Active antennal movements in Drosophila can tune wind encoding

Summary

Insects use their antennae to smell odors^1,2, detect auditory cues^3,4, and sense mechanosensory stimuli such as wind⁵ and objects^6–8, frequently by combining sensory processing with active movements. Genetic access to antennal motor systems would therefore provide a powerful tool for dissecting the circuit mechanisms underlying active sensing, but little is known about how the most genetically tractable insect, Drosophila melanogaster, moves its antennae. Here we use deep learning to measure how tethered Drosophila move their antennae in the presence of sensory stimuli, and identify genetic reagents for controlling antennal movement. We find that flies perform both slow adaptive movements and fast flicking movements in response to wind-induced deflections, but not the attractive odor apple cider vinegar. Next, we describe four muscles in the first antennal segment that control antennal movements, and identify genetic driver lines that provide access to two groups of antennal motor neurons and an antennal muscle. Through optogenetic inactivation, we provide evidence that antennal motor neurons contribute to active movements with different time courses. Finally, we show that activation of antennal motor neurons and muscles can adjust the gain and acuity of wind direction encoding by antennal displacement. Together, our experiments provide insight into the neural control of antennal movement and suggest that active antennal positioning in Drosophila may tune the precision of wind encoding.

Graphical Abstract

eTOC blurb:

Suver et al. characterize active antennal movements in Drosophila and provide genetic access to several antennal motor neurons and antennal muscles. Active positioning of the antenna can change how wind is encoded by antennal deflections.

Results and Discussion

Flies exhibit diverse active antennal movements that are promoted by wind and flight

To measure active antennal movements in Drosophila, we presented head-fixed flies with wind stimuli of various speeds (0–200 cm/s) and directions (−90°, −45°, 0°, +45°, and +90°) while recording antennal movements using a camera positioned in front of the fly (Figure 1A, Methods). Flies were mounted at the center of 5 airstreams⁹, and flew on a subset of trials. The three largest segments of the Drosophila antenna (Figure 1B) are the first (scape), second (pedicel), and third (funiculus). A large sail-like bristle, known as the arista, is stably fixed to the third segment and amplifies air currents due to wind or sound, rotating the third segment in response to these stimuli^9–11. These rotations are detected by a set of stretch receptor neurons known as Johnson’s Organ Neurons (JONs)^12–14 that span the joint between the second and third segments. As no muscles insert in this joint, its movements are entirely passive^15–17. In contrast, the joint between the first and second segments is spanned by several muscles and can be actively controlled by the fly^16,18.

Figure 1. Active antennal movements in response to wind and flight.

(A) Schematic of experimental setup showing wind manifold and camera position. Fly was head-fixed and held at the center of 5 airstreams of the manifold (red star). (B) Schematic of the fly antenna, showing three main segments: first (scape), second (pedicel), and third (funiculus). Active and passive joints are shown in orange and purple, respectively. Negative angles represent movements forward away from the head and positive angles represent movements backwards towards the head. (C) Still image of the fly head showing the segment and joint angles tracked. Active joint movements were computed as the angle of the second segment relative to the midline (orange blue). Passive joint movements were computed as the difference between the third and second segment angles (purple, cyan). (D) Single-trial examples of all measured joint angles with colors as in (C). A 2 s frontal (0°) wind stimulus at 200 cm/s elicits posterior (positive) deflections of both left and right passive joints (purple and cyan traces) that are tightly locked to the wind stimulus, shown for three example flies. In contrast, active joint movements (orange and blue) show diverse waveforms that are not tightly locked to the wind stimulus. Examples illustrate no active movement (left), a rapid unilateral active movement (middle), and bilateral slow active movements (right). (E) Average passive joint deflections in response to a 2 s contralateral wind (+45°) stimulus across windspeeds (top row) and in response to 200 cm/s wind across five directions (bottom row). Thin lines represent single trial averages (n = 3–6 trials per fly), thick lines indicate the mean across fly averages (N = 12). (F) Left: All active joint movements in response to 200 cm/s frontal wind, filtered to frequencies below 2.5 Hz. (n = 84 traces from 12 flies). Right: mean and standard error of filtered active joint movements. (G) Left: All active joint movements in response to 200 cm/s frontal wind, filtered to frequencies above 2.5 Hz (same traces as in F). Right: mean and standard error of the absolute value of filtered active joint movements. (H) Raster of active joint movements (both antennae) for different speeds of a single wind direction (+45° ipsilateral). Different colors represent trials from different flies (N = 12 non-flying flies). Peristimulus time histogram of active joint movements shown in gray below raster. (I) Average number of active joint movements per trial as a function of wind direction. Dots represent single flies (N= 12), lines represent average across flies. (J) Frequency of active joint movements in odorized (1% apple cider vinegar) versus non-odorized wind, in both non-flying and flying flies. (K) Violin plots showing the average number of active joint movements during the wind stimulus (circles), the standard deviation (black line), and the range (shaded region). For each condition shown, average number of movements per second during the stimulus was 1.69+/3.25, 1.88+/−3.57, 2.79+/−4.23, and 2.78+/−4. 08, respectively. Flying flies respond with more active joint movements regardless of the presence of odor (p<0.0001, student’s t-test), whereas no significant change in movements was observed in the presence of odor (p=0.23 and p<0.0001 during nonflying and flying trials, respectively). Flight data is from N = 6 of the 7 total flies. Averages represent the mean of n = 206, 198, 82, and 88 trials, respectively. All flies in Figure 1 are Canton-S>Chrimson and experiments were performed in the presence of red light to serve as controls for later optogenetic activation experiments. See also Figure S1, Video S1.

We used DeepLabCut¹⁹ to track the position of multiple points on the head and antennae (Figure S1A), and computed the angles of the passive and active joints from a subset of these (Figure 1C). Consistent with the anatomy described above, we found that passive joint movements were tightly locked to the wind stimulus (Figure 1D, S1B, magenta and teal traces), and changed systematically in magnitude and direction with wind speed and direction (Figure 1E). In contrast, active joint movements had more variable timing and magnitude across trials and flies. These active movements could be rapid (Video S1) or slower (Figure 1D, orange and blue traces).

Although active movements (second segment angle versus midline) spanned a range of frequencies (Figure S1C), we observed a distinction between slow movements (<2.5Hz) and faster movements (>2.5Hz). Slow movements generally brought the antennae forward (decreased angle), and opposed the backward deflections produced by wind (Figure 1F). These movements grew gradually in response to wind (Figure S1D), and resemble adaptive positioning observed in moth antennae in the presence of frontal air currents^16,20,21. In contrast, fast movements occurred in both directions, and were concentrated at the onset of the wind stimulus (Figure 1G). By detecting active movements as peaks in the active joint angle trace (Methods), we found that they increased in frequency with windspeed (Figure 1H) and frontal direction (Figure 1I). Although there was a weak correlation between active movements of the two antennae (Figure S1E-F), they could move independently, as shown in the second example in Figure 1D.

In several arthropods, such as locusts¹ and cockroaches²², antennal movements increase in the presence of odor and are thought to function like sniffing²³ to increase the persistence of odor at olfactory sensilla. We compared active joint movements in response to odorized and non-odorized wind, but found no difference between these two conditions (Figure 1J). In contrast, we did observe an increase in active joint movements during flight (average increase of 1.1 movements/s, p<0.0001, student’s t-test; Figure 1K) as previously observed²⁴. Blocking active joint movements by stabilizing the first and second segments did not impair odor-evoked upwind walking (Figure S1G). Together, these data argue that active antennal joint movements in Drosophila do not function to improve odor detection and are not required for wind orientation, at least when walking in laminar airflow. Rather, they suggest a role in detecting dynamic air currents in flight, when effective airspeeds are higher and antennal deflections are larger.

Distinct motor neurons promote antennal movements with different time courses

While mechanosensory neurons of the fly antenna have been extensively characterized^12–14, little is known about the muscles and motor neurons controlling antennal movements. To investigate motor control of the antennae, we visualized antennal muscles and tendons (see Methods, Figure 2A-C). We observed four distinct muscles within this structure (Figure 2B, C), in contrast to classical studies¹⁸, but consistent with findings in blowflies¹⁷. Tendon labeling (Figure 2C) suggests that these muscles insert at the interior portion of the second segment, where it joins with the first. Next, we used the FlyLight collection²⁵ to search for genetic lines that label antennal motor neurons, selecting candidate lines with expression in the antennal mechanosensory and motor center (AMMC) where motor neuron inputs reside^26,27, and in the lateral margin of the antennal nerve, where motor neuron axons are found^28–30. We then screened candidate lines for anatomical innervation of antennal muscles (Figure 2D-F) and for antennal deflections during optogenetic activation (Figure 2G-J, S2A-D). For these analyses only, we measured second and third segment angles relative to the midline, rather than joint angles. Through these approaches, we identified two genetic lines labeling antennal motor neurons (Figure 2D, E, H, I) and one line labeling an antennal muscle (Figure 2F, J).

Figure 2. Anatomy of antennal muscles and genetic lines labeling antennal motor neurons.

(A) Antennal preparation with first antennal segment outlined as an inset, indicating the location of the four antennal muscles, numbered 1 to 4 from anterior to posterior. Schematic depicts coronal (frontal) projection of the antennae. Gray lines indicate outer and inner cuticle surrounding the first antennal segment. (B) Four muscles in the first antennal segment. Left: phalloidin Alexa 568 stain for muscle (magenta). Center: cuticle (gray). Right: overlay of phalloidin stain and cuticle with outlines of muscles. (C) Tendon expression using sr-GAL4>10xUAS-GFP (green) with the four muscles in the first antennal segment. Yellow arrows indicate tendons; muscles numbered as in (A). (D) Motor neuron processes labeled by 18D07. Left: muscle 1 innervation. Center: muscle 4 innervation. Right: schematic of innervated muscles. (E) Motor neuron processes labeled by 91F02. Left: muscle 3 innervation. Center: muscle 4 innervation. Right: schematic of innervated muscles. (F) Muscle labeled by 74C10. Left: phalloidin stain of muscles. Center: GFP expression overlaid on phalloidin stain. Right: schematic of labeled muscle. (G-J) Third segment deflections produced by Chrimson activation in control flies (Canton-S>Chrimson, N=12 flies) and three experimental lines (18D07, N=9 flies; 91F02, N=8 flies; and 74C10, N=10 flies). Thin traces represent average response of one fly, thick traces are means across flies. Anatomy and activation for 91F02 used eyFLP to suppress expression in JONs in this line (see Methods). See also Figure S2, Video S2-4.

One driver (18D07-GAL4, Figure 2D) labeled two motor neurons synapsing onto the dorsomedial antennal muscles 1 and 4. Consistent with this anatomy, activation of this line drove the antennae forward and up (decreased antennal angle, Figure 2H, Video S2). A second driver (91F02-GAL4, Figure 2E) labeled motor neurons innervating lateral muscles 3 and 4, as well as some JONs, which we suppressed using eyFLP>tub(FRT.stop)Gal80³¹. Curiously, activation of this line evoked only small, transient antennal movements at light onset and offset (Figure 2I, Video S3)³¹. Finally, one driver line produced large antennal deflections down and backward (positive antennal angle, Fig. 2J), but did not label any motor neurons. Instead, this driver (74C10-GAL4, Figure 2F, Video S4) labeled the large ventro-lateral muscle 3 which pulls the antennae down and back. Thus, these three lines allow for differential experimental control over antennal positioning.

To determine the role of these motor neurons in active movements, we expressed the inhibitory opsin GtACR1³² in each line, and measured wind-evoked active joint movements in the presence and absence of light. In control flies (empty-GAL4>GtACR1), light did not alter the frequency of active joint movements (Figure 3A, S3A), although it reduced mean active joint deflection (Figure S3E-F). To analyze the time course of active joint movements, we filtered them into low and high frequency components, as performed above. Similar to our previous observation (Figure 1F, G), slow active joint movements in control flies during light predominantly brought the antennae forward (negative angle), while fast active joint movements occurred throughout the wind stimulus and after (Figure 3D).

Figure 3. Motor neurons contribute to active movements with different time courses.

(A) Effects of optogenetic silencing on active joint movements in control flies (empty>GtACR1). Left: example active joint traces in response to wind (45° contralateral, 200cm/s) with no light. Center: example active joint traces in response to wind with light. Right: No significant difference in the number of active movements in light vs no light (average difference = −0.109+/−0.621 std, p=0.59, N = 10 flies). Thin lines represent single fly averages; thick line represents cross-fly average. (B) Effects of optogenetic silencing of 18D07 on active joint movements. Left: examples, wind with no light. Center: examples, wind with light. Right: Significant decrease in active movement number with 18D07 silencing (average difference = −1.138+/−0.871, p=0.002, N = 11 flies). (C) Effects of optogenetic silencing of 91F02 on active joint movements. Left: examples, wind with no light. Center: examples, wind with light. Right: Significant decrease in active movement number with 91F02 silencing (average difference = −0.454+/−0.482, p=0.014, N= 11 flies). All p-values for (A-C) from paired t-test. (D) All active joint movements in control flies (empty>GtACR) during light (n = 120 trials in N = 10 flies), filtered below 2.5 Hz (top) or above 2.5 Hz (bottom). (E) Low frequency component of active joint movements in 18D07>GtACR flies during light (<2.5 Hz). Left: all traces (n = 110 traces from N = 11 flies). Right: Mean and standard error for 18D07>GtACR (green) versus empty>GtACR (black). (F) High frequency component of active joint movements in 18D07>GtACR flies during light (>2.5 Hz). Left: all traces. Right: Mean and standard error of the absolute value in 18D07>GtACR (green) versus empty>GtACR (black). (G) Low frequency component of active joint movements in 91F02>GtACR flies during light (<2.5 Hz). Left: all traces (n = 106 traces from N = 11 flies). Right: Mean and standard error for 91F02>GtACR (green) versus empty>GtACR (black). (H) High frequency component of active joint movements in 91F02>GtACR flies during light (>2.5 Hz). Left: all traces. Right: Mean and standard error of the absolute value in 91F02>GtACR (green) versus empty>GtACR (black). See also Figure S3 and Video S3.

Silencing the first motor neuron line (18D07>GtACR1) significantly reduced the frequency of active joint movements, and altered their time course (Figure 3B, S3B). In the absence of light, active joint movements often persisted for the duration of the wind stimulus and beyond. During light, the remaining active movements were rapid and clustered near the beginning of the wind stimulus. To quantify the effects of 18D07 silencing on active joint movement time course, we filtered each trace into a high and low frequency component (Figure 3E, F) and compared these to active joint movements of control flies in the presence of light. Low frequency movements in 18D07-silenced flies were almost exclusively backwards (increased antennal angle), the opposite of what we observed in control flies (Figure 3E) and in the absence of light (Figure S3G). This result supports the idea that 18D07 motor neurons are required for the adaptive frontal movements observed in control flies during wind. High frequency movements were also strongly reduced, and the remaining movements were clustered near the beginning of the wind stimulus (Figure 3F). Silencing of 18D07 motor neurons also produced a tonic offset in second segment position (Figure S3B, D), suggesting that these neurons contribute to steady-state antennal positioning. These data indicate that 18D07 motor neurons contribute to slow frontal positioning and to rapid movements later in the wind stimulus.

Silencing the second motor neuron line (91F02>GtACR1) also reduced active joint movement frequency. However, the remaining active joint movements were longer in duration and tended to occur later in the stimulus (Figure 3C, S3C). Analysis of low and high frequency components (Figure 3G, H, S3I, J) shows that 91F02 silencing affects both fast and slow movements, but delays the occurrence of active joint movements, so that most movements arise later after wind onset. Silencing 91F02 also produced tonic changes in antennal position, of a similar magnitude to those produced by 18D07 silencing (Figure S3C, D). It remains possible that some of these effects are due to silencing of JONs in this line, however, these results suggest that 91F02 contributes more to movements triggered rapidly by wind onset.

Active antennal movements can alter the gain and precision of wind encoding

Flies and other walking insects compute wind direction from differential displacements of their antennae, which are decoded by central neurons in the brain to subserve wind orientation behavior^{5,9–11,33–35}. Because this differential displacement depends on antennal angle, we reasoned that changing antennal position should change peripheral encoding of wind direction. To test this hypothesis, we expressed the light-gated cation channel Chrimson in each of our motor neuron and muscle lines, and compared the encoding of wind direction by displacements of the passive joint (third segment minus second segment angle) to a genetic control (Canton-S>Chrimson, Figure 4A-D).

Figure 4. Antennal positioning modifies the gain and precision of wind encoding.

(A) Steady-state passive joint deflection (third segment angle minus second segment angle) across wind directions and speeds in control flies (Canton-S>Chrimson). Left: blue-purple hues indicate left antennal responses, orange-red hues indicate right antennal responses. Right: difference between right and left antennal deflections shown in gray hues. Dots represent single flies (N=12 flies). Lines represent cross-fly averages. (B) Effects of activating 18D07 on wind encoding by passive joint displacement differences (N=9 flies). (C) Effects of activating 91F02 on wind encoding by third segment displacement differences (N=8 flies). For these experiments, we suppressed expression in JONs using eyFLP (eyFLP>GAL80;91F02>Chrimson, see Methods). (D) Effects of activating 74C10 on wind encoding by passive joint displacement differences (N=10 flies). Colors and symbols in (B-D) as in (A). (E) Gain (average passive joint displacement difference at −45° minus +45°) during light activation across speeds for the four genotypes. (F) Passive joint deflection gain (average at −45° minus +45°) across flies during light activation at 200 cm/s (subset of data in (E)). Gain significantly changes with 18D07 activation (p=0.025), 91F02 activation (p=0.007), and 74C10 activation (p<0.0001). Comparison made with two-sided student’s t-test. (G) Tuning curves (see Methods) fit to passive joint displacement differences for each genotype at 200 cm/s. (H) Fisher information computation for each genotype at 200 cm/s. Shaded region indicates standard error of the mean. (I) Hypothesized function of active antennal movements: acuity of wind direction encoding can be modulated by actively positioning the antennae. See also Figure S4 and Video S4.

As described above, activation of the 18D07 motor neuron line moved the antennae forward and up, similar to natural movements induced by the start of flight (Figure 2H). Wind-evoked displacements of the third segment were slightly larger in this position, leading to a wind direction tuning curve with somewhat higher gain (Figure 4B, E, F). In contrast, while activation of the 91F02 motor neuron line produced little net change in antennal position (Figure 2I), 91F02 activation reduced the gain of wind encoding (Figure 4C, E, F). Finally, activation of the muscle driver line 74C10 pulled the antennae down and away from the midline (Fig. 2J). This manipulation strongly reduced wind encoding gain and shifted the tuning curve so that the peak difference between antennal displacements occurred at more eccentric wind angles (+/− 90° instead of +/− 45°, Figure 4D). These results support the hypothesis that active changes in antennal position can alter the encoding of wind direction at the periphery (Figure 4E, F).

To understand whether antennal position can alter the information a fly can encode about wind direction, we computed the Fisher Information, a measure of discriminability between nearby stimulus directions³⁰. The Fisher Information is equal to the slope of the tuning curve normalized to the standard deviation of the response, and measures how much the response changes (relative to its variability) when the stimulus changes. To compute the Fisher Information, we fit a smooth function to the displacement tuning curves for the fastest wind direction (Figure 4G) for each genotype, differentiated this function, and divided by the standard deviation across trials for each genotype (see Methods). This analysis (Figure 4H) shows that 18D07 activation increases the Fisher Information for frontal wind directions, whereas 91F02 and muscle activation (74C10) decrease Fisher Information. These effects arise both because of the changes in tuning curve slope, and because responses during 18D07 activation were more reliable across trials. This analysis argues that flies could use active antennal positioning to tune their acuity for wind direction (Figure 4I).

Discussion

Many animals actively move their sensors to alter the information they can obtain from their environment. For example, humans actively move their hands and fingers to determine the size and texture of objects^36,37, rodents whisk to sense their surroundings in the dark³⁸, and both mammals and lobsters sniff to alter the flow of olfactory information^23,39. While such active movements are generally proposed to increase sensory information or acuity, this is challenging to measure quantitatively without experimental control over animal movement.

Insect antennal movements provide an attractive model for studying active sensing. Many insects move their antennae to explore objects^7,8, to amplify odors¹, or during the transition to flight^26,40. The antennae play a central role in flight control; the forward position of the antennae in flight allows them to sense air currents, set flight speed^16,41–44, and provide feedback on wingbeat amplitude^45–46. Further, antennal position reflects the integration of signals from multiple modalities. For example, honeybees advance their antennae with increased airflow but retract them in response to optic flow²⁰. Thus, antennal motor neurons are a likely site for multisensory integration^20,47. Antennal movements are controlled by a variable number of muscles across species, each innervated by motor neurons originating in the AMMC^{27–30,48,49}. These motor neurons receive fairly direct synaptic input from mechanosensory neurons at each of the antennal joints (JONs and Böhm’s bristles)^26,27, as well as indirect input from olfactory²², visual^20,47, and descending motor systems⁵⁰. However, no studies have yet provided genetic access to antennal motor circuitry.

In this study, we set out to characterize antennal movements in Drosophila melanogaster, and to identify genetic lines for experimental control over antennal movement. We observed slow movements (<2.5Hz) that move the antennae forward in response to airflow, similar to what has been previously observed in bees, moths, and locusts^26,40. We also observed more rapid movements that could help calibrate the antennal air current sensing system⁵¹. Surprisingly, we found that active joint movements increased strongly during wind and flight but not during odor. This suggests that flies may use antennal movements primarily to tune mechanosensory input, particularly during flight^45,46. Future experiments measuring active movements during flight – particularly in 3 dimensions – and manipulating them using the genetic reagents described here, will help resolve the function of active antennal movements in Drosophila.

In this study we also provided insight into Drosophila antennal motor control. We found that Drosophila, like the blowfly Callifora¹⁶, controls antennal position using 4 muscles. We identified two genetic lines that label 4 motor neurons, innervating 3 out of 4 antennal muscles. Our silencing data suggest that different antennal motor neurons subserve movements with different time courses. Because we did not identify any motor neurons innervating muscle 2, and because antennal muscles in insects are typically innervated by more than one motor neuron^29,30,48, we expect that Drosophila possesses additional antennal motor neurons. We used a pan-motor neuron line (VGlut-GAL4), to try to count the total number of antennal motor neurons but were unable to visualize individual antennal axons in this line. Future studies, using electron microscopy and x-ray images of the antenna and ventral brain, will allow for a more complete mapping of antennal motor circuitry, and a more detailed understanding of the mechanics of antennal joints.

Finally, we examined how active changes in antennal position might change the encoding of wind direction. We showed that experimentally altering antennal position could change the shape of the wind tuning curve, as measured at the passive sensory joint between the second and third antennal segments. Moving the antennae forward, as insects do during flight, increased frontal wind tuning gain and acuity. This may represent an adaptation to higher airspeeds encountered during flight (typically 0.5–1 m/s⁵²) in contrast to walking, where windspeeds above ~15–20 cm/s cause freezing behavior¹¹. This result is consistent with previous studies showing that insects can dynamically position their antennae depending on the speed of air currents^{16,20,41–43}. In contrast, the more lateral position of the antennae during walking could promote sensation of lateral and posterior mechanosensory stimuli, such as conspecific song. We showed that forward movement allows the fly to “foveate” its wind sensing and also that the two antennae can move independently. Thus, flies may be able increase wind acuity in different regions of space by differentially positioning their antennae.

Although we show here that active antennal positioning could enable the fly to dynamically tune its encoding of wind direction, this presents a challenge for central neurons that must decode wind direction to drive orientation behavior^9,10,34. To compute wind direction in the presence of active movements, central neurons might receive predictive motor signals that would allow them to correct for the angular position of the antenna. In several mechanosensory systems, neurons downstream of the sensory periphery receive input from the motor system^53–55. Future studies combining genetic control of antennal position with recordings from central neurons will allow us to pinpoint how wind direction is dynamically computed during active sensing.

STAR Methods

RESOUCE AVAILABILITY

Lead contact

Further information and requests for resources and reagents should be directed to and will be fulfilled by the lead contact, Katherine Nagel (katherine.nagel@nyumc.org).

Materials availability

This study did not generate new unique reagents.

Data and code availability

Data generated in this study is available at Zenodo at DOI: 10.5281/zenodo.7508037. Code generated during this study is available on Github at https://github.com/nagellab/SuverEtAl2023.

EXPERIMENTAL MODEL AND SUBJECT DETAILS

For all experiments, we used adult Drosophila melanogaster, between 1 and 10 days old (typically 2–4 days old). For activation experiments (Canton-S>Chrimson, 18D07>Chrimson, 91F02>Chrimson, and 74C10>Chrimson), we used female to male ratios of 7:5, 5:4, 4:4, and 8:2, respectively. We observed no obvious difference in antennal movements between male and female flies. For all other experiments, we used only female flies. Exact genotypes for each figure panel are listed in the table below. Parental strains and RRIDs are listed in the Key Resources Tables.

KEY RESOURCES TABLE

REAGENT or RESOURCE	SOURCE	IDENTIFIER
Antibodies
Alexa Fluor 1568 Phalloidin	Invitrogen	A12380
Anti-GFP Polyclonal Antibody, Alexa Fluor 488	Invitrogen	A-21311
Chemicals, peptides and recombinant proteins
All trans retinal	Sigma	RRID: BDSC_R2500
Experimental Models: Organisms/strains
Canton-S	Dickinson Lab	N/A
sr-GAL4	BDSC	RRID: BDSC_26663
R18D07-GAL4	BDSC	RRID: BDSC_48813
R91F02-GAL4	BDSC	RRID: BDSC_40584
R74C10-GAL4	BDSC	RRID: BDSC_39848
+(HCS);+;P{10xUAS-IVS-Syn21-GFP-p10}attP2	Michael Dickinson (wild-type backcrossed onto chromosome I using GFP construct from Pfeiffer et al., 2012)	N/A
ey-FLP	BDSC	RRID: BDSC_5576
w[*];P{w[+mC]=tubP(FRT.stop)GAL80}2; MKRS/TM6B, Tb[+]	BDSC	RRID: BDSC_38878
w[1118]; P{y[+t7.7] w[+mC] = 20XUAS-IVS-CsChrimson.mVenus}attP40 (UAS-Chrimson)	BDSC	RRID: BDSC_55135
w+;s/Cyo;UAS-GtACR1-eYFP	Desplan Lab	N/A
empty-GAL4	BDSC	RRID:BDSC_68384
Software
Python version 3.7.6	https://www.python.org/downloads/release/python-376/	version 3.7.6
Matlab version 2018b	MathWorks	RRID: SCR_001622
DeepLabCut version 2.0.5.1	http://www.mackenziemathislab.org/deeplabcut	version 2.0.5.1
ImageJ version 2.1.0	https://imagej.nih.gov/ij/	RRID: SCR_003070
Custom Code for this project	https://github.com/nagellab/SuverEtAl2023
Data for this project	DOI: 10.5281/zenodo.7508037

Fly strains

Figures 1A-I	Canton-S/UAS-20xChrimson
Figures 1J-K	Canton-S
Figures S1A-F	Canton-S/UAS-20xChrimson
Figure S1G	Canton-S
Figure 2B	tub(FRT.stop).GAL80/ eyFLP;91F02-GAL4/10xUAS-GFP
Figure 2C	sr-GAL4;UAS-10xGFP
Figures 2D, H, S2B	18D07-GAL4×10xGFP
Figures 2E, I, S2C	tub(FRT.stop).GAL80/ eyFLP;91F02-GAL4/10xUAS-GFP
Figures 2F, J, S2D	74C10-GAL4/10xGFP
Figures 2G, S2A	Canton-S; UAS-20xChrimson
Figures 3A, D, E-H, S3A, D, E-J	empty-GAL4/UAS-GtACR1
Figures 3B, E-F, S3B, D, G-H	18D07-GAL4/UAS-GtACR1
Figures 3C, G-H, S3C, D, I-J	91F02-GAL4/UAS-GtACR1
Figures 4A, 4E-H, S4A	Canton-S/UAS-CsChrimson
Figures 4B, 4E-H, S4B	18D07-GAL4/UAS-CsChrimson
Figures 4C, 4E-H, S4C	tub(FRT.stop).GAL80/ eyFLP;91F02-GAL4/UAS-CsChrimson
Figures 4D, 4E-H, S4D	74C10-GAL4/UAS-CsChrimson

METHOD DETAILS

Wind Apparatus and Antenna Tracking

To mount flies in the behavior apparatus, we cold-anesthetized flies and cut off each leg at the femur-trochanter joint. Next, we glued the anterior edge of the thorax and the dorsal rim of the head to the edges of a thin stainless steel cutout attached to a plastic holder using a small amount of UV-cured glue. This holder was positioned in the center of a wind manifold with channels directed to the fly from 5 directions (−90°, −45°, 0°, +45°, and +90° relative to the fly’s midline). Wind speed was controlled using a mass flow controller (Aalborg GFC17A-VAL6-A0) and the timing and direction by a series of solenoid valves (Lee LHDA1233115H) controlled by an Arduino MEGA 2560 and custom Matlab software. We measured wind speed with a calibrated hot wire anemometer (Dantec MiniCTA with wire probe).

For odor delivery experiments, we used the same wind manifold but with an additional set of valves controlling odor delivery (same as used in a previous study: Suver et al., 2019). For these experiments, we delivered a 2 s wind pulse (60 cm/s) to 3–4 day old female flies from −45 or +45 deg. This wind contained either humidified air or 1% apple cider vinegar (Pastorelli brand).

We measured movements of the antennae with a camera mounted in front of the fly (head-on) at a framerate of 60 Hz. The fly was illuminated by IR light (880 nm) delivered by two bare fiber optic cables aimed towards the left and right of the fly from behind (Thorlabs). Wingbeats were measured using reflected IR light detected by a tachometer affixed with an IR filter (https://github.com/janelia-kicad/light_sensor_boards) mounted below the fly.

To track movements of the antennae, we used DeepLabCut⁴⁶ for pose estimation (version 2.0.5.1, Mathis et al., 2018). We trained our network using a set of 59 frames from 35 flies across genotypes. After an initial training of the network using 30 randomly selected frames, we selected a few additional frames by hand from videos in which we observed more tracking errors (e.g., frames when the fly was flying or those with particularly large antennal movements). We tracked 32 points on the head of the fly, including several stationary points on the head to determine the head axis, and several along the second and third segments of the antenna (Fig. S1A). We set a high training/test fraction of 0.95 (95% frames used for training, 5% for test) because our dataset was relatively unvarying (head-fixed flies with similar lighting). We used a ResNet-50-based neural network, and our network training yielded a test error of 3.48 pixels and train error of 1.31 pixels in an image size of 640×480 pixels.

Anatomy

For the antenna image in Figure 2A, we placed a freshly removed pair of antennae from the head still attached to cuticle using fine forceps, and placed these on a damp Kimwipe. We took this image using a Google Pixel 4a aimed by hand into one eyepiece of a stereomicroscope (Leica M80).

To visualize antennal muscles and the motor neurons, we dissected whole antennae in PBS using fine forceps. We took care to avoid damaging the antennal nerve, leaving them connected to central brain. After removal, we placed the antennae in 4% paraformaldehyde solution (in PBS) for 50 minutes on an oscillator at room temperature. Following this, we washed the antennae in PBST for 15 minutes, followed by a wash in PBS for another 15 minutes. To stain antennal muscles, we placed antennae in a solution of PBS, Alexa Fluor Phalloidin 568 (10μL stock in 1 mL PBS, Invitrogen A12380) and rabbit anti-GFP 488 (10µL in 1mL PBS, Thermofisher A6455) for approximately 48 hours. We then removed the antennae from the solution and placed it in PBS for 15 minutes on an oscillator at room temperature. Afterwards, we mounted the antennae anterior side up on a glass microscope slide in Vectashield (Vector Labs H-1000–10), covered with one #0, 22 × 22 mm coverslip, and sealed with nail polish. We imaged antennae at 20x magnification on a Zeiss LSM 800 confocal microscope at 1 μM depth resolution. Final images are presented at maximum z-projections over relevant depths.

Optogenetics

We placed adult flies on food enriched with all-trans retinal for 24–48 hours prior to optogenetics experiments. To make enriched retinal food, we placed 50 μL all-trans retinal (35 mM stock, Sigma R2500–100MG dissolved in ethanol, stored at −20°C) in 1 tsp rehydrated potato flakes placed on top of standard cornmeal-molasses fly food. For Chrimson activation, we used red light with an intensity sufficient to produce antennal deflections in our initial motor neuron screen (3.23 μW/mm² measured at 658 nm at the location of the fly when mounted). For GtACR1 inactivation experiments, we used green light at 70 μW/mm² measured at 530 nm.

Walking arena behavior

For walking behavior, we used wild type Canton-S flies that were 2–5 days old (roughly equal numbers of males and females). These flies were raised on standard food and a 12:12 light cycle at 25°C. These flies were then starved (housed in an empty vial with a damp kimwipe) for 24 hours prior to the experiment to encourage odor-seeking. We blocked movements of the first and second antennal segments by applying UV-curing glue to a tiny portion of the head to the first segment and dorsal edge of the second segment in approximately rest position. For each fly, after the experiment finished, we determined whether our gluing procedure was successful by verifying under a light microscope that the third segment was free but first and second remained fixed to the head. We used previously described walking wind tunnels⁵⁶ to present controlled wind and odor stimuli to freely walking flies. We placed cold-anesthetized single flies in individual walking tunnels. We let these flies acclimate for several minutes, then presented them with either no wind, wind, or constant wind with a 10 s pulse of 10% apple cider vinegar (made with Pastorelli brand).

QUANTIFICATION AND STATISTICAL ANALYSIS

We analyzed all data in Python 3.7.6 and Matlab 2018b.

To analyze active antennal movements, we first computed second and third segment angles with respect to the midline based on the points shown in Figure S1A. Transient tracking errors (angles greater than 35 deg, which we only ever observed as being caused by tracking errors) were detected and omitted from analysis. Active joint displacements were equal to the second segment angle relative to the midline. Passive joint displacements were equal to the third segment angle minus the second segment angle. We used joint angle measurements for all analyses shown except for Figure 2, where we show raw second and third segment angles relative to the midline.

To analyze active joint displacements, we low- or high-pass filtered the active joint displacement traces with a 2-pole Butterworth filter with a cutoff frequency of 2.5 Hz. Mean joint displacements were computed by averaging across all trials for a given stimulus condition.

To detect rapid active movements, we first baseline subtracted each active joint measurement and low-pass filtered with a butterworth filter (cutoff frequency = 10 Hz). We then found peaks in the resulting trace using signal.find_peaks with prominence of 1.5 degrees. We omitted transient movements at the light or wind stimulus on and offset (within 2 frames at onset for both, and 8 or 10 frames at offset, respectively).

For active movement peri-stimulus time histograms (Figure 1H, J,), we mirrored left antennal movements and pooled these with data from the right antenna so that stimuli ipsi- and contralateral to the antenna were correct for each antenna. We normalized histograms to the number of trials, converted to movements per s, and low-pass filtered with a Butterworth filter with cutoff frequency of 5 Hz. We computed auto- and cross-correlograms (Figure S1F) across all raw antennal movements for the left or right active joint using scipy.signal.correlate. To compare the variation in this data set, we computed the mean cross-correlation for each fly, and compared across conditions with a paired student’s t-test.

Most statistical analyses were performed using a paired two-sided student’s t-test (scipy.stats.ttest_rel). For Chrimson activation quantification (Figure 4F), we used an unpaired two-sided student’s t-test (scipy.stats.ttest_ind) test as we were comparing across genotypes. To compare the distribution of antennal deflections during inactivation, we used Levene’s test to assess variance (scipy.stats.levene; Fig. S3D).

To compute the Fisher Information, we first fit a smooth tuning curve, d_smooth($\mathrm{\phi }$) to the plot of mean passive joint displacement difference (in degrees) as a function of wind direction for each genotype (also in degrees, using 200 cm/s wind) with one fly left out for jackknifing (see below). The tuning curve had the form of a Gaussian function multiplied by a ramp:

$$d\_smooth\left(\mathrm{\phi }\right)=m\left(\mathrm{\phi }-\delta \right)*e{}^{\frac{-{\left(\mathrm{\phi }-\mu \right)}^2}{2{\sigma }^2}}$$

where $\mathrm{\phi }$ is the wind direction, and 𝑚, 𝛿, 𝜇, and 𝜎 are free parameters. The parameters were determined by minimizing the mean squared error between this tuning curve and the experimentally measured values of $d\left(\mathrm{\phi }\right)$. The Fisher Information was computed as the mean across flies of

$$FI=\frac{{\left(d\_smooth\prime \left(\mathrm{\phi }\right)\right)}^2}{std\left(d\right)}$$

The numerator was computed by differentiating d_smooth and squaring. The denominator was computed by first computing the variance of d for each wind direction, then averaging across wind directions and flies to obtain a single measurement of variance. The standard deviation was the square root of that quantity. The mean and error bars of FI, the quantities plotted in Figure 4H, were computed by jackknifing across flies, using the formula:

$$\sqrt{\frac{n-1}{n}var\left(FI\right)}$$

where 𝑛 is the number of flies.

We measured movement parameters for walking behavior using custom Matlab scripts as previously reported^35,57. Trials in which flies moved less than 25mm total or for less than 5 trials were omitted from our analysis. Groundspeed was computed by taking the distance between adjacent samples divided by the 20 ms frame interval. Probability of moving was the probability of groundspeed exceeding 1 mm/s. Upwind velocity was the absolute value of the difference in y-coordinates divided by the frame interval. Angular velocity was computed by dividing the absolute value of the difference in unwrapped orientation by the frame interval. Curvature is plotted as the angular velocity divided by the groundspeed. Variance is presented as standard error of the mean.

Supplementary Material

2

Video S1: Example of a unilateral active movement of one antenna, related to Figure 1. Red dot indicates time of light on and off (genotype is Canton-S>Chrimson). White dot indicates time of wind on and off.

3

Video S2: Example of bilateral antennal movements generated by activating 18D07 motor neurons with Chrimson, related to Figure 2. Red dot indicates time of light on and off.

4

Video S3: Example of bilateral antennal movements generated by activating 91F02 motor neurons with Chrimson, related to Figure 3. Red dot indicates time of light on and off.

5

XVideo S4: Example of bilateral antennal movements generated by activating 74C10 muscle with Chrimson, related to Figure 4. Red dot indicates time of light on and off.

Highlights:

1. Drosophila make diverse active antennal movements in wind and during flight

2. Genetic driver lines provide access to motor neurons and an antennal muscle

3. Two sets of motor neurons contribute to active movements with different time courses

4. Active antennal positioning can tune the gain and precision of wind encoding