at its baseline value (and in the opposite manner with the second stimulus). Additionally, we presented both wind and visual stimuli oscillating simultaneously within their respective ranges.

To confirm we were not saturating the antennal motor system with our stimulus (i.e. active antennal movements reached their maximum range), we first presented each stimulus type at half amplitude (125–175 cm/s for wind, 12.5–27.5 cm/s for visual) and full amplitude (100–200 cm/s for wind, 5–35 cm/s for visual) (Figure S6C). We observed that wind oscillation-dependent movements scaled approximately linearly across oscillation frequencies in nonflying flies (Figure S6D, blue), while visual oscillation and wind plus visual oscillation-dependent movements exhibit nonlinear properties (however we note that visual responses were, in general, low in nonflying flies; Figure S6D, red and purple). For flying flies, all three stimulus combinations elicited mildly nonlinear responses (and highly variable responses for vision-only), but generally scaled (roughly) by a factor of 2, indicating that the stimuli were not saturating the motor response (Figure S6E).

In response to oscillating airflow and optic flow, we observed changes in active antennal responses in both nonflying and flying flies that depended on the frequency of oscillation and the sensory modality that was oscillating (Figure 6A). Antennal motions in nonflying flies were smaller across all oscillation frequencies and oscillatory conditions than in flying flies (compare gains in Figure 6B vs. 6D). This was especially true when optic flow speed alone was oscillating, with nonflying flies exhibiting near-zero active movements (Figure 6A and 6B, red). Notably, these movements in response to optic flow oscillations also lagged in phase when compared to the other stimulus conditions (Figure 6A and 6C, red).

Figure 6. Active antennal responses to oscillating combinations of airflow and optic flow reveal flight-gated nonlinear integration.

A) Average baseline-subtracted active (pedicellar) inter-antennal angle over a 6 sec period in response to oscillating wind speed (with constant 20 cm/s optic flow, blue, N=14–15 flies), oscillating optic flow (with constant 150 cm/s airflow, red, N=15 flies), or synchronous oscillation of wind speed and optic flow (purple, N=15 flies) for nonflying (upper middle) and flying (bottom) flies. Black lines (top) represent time-dependent location of stimulus oscillation on an arbitrary y-scale. B) Gain of active antennal movements across oscillatory conditions in nonflying flies. Error bars are generated by a 100-epoch Monte Carlo estimation with replacement at each frequency. C) Phase of active antennal movements relative to time-dependent phase of stimuli presentation (black, top of (A)). Error bars are generated by same method as (B). D) Same as (B), in flying flies. E) Same as (C), in flying flies. F) Wingbeat frequency response (upper middle) and baseline-subtracted wingbeat amplitude (bottom) relative to oscillating stimuli (black, same as (A)) in flying flies. Dataset originates from flying flies in (A). G) Gain of wingbeat frequency response across oscillatory conditions. Error bars are generated by a 100-epoch Monte Carlo estimation with replacement at each frequency. H) Phase of wingbeat frequency response relative to time-dependent phase of stimuli presentation (black, top of (F)). I) Gain of baseline-subtracted wingbeat amplitude across stimulus oscillation frequencies. J) Phase of baseline-subtracted wingbeat amplitude across stimulus oscillation frequencies. In subfigures A and F, the shaded regions represent one standard error of the mean. Range of trials per oscillatory condition shown on right. See also Figures S2 and S4–S7.

Flying flies performed larger antennal movements than nonflying flies, and these movements depended strongly on the oscillation frequency (Figure 6A, bottom). At 0.3 Hz, the gain (Figure 6D) and phase (Figure 6E) of visually driven movements (Figure 6Ai, bottom, red) were nearly identical to that of the responses to the combined (wind and visual) cue (Figure 6Ai, bottom, purple). In contrast, at 2.3 Hz, the visually driven movements were much smaller (Figure 6D, red) and wind-oscillation-dependent movements (Figure 6Aiii, bottom, blue) were nearly identical to the responses to the combined cue (Figure 6Aiii, bottom, purple). As oscillation frequency increased, the phase lag (Figure 6E) of wind-driven dependent movements remained steadily antiphase (blue), suggesting a flat mechanosensory gain of -1, while visually driven (red) and wind- plus visually driven (purple) movements lagged slightly in phase at the lowest frequency, and trended towards antiphase. Critically, these results indicate that for tethered, flying flies, low-frequency movements were dominated by vision, whereas high-frequency antennal movements by mechanosensation, crossing over at the central frequency we tested (see also Figure S7). In contrast, for nonflying flies, antennal responses followed mechanosensory cues across all frequencies of sensory oscillations.

We also observed differences in both wingbeat frequency and wingbeat amplitude depending on the type of oscillating stimuli and oscillation frequency (Figure 6F). The gain (Figure 6G) and phase (Figure 6H) of wingbeat frequency oscillations closely matched each other in the wind oscillation (blue) and wind- plus visual- oscillation (purple) conditions, with a significant decrease at 2.3 Hz. However, the wind- plus visual- oscillation condition (purple) tonically increased wingbeat frequency at all oscillation frequencies compared to wind-only (blue) oscillations (Figure 6F, ii); we note that this tonic increase does not manifest as a change in the frequency response gain or phase (Figure 6G). With oscillating visual stimuli, the magnitude of wingbeat frequency changes is near zero, and phase lag increases dramatically as oscillation frequency increases (Figure 6G–H, red).

Similarly, we measured minimal changes in wingbeat amplitude in response to oscillating visual stimuli (Figure 6I, red) with phase lag increasing significantly as oscillation frequency increases (Figure 6J, red). In the wind-only (blue) and the wind- plus visual- (purple) oscillation conditions, we observed similar oscillations in wingbeat amplitude responses (Figure 6I). The phase lag of these two conditions mirrors that of wingbeat frequency, albeit in antiphase.

DISCUSSION

Visual environment modulates active antennal sensing

In our study, we investigated how airflow and wide-field visual stimuli are integrated to control the movements of (and by extension, sensation by) the multimodal antennae of Drosophila melanogaster. We observed more frontal positioning of the antennae as frontal airflow increased in darkness (Figure 2D), similar to other flying insects such as hawkmoths (D. nerii)⁸⁵ and honeybees (A. mellifera)⁷⁷. However, the fly’s response to wind was largely inverted in the presence of a static visual grating, with antennal position shifting backwards with increasing frontal airflow (Figure 2E). These inversions in behavioral responses to broad changes in visual environment are similar to previously reported work. For instance, in a closed-loop tethered experiment⁸⁶, flies oriented downwind in response to airflow in an enclosed, dark environment. In a separate closed-loop tethered study, flies instead exhibited an inverted response, orienting upwind in the presence of airflow with some ambient light present⁵⁶. Not surprisingly, light levels profoundly affect behavior in other insect species like the honeybee Apis mellifera, which halt flight in dark conditions⁸⁷. Our results support the conclusion that both the presence and content of the visual environment strongly influence flight behavior. Notably, the presence of progressive visual motion resulted in an opposite wingbeat amplitude response compared to a static visual environment. Furthermore, the active antennal movements increase the gain of the passive joint deflection as a function of windspeed, suggesting that the animals are tuning antennal sensitivity (Figure 2 and S1).

Why flies (or any animal) alter their active sensing strategy based on the visual environment, remains a key question. We hypothesize that active sensing behaviors enable the animal to glean additional information from one sensory modality when information from another is diminished or entirely absent. Our results appear to fit into a widespread phenomenon in which animals alter sensory acquisition strategies in one modality in the face of reduced sensory input in another. For example, electric fish change movements of their entire body movements in the dark, but also begin actively repositioning their tail (where the electric organ is) in this condition^88,89. Work in humans similarly shows that haptic search strategies depend strongly on vision⁸⁸. Further, vision alters muscle spindle responses and movement-amplitude discrimination, providing direct evidence that visual context can tune muscle spindle sensitivity⁹⁰. The present study shows that active antennal movements are driven by a combination of visual and mechanosensory inputs, and future studies aimed at directly measuring the sensory consequence of these movements will provide valuable insight into mechanisms of cross-modal interactions during active sensing.

In response to visual motion, we consistently observed an increase in the magnitude of antennal movements in flying flies compared to nonflying flies. This was true with steady-state optic flow in the absence of airflow, where only flying flies performed significant visually driven antennal movements (Figure 5). Additionally, we observed an increased gain across all oscillatory sensory conditions in flying flies as compared to their nonflying counterparts (Figure 6A–E). However, we did not observe flight-dependent changes in antennal responses to airflow in different visual environments, in the absence of visual motion (Figure 3C). This suggests that increased magnitude of antennal movements during flight requires the presence of visual motion. One potential explanation for this phenomenon is the direct modulation of upstream visual motion circuitry during flight by octopaminergic neurons^91–93, which might increase the salience of visual information integrated by antennal motor circuits. It is also possible that a nonlinear interaction between visual motion information and an internal representation of flight in premotor antennal circuits exists; evidence that visual input modulates antennal control circuitry in other flying insects, such as the hawkmoth (D. nerii)⁹⁴ and honeybee (A. mellifera)⁷⁷, may support this explanation. In this scenario, reafferent feedback from wingbeats detected by Johnston’s Organ Neurons (JONs)^54,75,95 could be integrated directly by premotor circuitry. Additionally, flight state information from ascending neurons may directly convey behavioral states to central brain premotor regions, similar to those found in grooming and walking behaviors⁹⁶. Further investigation through connectomics analyses and in vivo physiology in behaving animals may enable us to discern how visual motion information is relayed to antennal motor circuits in a state-dependent manner.

Control of antennal position at the onset of flight

In our study, we observed a steady-state frontal positioning of the antennae during flight, consistent with previous work^{31,73,76,79,80} (Figure 3A–B, 4B–D). This flight-state-dependent positioning was invariant between flies in darkness and those presented with a static grating (Figure 2E, left) in the absence of airflow, which is surprising based on our observation that they perform different antennal movements as airflow increases (Figure 2E, right). In other words, the visual context seems to set a baseline active antennal posture but inverts the wind-speed-dependent positioning relative to that baseline posture. We also found that silencing JON input both mechanically and optogenetically shifted this flight-state-dependent positioning to a more extreme frontal position across all wind speeds (Figure 4B, 4D). This result is in contrast to previous work in hawkmoths, where mechanical silencing produced no difference in flight-based forward antennal positioning⁷³. Rather, in this previous study, the Böhm’s bristles (hairs positioned between cuticular segments thought to aid in proprioception) controlled flight-based positioning of the antennae. Additionally, silencing hawkmoth and honeybee JON activity with glue halts wind speed-dependent antennal movements^77,85. Surprisingly, our results do not support the conclusion that airflow-dependent changes to antennal positioning in flight is mediated primarily by JONs (Figure 4D). However, our finding that wind-speed-dependent responses persist in flies with silenced JONs (mechanically and optogenetically) suggests that a sensor analogous to Böhm’s bristles, such has hair plates detecting relative position between the scape and pedicel (such as those noted in Eichler et al. 2024⁹⁷, and possibly similar to those recently described in Drosophila legs⁹⁸), may contribute to antennal positioning in Drosophila. Nonetheless, we observed a steady dysregulation of medial antennal positioning in flies with silenced JONs, largely independent of wind speed (Figure 4B), suggesting that a feedforward motor program may be involved in this behavior.

Unlike with antennal movements, whose airflow-dependent responses appear to be independent of JON sensation, we observed a clear effect of blocking antennal sensation on wind speed-dependent wingbeat frequency (Figure 4E). Markedly, blocking antennal sensation significantly increased wingbeat frequency at higher windspeeds, and we observed similar trends as a function of windspeed in both manipulations. However, this effect was somewhat more pronounced in flies with mechanically blocked JONs. This difference may stem from the genetic driver line (nan-GAL4) we used that labels all chordotonal organs^82,83, including stretch-receptive sensors located in the femur, haltere, and wings. Here, it is important to note that silencing all chordotonal cells across the body could influence behavior in ways that are difficult to measure or interpret in this manipulation. Silencing these additional sensory cells outside the antenna, particularly in the wings, may explain why we observe an initial decrease in wingbeat frequency in genetically silenced flies in the absence of wind (Figure 4E, left). However, the residual difference in the magnitude of the effect of wind speed on the change in wingbeat frequency depending on the manipulation (mechanical versus genetically silenced), could have two reasonable explanations worth investigating in future studies. First, because the sensitivity of any system to a perturbation is often a function of the overall set point, the substantially lower wingbeat frequency in the absence of wind (i.e., the wingbeat frequency set point) that we measured in genetically silenced flies could explain the overall downward shift in the nominal wingbeat frequency compared to flies with glued antennae. A second explanation is that the antenna is likely not the only sensor influencing wingbeat frequency as a function of windspeed, thus overall silencing of chordotonal organs may affect the set point and change the fly’s sensitivity to wind. Overall, the two antennal sensory manipulations (mechanical and genetic silencing) elicit similar trends in antennal response at flight onset and in response to varying airflow, supporting the conclusion that antennal sensation is crucial for guiding these active behaviors.

Mechanisms of visual and mechanosensory integration

One caveat of our findings is that the open-loop tethered preparation does not allow the animal’s intended changes in speed to change the resulting wind flow patterns on the antenna, nor the perceived optic flow. Thus the flight control system is operating in open loop (Figure 1D), and because the visuomotor transform is thought to act like an integrator for forward flight speed control⁵⁸, the small effect of visual oscillations on wing kinematics may arise, at least in part, due to this lack of reafferent feedback, which is known to heavily attenuate sensorimotor responses⁹⁹. Similarly, it is possible that the amplitude of both wing and antennal movements in our experiments could be influenced by head fixation, a condition that alters visually evoked changes in wingbeat amplitude^100–102. Further, active antennal movements during saccades amplify wingbeat-induced oscillations of the antennae and are thought to aid in regulating visually evoked turn magnitude⁶⁷; fixing the head could potentially reduce the animal’s ability to modulate these movements relative to head free flies. However, previous work has shown that wingbeat amplitude responses to frontally expanding stimuli are not altered by head fixation¹⁰², so we suspect that any differences in our experiments, also featuring symmetric stimuli, would be relatively small.

When we provided nonflying flies with oscillatory airflow and optic flow, antennal movements generally mirrored the responses to mechanosensory oscillations alone across all frequencies (Figure 6B and 6C). This broad-band mechanosensory winner-takes-all computation is illustrated in Figure 7A. In contrast, we observed an unexpected nonlinear effect of oscillation frequency on antennal movements in flying flies; as the oscillation frequency increases, antennal movements in response to synchronous airflow and optic flow oscillations (Figure 6D,E, purple) shift from mirroring the antennal response to visual oscillations (Figure 6D,E, red) to mirroring the antennal response to oscillating wind stimuli (Figure 6D,E, blue). This frequency-dependent “winner-takes-all” effect between visual information at lower frequencies and mechanosensory information at higher frequencies (Figure 7B) is similar, for instance, to hawkmoth flight tracking of food source (M. stellatarum)¹⁰³ and head stabilization during body rolls (D. nerii)¹⁰⁴. Additionally, our findings may be analogous to multimodal integration of visual and haltere signals during turning behavior in fruit flies, where slower turns rely on visual information and faster turns use mechanosensory feedback from the angular velocity encoding halteres¹⁰⁵. Mechanical transduction is lower latency than visual transduction making it more suitable for rapid (high frequency) responses. Although the visuomotor response may dominate at low frequencies^58,106, it is also slower than mechanosensory feedback due to phototransduction¹⁰⁷. This makes visual information inherently less useful as motion frequency increases, evidenced by the increasing phase lag we observed (high phase lag feedback can, if not attenuated, destabilize closed-loop dynamics). Thus, the faster mechanosensory pathway may guide appropriate antennal positioning at higher frequencies, as is likely essential for highly variable environments. Altogether, our results demonstrate previously uncharacterized multisensory control principles for active antennal sensation and pave the way towards understanding the precise neural circuits responsible for performing these computations.

Figure 7. Integration of mechanosensory and visual cues for antennal control is nonlinear and state-dependent.

A) Block diagram for nonflying flies, following a similar format to Figure 1E. Colors indicate sensory pathways. Gray feedback arrow from iaa that passes through the mechanosensory block is modulatory. Mechanosensory cues created by oscillating wind speed elicited robust antennal responses. Time-varying optic flow stimuli, by contrast, elicited weak antennal responses that were overridden by responses to mechanosensory cues when both were present. B) A simplified control theoretic model illustrates the potential role of active antennal movements during forward flight. Same format as (A). Based on prior work⁵⁸, forward flight speed control is thought to be well approximated by the linear summation of a proportional, lower-latency mechanosensory response and an integral, longer-latency visual response (lower multisensory feedback loop in the diagram). Our work shows that antennal movements were regulated in a frequency-dependent, “winner-takes-all” manner, where low-frequency antennal motions are dominated by visual feedback and high-frequency antennal motions are dominated by mechanosensory feedback (upper multisensory feedback loop). We hypothesize that this active sensing control, illustrated by the diagonal arrow through the mechanosensory response block, is modulatory in nature, altering antennal neuromechanical processing of wind speed. This suggests that the dynamics of forward flight control emerge from a neuromechanical system that includes inner-loop sensor modulation.

RESOURCE AVAILABILITY

Lead contact:

Any requests for resources or further information should be directed to and will be fulfilled by the lead contact, Marie P. Suver (marie.suver@vanderbilt.edu).

Materials availability:

No new materials or reagents were generated in this study.

Data and code availability:

Data and code generated in this study are available on Zenodo (https://doi.org/10.5281/zenodo.17535569), and all code is additionally available on Github (https://github.com/suverlab/MillsEtAl2025). Additional raw data will be made available upon request.

STAR METHODS

EXPERIMENTAL MODEL AND STUDY PARTICIPANT DETAILS

All experiments used adult female Drosophila melanogaster between 3 to 7 days post-eclosion. All flies were raised on standard yeast-cornmeal medium at 25° C on a 12-hour light-dark cycle. Experimental genotypes for each figure are listed in the table below. Parental genotypes and RRIDs are available in the Key Resource Table.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Chemicals, peptides, and recombinant proteins
all-trans retinal	Sigma-Aldrich	Cat #R2500
Deposited data
Raw and analyzed data	This paper	https://doi.org/10.5281/zenodo.17535569
Experimental models: Organisms/strains
D. melanogaster: Canton-S	Dickinson Lab	N/A
D. melanogaster: w[*]; P{y[+t7.7] w[+mC]=UAS-GtACR1.d.EYFP}attP2 (UAS-GtACR1)	Bloomington Drosophila Stock Center	RRID: BDSC_92983
D. melanogaster: w[*]; P{w[+mC]=nan-GAL4.K}2 (nan-GAL4)	Bloomington Drosophila Stock Center	RRID: BDSC_24903
Software and algorithms
MATLAB version 2023b	MathWorks	RRID: SCR_001622
Psychophysics Toolbox	https://psychtoolbox.org/	RRID: SCR_002881
Python version 3.10.11	Python Software Foundation	https://www.python.org/downloads/release/python-31011/
seaborn	https://seaborn.pydata.org/	RRID: SCR_018132
SciPy	http://www.scipy.org/	RRID:SCR_008058
statsmodels	http://www.statsmodels.org/	RRID:SCR_016074
emd	https://emd.readthedocs.io/en/stable/	N/A
Custom code for project	This paper	https://github.com/suverlab/MillsEtAl2025

Fly genotypes used in experiments

Figure 2	Canton-S
Figure 3	Canton-S
Figure 4	Canton-S>UAS-GtACR1; nan-GAL4>UAS-GtACR1
Figure 5	Canton-S
Figure 6	Canton-S
Figure S1	Canton-S
Figure S2	Canton-S
Figure S3	Canton-S
Figure S6	Canton-S
Figure S7	Canton-S

METHOD DETAILS

Behavioral apparatus

To present airflow, humidified air was sent through a tube (McMaster 89895K657) positioned approximately 5 cm anterior to the fly. Wind speed was controlled by a mass flow controller (Dakota Instruments 6AGC1AL55–08AB) and toggled on and off by a solenoid valve (Lee Company LHDA1233115H) controlled by a custom MATLAB script. To measure and calibrate wind speed, we used a calibrated hot wire anemometer (Dantec MiniCTA with 55P11 probe). Additionally, we used a manual wind puffer to elicit flight, with an output tube positioned dorsally to the primary airflow tube. Puff occurrences were measured by an airflow sensor (Honeywell Sensing Solutions AWM3300V).

To block external light, we created a 24”x24” lightproof box with foamcore walls and ceiling mounted on a 1/4”-20 breadboard (Thorlabs MB2424). To present unified optic flow, two 308.448 x 93.43 mm (2880 x 864 pixels) 120 Hz LCD screens were placed parallel to the fly on both left and right side at a distance of 87 mm. Visual stimuli presented were bilaterally and vertically symmetric, so for practical purposes the right screen displayed a vertically flipped view of the left screen. All vertical bars within the visual gratings were 35.343 mm across (330 pixels), totaling approximately 22 degrees within the fly’s field of vision when an edge is directly adjacent to the fly. Reflections of light on the LCD screens were minimized with custom-fitting anti-glare screen covers.

We created all visual stimuli in PsychoPy, and all visual stimuli were presented using Psychophysics Toolbox in a secondary MATLAB instance connected to the primary MATLAB instance through TCP/IP protocol. To synchronize visual stimuli and wind stimuli presentation, we built a custom photodiode-based circuit that would trigger the primary instance of MATLAB to both begin data acquisition and initiate the wind stimulus upon detection of a localized, time-synchronized black-white oscillation on the LCD screen out of view of the fly (Figure S4). The photodiode sensor was placed facing a corner of the visual display outside the field of view of the animal; the sensor produced a voltage proportional to the intensity of light detected. When the voltage output from the sensor increased above a threshold of 2.5 V, it results in a trigger signal sent to the computer/data acquisition (DAQ) object, which would instantaneously trigger airflow.

All voltage data was collected and transmitted at a sampling rate of 20 kHz through a 4 AO PCIe card (National Instruments PCIe-6323) attached to two BNC breakout boards (National Instruments BNC-2110).

Experimental design

In the experiments shown in Figure 2–4 and Figure S1–S3, each fly experienced 10 airflow trials either increasing from 0 cm/s to 300 cm/s (5 trials) or decreasing from 300 cm/s to 0 cm/s (5 trials). Airflow was presented in 50 cm/s steps over 42 sec, with 6 sec of steady state airflow for each wind speed. The order of increasing and decreasing trials was pseudo-randomized for each flight. For each wind speed, averages were taken over the last 4 sec as the mass flow controller did not adjust instantaneously. 6 sec of pre-trial at the initial wind speed (with the solenoid valve off, to prevent airflow) was fed to the mass flow controller to allow for adjustment and settlement at the correct speed before the trial began. Additionally, at 0 cm/s wind speed the solenoid valve was shut off to prevent airflow.

In the experiments shown in Figure 5, each fly experienced 60 optic flow trials ranging from 0–50 cm/s in 5 cm/s increments (5 trials at each optic flow speed, pseudo-randomized order). Each trial consisted of an 8 sec period of constant optic flow. For each optic flow speed, averages were taken over the last 6 sec of each trial. The interlude between trial periods consisted of a non-moving static grating, equivalent to the 0 cm/s optic flow condition. The solenoid valve for airflow was shut off for the entirety of these visual stimulus-only experiments.

In the experiments shown in Figure 6, each fly experienced 36 trials at 3 oscillatory frequencies (0.3, 1.3, and 2.3 Hz) with wind speeds ranging from 100–200 cm/s and optic flow speeds from 5–35 cm/s. Trials were presented in 12 blocks, with each frequency represented once pseudo-randomly organized within each block. Wind and visual stimuli were presented synchronously during the experiment. If oscillations were not applied to a stimulus type (wind or visual) in a condition, the non-oscillating stimulus was held at a constant value in the middle of its oscillatory range (150 cm/s for wind, 20 cm/s for visual). Trials were 16 sec in length, with 2 sec before each trial used for synchronization of visual and wind stimuli through the photodiode circuit and 1 sec after each trial used to shut off airflow (via solenoid valve) and the camera recording. Pre-trial and post-trial data was omitted during post-processing. Each trial consisted of 4 sec of baseline presentation, which consisted of both wind and visual stimuli being held at the middle of their respective oscillatory ranges (150 cm/s for wind, 20 cm/s for visual). The following 12 sec consisted of stimuli oscillation or stimuli being held in the middle of its oscillatory range depending on the condition.

To confirm our accurate manipulation of wind speed at high frequencies, we performed directed calibration experiments using the hot wire anemometer (Dantec MiniCTA with probe). The anemometer was placed in the position of the tethered fly during experiments. First, we performed a sweep of all wind speeds in our experimental range (0–300 cm/s in 1 cm/s increments) and collected the average anemometer value at each speed (Figure S5A). Using this, we fit a high-order polynomial to the data (overfitting was not a concern, as we did not extrapolate nor leave the original experimental range) and found the inverse of the polynomial to use as a transfer function for anemometer values to wind speed. When applying the same voltage oscillation (± 1.1 V) from a baseline (~2.49 V, equivalent to 150 cm/s) at 0.1 Hz and 1.1 Hz, a smaller wind speed oscillation amplitude was observed at the higher frequency (Figure S5B). This was observed across frequencies between 0.1 and 1.1 Hz and multiple oscillation voltages (± 0.55 V, ± 1.1 V, and ± 2.2 V), with negative linear relationships between oscillation frequency and oscillation amplitude consistently found (Figure S5C). With the knowledge that oscillation voltage does not equivalently scale oscillation amplitude, we hand-tuned (by adjusting oscillation voltage) oscillation amplitude across various frequencies to be equivalent to ± 50 cm/s (Figure S5D).

We also measured the delay between the command sent to mass flow controller and the arrival of the wind stimulus. To account for this in our stimulus timing, we experimentally measured the phase of the wind stimulus at each frequency. This phase delay was then applied to the visual stimulus so that both wind and visual stimuli would be synchronized when oscillating together.

Fly preparation

Flies were first anesthetized on ice and placed on a metal plate liquid-cooled by ice-cold water flowing through a peristaltic pump. We then dissected all 6 legs at the femoral-trochanter joint to ensure no grooming of antennae or wings would occur over the course of the experiment. Ultraviolet-curing glue (TOPCHASE B08YMTFM7D) was then placed in the gap between head and thorax to fix the head in place during experiments. The rigid tether was then lowered at an 85-degree angle relative to the rostral-caudal axis of the fly and fixed to the mid-thorax using ultraviolet-curing glue. The tethered fly was then placed into the behavioral apparatus at a 60-degree angle relative to ground.

All flies were rigidly tethered and head-fixed using the same protocol, except for dead flies (Figure 2A–B) and those with glued passive antennal joints (Figure 4B–E). Dead flies were frozen at approximately -25° C for 5 minutes before tethering, following a method similar to previous work⁵⁶. We confirmed that the flies did not reawaken after this freezing period. To glue the passive antennal joint, we applied a small drop of ultraviolet-curing glue to the medial edge of the pedicel and funiculus of each antenna⁸⁴. To confirm successful glue application, we nudged each arista with a single hair from a small paintbrush after the glue was cured to verify the pedicel moved in tandem with the funiculus.

Antennal tracking

To record antennal movements, the fly was illuminated from above by infrared light (850 nm) emanating from a 10x2 grid of infrared LEDs (ORDRO LN-3 Studio IR Light Accessory). Antennal movements were measured using a camera (Allied Vision Guppy Pro F-031 with InfiniStix 44mm/3.00x lens) rotated ten degrees backwards from perpendicular to ground. Videos were recorded at 60 frames per sec at a resolution of 640x480 pixels.

We used DeepLabCut^68,69 (version 2.3.6) to track antennal movements over time. The network was trained using 670 frames from 67 flies across genotypes and experimental designs. We tracked 20 points on the head and antennae of the fly, including 6 static points on the head to determine a head axis, 1 on each funiculus, 2 on each arista and 4 on each pedicel. We used a ResNet-50 neural network, trained over 500,000 iterations with a training-to-test fraction of 0.8. The trained network demonstrated a training error of 1.94 pixels and a testing error of 2.24 pixels, respectively.

Wingbeat data collection

To gather information on wingbeat frequency and amplitude, infrared light reflected off the beating wings was collected via a liquid light guide (Thorlabs LLG3–4Z) and fed into a custom circuit board. The signal was smoothed via a second-order bandpass Butterworth filter (cutoff frequencies: 150–250 Hz) followed by a second-order Savitsky-Golay filter (window size: 47 samples). We determined if the fly was flying or quiescent by taking the absolute value of a Hilbert transformation of the processed wingbeat signal followed by a second-order lowpass Butterworth filter (cutoff frequency: 6 Hz); this was then compared to a threshold value. Instantaneous frequency was estimated from the processed wingbeat signal using an amplitude-normalized Hilbert transformation¹¹².

To assess whether our photodiode-based measurement of wingbeat amplitude correlated with true amplitude measured using video recording, we tracked the wing envelope at 60 frames per second using a camera (Allied Vision Guppy Pro F-031 with InfiniStix 94mm/0.45x) mounted above the fly (Figure S2A). For our photodiode-based measurement of instantaneous wingbeat amplitude, we took the maximum amplitude values of the original wingbeat signal over 100 sample (0.05 sec) window periods (Figure S2B). For our camera-based measurement, we tracked the tip and base of the leading edge of the fly’s wings, as well as the midline of the fly using DeepLabCut^68,69 (Figure S2C). We trained a ResNet-50 neural network over 500,000 iterations on 330 frames, sourced from 22 videos from 11 flies (2 per fly). The training-to-test fraction for this network was 0.8. The trained network demonstrated a training error of 2.03 pixels and a testing error of 3.92 pixels, respectively. To compare the camera measurement with our photodiode measurement, we used the combined left- and right- wing angle from leading edge to fly midline (‘wingstroke angle’). We observed a negative linear relationship between wingstroke angle and wingbeat sensor amplitude (Figure S2D, r=-.7754, r²=.6013, p<.001) and accordingly sign-inverted our wingbeat sensor amplitude so that lower photodiode-based, raw measurements display as larger wingbeat amplitudes.

Optogenetic inactivation

All flies used in optogenetic experiments were placed on food saturated with all-trans retinal 24–48 hours prior to tethering. To create the retinal saturated food, 50 µL all-trans retinal (Sigma-Aldrich R2500) was mixed with 0.75 g rehydrated potato flakes and placed on top of standard fly-food mixture. For GtACR1⁸¹ inactivation experiments, we used 505 nm cyan light (Thorlabs, Driver: LEDD1B, LED: M505F3) calibrated at an intensity of 7.00 mW/cm². Light was delivered by a raw fiber optic patch cable (Thorlabs M118L02) positioned dorsally to the fly, with intensity measured at the location of the tethered fly. This optogenetic light stimulus was centered on the fly’s head and illuminated the entire fly.

QUANTIFICATION AND STATISTICAL ANALYSIS

All data was collected in MATLAB 2023b and analyzed using custom Python scripts. Alpha values were set at .05 for all statistical tests, with * representing p<.05, ** representing p<.01, and *** representing p<.001.

To analyze active and passive antennal movements, we computed the angles for both the pedicel and arista relative to a head axis midline using the points in Figure 1C. Inter-antennal angle for the pedicel and arista segments was calculated by adding the angle of the left and right antenna for each segment respectively.

Across all behavioral groups, the Kolmogorov-Smirnov test (scipy.stats.kstest) was used to confirm normality of data.

For most statistical analyses (Figure 2D–G, 3C, 4B–F, 5D, S1H–I, S3B–E), we used a linear mixed-effect model (statsmodels.formula.api.mixedlm) with post-hoc tests to determine individual differences between groups at each stimulus level. All but one model included the stimulus (wind speed or optic flow speed) as a continuous variable (in increments of 50 cm/s and 5 cm/s respectively) and the conditional groups (e.g., darkness v. static grating, flying v. nonflying, and JON silencing) as categorical variables. Both stimulus speed and conditional group were considered fixed effects (estimated using restricted maximum likelihood, REML); individual flies were included as random intercepts. The model was specified as follows:

$$\mathrm{m}\mathrm{e}\mathrm{a}\mathrm{s}\mathrm{u}\mathrm{r}\mathrm{e}~\mathrm{s}\mathrm{t}\mathrm{i}\mathrm{m}\mathrm{u}\mathrm{l}\mathrm{u}\mathrm{s}+\mathrm{c}\mathrm{o}\mathrm{n}\mathrm{d}\mathrm{i}\mathrm{t}\mathrm{i}\mathrm{o}\mathrm{n}+\left(1\left|\mathrm{f}\mathrm{l}\mathrm{y}\right.\right)$$

Where measure is inter-antennal angle, wingbeat frequency, or wingbeat amplitude.

For Figure 4B, we used two categorical variables: flight state (i.e. flying or nonflying) and condition (i.e. intact antennae, glued antennae, or inactivated JONs). Given the relative simplicity of the model, we included random slopes per fly to increase model accuracy. The results of all linear mixed-effect models and their descriptive equations can be found in Table S1–S6.

To test for linearity and homoscedasticity, we generated residuals vs. fits plots and looked for abnormality in variance. While some “pinching” was observed near zero in models using baseline-subtracted data (which was expected, given reduced variability near baseline), we did not observe any major abnormalities. We also checked for normality using Q-Q plots, and did not detect any substantial violations of assumptions.

Post-hoc analysis was performed through model-derived estimated marginal mean (EMM) contrast comparisons across conditional groups at each stimulus speed (wind speed or optic speed). To control Type I error, we then performed a Holm-correction (statsmodels.stats.multitest.multipletests) on generated p-values on a per model basis.

Other analyses performed included independent Student’s t-tests without correction (Figure 2D–E), paired Student’s t-tests (scipy.stats.ttest_rel, Figure 3C, 5C), one-sample Student’s t-tests with Holm-correction (scipy.stats.ttest_1samp, Figure 5D), independent t-tests with Holm-correction (scipy.stats.ttest_ind, Figure 4E) and a linear regression (scipy.stats.linregress, Figure S2D).

In Figure 6, Figure S6, and Figure S7, we performed a Fast Fourier Transform (scipy.fft.fft and scipy.fft.fftfreq) on all average traces to obtain oscillation gain and phase lag information (Figure 6B–E, 6G–J, S6D–E, S7). We normalized the fft output by dividing it by a normalized-to-one unitless sinewave in phase with the oscillating stimulus. We calculated oscillation gain by taking the absolute value (np.abs) of this measure and oscillation phase by taking the angle of this measure (np.angle). To account for phase wrapping, all phase lags above 0 (i.e., phase lead) were subtracted by 2π radians. To estimate the error of gain and phase, we performed a 100-epoch Monte Carlo simulation with replacement at each oscillation frequency for each condition. Error bars derived from all Monte Carlo simulations represent the range of the 95 out of 100 most central epochs relative to the mean value.

Shaded error bars unless otherwise noted represent a 95% confidence interval.

Supplementary Material

Document S1. Figures S1–S7, Tables S1–S6.

Highlights:

• Mechanosensory input guides active antennal positioning at flight onset

• Airflow-dependent active antennal movements depend on the visual environment

• Flying flies perform visually driven active antennal movements

• Integration of airflow and optic flow follows a “winner-takes-all” computation