Language Selection

Get healthy now with MedBeds!
Click here to book your session

Protect your whole family with Orgo-Life® Quantum MedBed Energy Technology® devices.

Advertising by Adpathway

         

 Advertising by Adpathway

An indirect wing steering muscle stabilizes flight by regulating wing pitch angle in Drosophila

1 day ago 1

PROTECT YOUR DNA WITH QUANTUM TECHNOLOGY

Orgo-Life the new way to the future

  Advertising by Adpathway

  • Loading metrics

Open Access

Peer-reviewed

Research Article

?

This is an uncorrected proof.

Abstract

Flapping flight is inherently unstable, requiring Drosophila to fine-tune their wing motions on a milliseconds timescale. Pioneering studies have shown that the direct steering muscles, which attach to the wing hinge, are important for generating these rapid flight reflexes. Recent connectome data, however, reveal that indirect steering muscles or tension muscles, which alter the mechanics of the thorax, receive some of the same synaptic inputs from the sensory apparatus as the direct steering muscles. This discovery suggests that the indirect steering muscles may also be important for flight control. Here, we show that the indirect tergopleural muscles indeed contribute substantially to stabilization, particularly for large pitch perturbations. We find that for small perturbations (less than 1,000 °/s), flies modulate their wing stroke amplitude to generate lift-based corrective torques, a strategy that has been previously documented. For larger perturbations, however, we observe that Drosophila engage an additional wing degree of freedom—the wing pitch angle—to leverage additional lift and drag forces during the corrective maneuver. Quasi-steady aerodynamic simulations reveal that this strategy minimizes power consumption, anaolgous to how some mammals (including humans) adjust their steady-state gaits in a near energy-optimal manner. Using optogenetics and a control theory framework, we demonstrate that the tergopleural muscle is activated by a proportional gain component of a nonlinear proportional-integral controller responsible for determining the wing pitch angle during large perturbations. A simplified torsional spring model for the wing hinge captures the changes in the wing pitch dynamics observed during correction maneuvers by using the tergopleural muscle to adjust the rest angle of the wing. These findings provide a striking example of reflex strategy selection in time-critical behaviors and underscore the vital role of indirect steering muscles in flight stabilization.

Citation: Teoh HK, Biswas D, Li H, Leung A, Ludlow BK, Whitehead SC, et al. (2026) An indirect wing steering muscle stabilizes flight by regulating wing pitch angle in Drosophila. PLoS Biol 24(11): e3003565. https://doi.org/10.1371/journal.pbio.3003565

Academic Editor: Anders Hedenström, Lund University, SWEDEN

Received: October 1, 2025; Accepted: October 22, 2026; Published: November 18, 2026

Copyright: © 2026 Teoh et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The data set and accompanying code are available at Zenodo via https://doi.org/10.5281/zenodo.22642575.

Funding: This work was supported in part by: National Institute of Neurological Disorders and Stroke Award Number: R01NS116595 | Recipient: I.C Website: https://www.ninds.nih.gov. National Institute of Neurological Disorders and Stroke Award Number: U01NS131438 | Recipient: I.C. Website: https://www.ninds.nih.gov. National Institute of Neurological Disorders and Stroke Award Number: R01NS136988 | Recipient: M.D Website: https://www.ninds.nih.gov. Della Martin Foundation Award Number: None | Recipient: S.C.W. Website: https://giving.caltech.edu/news/accelerating-mental-health-research-2. Mong Junior Cornell Neurotech Award Number: None | Recipient: H.K.T Website: https://neurotech.cornell.edu. Air Force Office of Scientific Research Award Number: FA9550-23-1-0722 | Recipient: I.C Website: https://www.afrl.af.mil/AFOSR/. National Science Foundation Award Number: EFRI BRAID 2318081 | Recipient: M.D Website: https://www.nsf.gov/. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Abbreviations: AIC, Akaike Information Criterion; FANC, Female Adult Nerve Cord; HCS, haltere campaniform sensilla; MIP, Maximum intensity projection; NGS, normal goat serum; PD, proportional–derivative; PI, proportional-integral; pwp, pleural wing process; RMSE, root mean squared error; VNC, ventral nerve cord.

Introduction

Flapping flight, an extreme form of locomotion, demands continuous monitoring of flight stability and precise adjustment of wing motions on a millisecond time scale to maintain aerial control [1–3]. Insects like Drosophila, renowned for their rapid reflexes, demonstrate response times as short as 5 ms—some of the fastest in the animal kingdom [4]. Previous studies showed that these rapid correction strategies can be effectively described using proportional-integral (PI) controllers [4–7], which is a powerful framework for modeling sensorimotor feedback control in various animals [8–21]. For example, when flies are perturbed about the pitch axis, they adjust the forward sweep angle of the wings, which generates appropriate corrective torques that can counteract the aerial disturbances [6]. Recent work has shown that a PI controller framework that incorporates the fly’s body angular velocity (P) and the total angular displacement (I) with appropriate coefficients can predict this change in forward sweep angle. Similar PI frameworks have been used to predict the wing modulations used to counteract perturbations along the yaw [5] and roll [4] axes. In addition to being predictive of the wing modulations, these frameworks provide a foothold for understanding the neuromuscular implementations of flight control.

To maintain flight while executing this rapid and robust aerial control, the Drosophila flight motor system is divided into large asynchronous muscles that generate the high-frequency wing strokes and small synchronous wing steering muscles that implement rapid, fine adjustments to the wing motions. The large asynchronous muscles—the dorsal-longitudinal and dorsal-ventral muscles—form an antagonistic oscillatory system that produces wing beat frequencies of 200–250 Hz through reciprocal stretch activation.

The small synchronous muscles can be classified as direct or indirect steering muscles based on their attachment within the wing hinge. Direct steering muscles attach to sclerites, the skeletal elements of the hinge, while indirect muscles influence wing motion via mechanical alteration of the thorax [22–25]. The direct wing steering muscles are anatomically divided into four major groups based on the sclerites they insert on: the basalars (b1, b2, and b3), the first axillaries (i1 and i2), the third axillaries (iii1, iii3, and iii4), and the fourth axillaries (hg1, hg2, hg3, and hg4). In the context of flight, a large amount of effort has focused on understanding the role of these direct steering muscles. For example, experiments in tethered flies have determined via Ca2+ imaging the activity of the steering muscles in response to various visual stimuli [23,25]. Such experiments have shed light on which muscles may be recruited to execute various aerial maneuvers associated with the pitch, roll, and yaw degrees of freedom. In addition, since many of these direct steering muscles are innervated by a single motoneuron, studies have been able to use optogenetic manipulation to determine the functional role of the muscle being activated in free flight maneuvers. Notably, it was recently shown that the basalar motor units b1 and b2, respectively, modulate the I and P elements of the PI controller for the pitch degree of freedom [7]. As these discoveries highlight, this active area of study is rapidly advancing our understanding of the neuromuscular organization that distributes wing actuation across this flight motor system to generate such precise maneuvering and control in Drosophila.

Surprisingly, however, the role of the indirect wing steering muscles—tergopleurals (tp), pleurosternals (ps), and tergotrochanter (tt)—in flight control remains largely unexplored. Based on past morphological studies, the indirect wing steering muscles’ influence on wing motion is thought to arise from alterations to the mechanics of the thorax. Recent research has started to elucidate their crucial role in courtship. For example, the indirect wing steering muscles were shown to be important in modulating the wing motions responsible for ‘pulse’ (200–280 Hz) and ‘sine’ (160 Hz) songs [26,27]. Both courtship song modes require sustained wing vibrations, similar to the wing flapping frequency during flight (200–250 Hz). These findings show that the indirect wing steering muscles are capable of patterning the wing profile within a wing beat cycle. Moreover, previous studies have shown that flies with chronically silenced tp muscles have their wing postures modified and their flight abilities significantly impaired [26]. Finally, premotor synaptic connectivity data [28] show that many of these muscles receive haltere sensory inputs that are shared with the direct steering muscles. Such results indicate that despite the lack of attention paid to the indirect steering muscles, they may, in fact, be important for flight.

Here, we examine the role of the tergopleural or tp muscles in corrective flight maneuvers. In most insect species, including Drosophila, the tp muscles consist of two primary components, tp1 and tp2, while in Tabanids (horse flies) and Eristalis (hoverflies), these muscles are fused into a single component [29]. The tp muscles are located laterally within the thorax, spanning the dorsal tergum and pleural wall near the wing hinge. Unlike other wing steering muscles that insert directly onto hinge sclerites, the tp muscles act indirectly on the wing hinge through deformation of the thorax. In particular, the tp muscle spans the pleural cleft, with its tendon crossing the pleural cleft superior to the notopleural cleft [30–33]. This arrangement suggests that tp contraction could tighten the pleural cleft and alter the geometry or stiffness of the thoracic structures surrounding the wing hinge. Such deformation provides a potential mechanical pathway through which tp1 activity could influence wing kinematics without acting directly on a hinge sclerite, for example, by altering the timing or extent of wing rotation and stroke reversal. Consistent with this indirect mechanical role, tp muscles have previously been proposed to modify wing angle of attack and wingbeat frequency by altering thoracic stiffness [22]. Unlike other wing steering muscles, which are each innervated by a single motoneuron, the tp1 and tp2 muscles receive innervation from their respective tp1 and tp2 motoneurons, as well as an additional neuron known as the tpn motoneuron that stochastically innervates both muscles [26,27,34]. We therefore hypothesize that tp1 contributes to the control of wing pitch and plays a role in stabilizing body orientation following perturbations. Consistent with this hypothesis, we find that these muscles are important for implementing one of the reflex strategies for stabilizing pitch, and that this tp-mediated reflex is engaged during large-amplitude pitch perturbations.

Results

Silencing the tp1 muscle alters pitch stabilization

To gain insight into the potential influence of the tp muscles (Fig 1A) on flight reflexes, we asked whether they receive shared premotor input from haltere-associated pathways with the b1 and b2 direct wing steering muscles, which have previously been implicated in pitch stabilization. We first examined recently published premotor synaptic connectivity data [28] reconstructed from the Female Adult Nerve Cord (FANC) connectome [34]. We searched for premotor neurons that receive haltere-related input—an important sensory pathway for rapid flight control [35–38]—and that innervate the b1, b2, and tp motoneurons (Fig A in S1 Appendix). We identified five premotor neurons that synapse onto either the b1 or b2 motoneurons and also innervate the tp1 motoneuron (Fig 1Bi and Fig B in S1 Appendix). Specifically, three premotor neurons innervate all three motoneurons (b1, b2, and tp1), one innervates b1 and tp1, and one innervates b2 and tp1. Notably, these shared inputs constitute only a small fraction of the total synaptic input to the tp1 motoneuron (1%), indicating that this connectivity is sparse.

thumbnail

Fig 1. The tergopleural muscle, tp1, is involved in stabilizing flight during a large pitch perturbation.

(A) Direct and tension wing steering muscles of Drosophila, with the tergopleural muscles (tp) highlighted in orange and blue. Data taken from [23]. (B) i) Identified haltere premotor neurons (white) within FANC dataset that simultaneously innervate b1(blue), b2(green), and tp1(orange) motoneurons; ii) Maximum intensity projections of the fly ventral nerve cord (VNC) expressing the split-line tp1-GAL4. Scale bars, 50 m. iii) Phalloidin-stained thoracic hemisection from a tp1 > CsChrimson fly showing wing musculature, with both phalloidin (red) and GFP (green) expression innervating the tp1 muscle (outlined with white dashed lines). Scale bars, 50 m. (C) Schematic of the experimental apparatus used to deliver optogenetic and/or mechanical perturbations to freely flying Drosophilas. The magnetic field from Helmholtz coils is used to perturb flies with a magnetic pin glued to their notum. These magnetic perturbation events are filmed at 8,000 frames/s. (D and E) Definitions of the body and wing Euler angles used to describe flight kinematics. (F, G) Body orientation along the pitch axis during correction maneuvers in the (F) small and (G) large perturbation regimes. Left: average time traces of normalized body pitch angle. The green bar indicates optogenetic inhibition; the cyan/blue bar indicates mechanical perturbation. Right: normalized body pitch angle at 40 ms. For the large perturbation condition, a Kruskal–Wallis test revealed a significant effect across groups (H = 28.7753, p < .001, ). Post hoc Dunnett’s tests showed a significant difference between control and tp1 (p = .0224), and a highly significant difference between control and tp1-alt (p < .001). For the small perturbation condition, no significant group differences were observed (H = 1.08, p = .584, ). Sample sizes for small perturbation: control n = 9, tp1 n = 16, tp1-alt n = 11; for large perturbation: control n = 26, tp1 n = 27, tp1-alt n = 26. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.22642575.

https://doi.org/10.1371/journal.pbio.3003565.g001

To determine whether this pattern is specific to FANC or represents a reproducible feature of the flight motor circuit, we extended the analysis to two additional independently reconstructed connectomes, MANC and Male CNS (Fig E in S1 Appendix). Across all three connectomes, we observed a consistent pattern: tp1 shares haltere-related premotor input with the b1/b2 steering muscles through the haltere ascending pathway, although this shared input remains relatively weak. In contrast, we did not identify equivalent shared haltere inputs between tp2 or tpn and the b1, b2, or b3 steering muscles in any of the three connectomes. The replication of this pattern across independently reconstructed datasets suggests that haltere-related input to tp1 is a reproducible feature of the flight motor circuit and further distinguishes tp1 from the other tp motoneurons. Nevertheless, given the relatively small contribution of these shared inputs to the total synaptic input onto tp1, we interpret this connectomic evidence as suggestive rather than definitive of a role for tp1 in pitch-related control.

Motivated by this connectivity, we next directly tested the functional role of the tp1 muscle in pitch stabilization. To do so, we took advantage of recently developed transgenic lines [26,27] that target the tp1 motoneuron (Fig 1Bii), which innervates the tp1 muscle (Fig 1Biii).

To examine the role of the tp1 muscle in the pitch stabilization reflex, we optogenetically silenced the tp1 motoneuron while applying a pitch-up perturbation in free flight (See “Pitch Down Perturbations” in S1 Text in S1 Appendix for the rationale behind focusing on pitch-up rather than pitch-down perturbations). We targeted the tp1 motoneuron using the split-GAL4 driver line [27], SS41052 (Fig 1Bii and 1Biii) to drive expression of GtACR1 (inhibition). In our free flight pitch perturbation behavioral assay (Fig 1C), we optogenetically silenced the tp motoneurons by delivering a 45 ms light pulse. 10 ms after the LED onset, we applied a 7 ms magnetic perturbation. We captured the fly’s pitch correction maneuver for various pitch deflection amplitudes using high-speed videography and extracted body and wing kinematics (Fig 1D and 1E; see High-speed videography in Methods) [6,39]. We discovered that for small pitch perturbations (deflection amplitude less than 10°), both the control and tp1-silenced flies were able to return to their original orientation within ~30 ms (Fig 1F). However, when presented with a large pitch perturbation (deflection amplitude greater than 20°), the control flies were able to return to their original orientation in ~40 ms, which is slightly longer due to the larger perturbation amplitude. Here, the tp1-silenced flies recovered more slowly than the control group (pink curve in Fig 1G). To control for potential contributions from off-target expression, we repeated our experiments using an alternative driver line, tp1-alt, generated from a different parental lineage [26] (see Fly stocks and fly handling in Methods). Both tp1 lines exhibit off-target expression in the VNC, central brain, and optic lobes, with some domains potentially overlapping between the two lines (Fig F in S1 Appendix). Although their overall expression patterns are not identical, this approach allows us to assess whether the observed phenotypes are robust across independent driver lines despite differences in off-target expression.

Across both driver lines, we observed a degradation in correction performance during large pitch perturbations (pink and dark blue curves in Fig 1G). This consistency supports the interpretation that the shared effects arise primarily from tp1 manipulation, although a contribution from overlapping off-target expression cannot be completely excluded. These observations led us to hypothesize that flies adopt a distinct pitch stabilization strategy in the large perturbation regime and that silencing of the tp1 muscle affects the implementation of this strategy.

Fruit flies employ an additional reflexive strategy to correct for large pitch perturbations

To assess whether fruit flies employ a distinct pitch stabilization strategy in response to large perturbations, we first characterize the wing kinematic responses across both small and large perturbation regimes (see schematics in Fig 2). We found that when a fruit fly is given a small-amplitude pitch-up perturbation of ~10°, the fly exhibits a reduction of ~20° in its front stroke angle (Fig 2A). This reduction in the front stroke angle limits the forward extent of wing motion, thereby decreasing the anteriorly generated aerodynamic torque that leads to a net corrective moment that pitches the fly forward. A consistent decrease in front stroke angle at the wingbeat corresponding to the maximum corrective response was observed across all 29 flies subjected to perturbations of less than 10° (Fig 2B). In contrast, comparisons of wing pitch angle trajectories before and during the perturbation revealed no significant differences (Fig 2C and 2D), indicating that corrective responses in this regime are primarily mediated through stroke angle modulation. These findings are consistent with previously identified pitch correction mechanisms in fruit flies [6].

thumbnail

Fig 2. Drosophila use two distinct modes to stabilize pitch.

(A, C) Left Panels: schematics showing the wing stroke and wing pitch angle definition. Right Panels: wing stroke and wing pitch angles of a fly encountering a small pitch perturbation. The fly modulates its front stroke angle in response to body pitch instability. (B, D) Averaged wing stroke and wing pitch profiles (n = 29 flies) for pre-maneuver motion (dashed line) and the largest corrective response (solid line). (E, G) Wing stroke and wing pitch angles of a fly encountering a large pitch perturbation. The fly modulates both the front stroke angle and backstroke wing pitch angle in response to the instability. (F, H) Averaged wing stroke and wing pitch profiles (n = 61 flies) for pre-maneuver motion (dashed line) and the largest corrective response (solid line). (I) backstroke wing pitch angle as a function of body deflection amplitude. The black curve represents a sigmoidal fit. Cyan points indicate experimental data for pitch deflections less than 10°, and purple points indicate data for pitch deflections greater than 20°. Gray points represent intermediate pitch deflections between 10° and 20°. (J) Change in front stroke wing angle ( front stroke) as a function of body pitch deflection amplitude. The solid black line corresponds to a linear fit to the full dataset, while dashed black lines correspond to linear fits within the small- and large-perturbation regimes. The inset shows the slopes of the linear fits in each regime for comparison. The color scheme for the data points is the same as in (I). (K) Angle of attack as a function of wing cycle for two wing profiles representing small (dashed) and large (solid) perturbation responses. Differences in the angle of attack during the backstroke are illustrated by ball-and-stick diagrams, with red and blue arrows denoting drag and lift forces, respectively. (L) Drag (red) and lift (blue) contribution to corrective pitch torque normalized by the fly’s mass, wing span, and gravitational acceleration mgR for two wing profiles representing small and large perturbation responses. The dashed line indicates no backstroke wing pitch modulation; the solid line represents a 20° backstroke wing pitch modulation. (M) Contribution of drag and lift forces to corrective torque as the backstroke wing pitch angle varies. (N) Lift force generated as front stroke and backstroke wing pitch angles vary from stable flight (black point). The yellow to red curves show combinations that produce the same corrective torques. The star marks the optimal wing kinematics based on a lift maximization strategy. (O) Similar to (N), but examining power consumption. The star marks the optimal wing kinematics based on a power minimization strategy. (P, Q) Backstroke wing pitch and front stroke responses generated using the power minimization strategy. Gray points represent experimental data from (I) and (J) for comparison. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.22642575.

https://doi.org/10.1371/journal.pbio.3003565.g002

When facing a large-amplitude pitch-up perturbation, however, flies implemented an additional corrective mechanism. For example, when a fly was subjected to a ~37° pitch perturbation, the fly not only reduced its front stroke angle by ~40°—similar to the small perturbation response (Fig 2E)—but also exhibited a ~40° decrease in wing pitch angle (Fig 2G), thereby increasing the angle of attack during the backstroke. This dual response was consistently observed across all 61 flies experiencing large pitch perturbations (Fig 2F and 2H). To characterize the transition between regimes, we plotted the change in backstroke wing pitch angle ( Backstroke Wing Pitch) against body pitch deflection ( Body Angle), which we found is reasonably well fit by a sigmoidal relationship that captures the expected asymptotic behaviors at high and low perturbation amplitudes (Fig 2I; Sigmoidal fit AIC: 686 < Linear fit AIC: 694; see Backstroke wing pitch response curve fitting in Methods). These data suggest that flies began to engage wing pitch modulation for perturbations greater than ~10°. Notably, changes in front stroke angle ( Front Stroke) were broadly similar across both perturbation regimes (Fig 2J), though the stroke response had a steeper slope for data restricted to the small perturbation regime. Together, these results revealed a transition from a low-amplitude response strategy that relied predominantly on front stroke modulation to a high-amplitude regime that engaged wing pitch modulation.

To understand how backstroke wing pitch modulation contributed to overall pitch corrective torque, we performed quasi-steady aerodynamic calculations (See Quasi-steady aerodynamic calculation in Methods) comparing the maximum torques for pitch perturbations with and without the added wing pitch response. In these simulations, we mixed and matched the wing kinematics for stroke, deviation, and wing pitch degrees of freedom. Specifically, we compared the pitch torques generated by wing kinematics when responding to a large pitch perturbation with those generated using the same stroke and deviation angles but retaining the wing pitch trajectories from pre-perturbation kinematics. Overall, the wing pitch angle contributes an additional corrective pitch torque. We examined the differences in angle of attack and aerodynamic forces between these two sets of wing kinematics (Fig 2K and 2L). We found that lowering the wing pitch angle in the large perturbation regime increased the angle of attack during the backstroke. To emphasize this effect, we included an accompanying illustration featuring a “ball-and-stick” diagram, representing the wing’s leading edge and chord, along with visualizations of the corresponding lift (blue arrows) and drag (red arrows) forces during the backstroke phase. This diagram highlights larger drag forces (shown as more prominent red arrows) and relatively unchanged lift forces (shown as comparable blue arrows) associated with wing pitch angle modulation. Further analysis of the contributions from drag- and lift-based torques across a range of wing pitch angles revealed distinct trends: drag forces contributed linearly to the corrective torques, whereas lift forces exhibited a quadratic relationship with wing pitch angle (Fig 2M). In prior investigations, it was shown that by operating near an angle of attack of 45°, where lift forces are maximal, flies are able to modify drag forces (linear dependence on angle of attack) while keeping lift forces (quadratic dependence on angle of attack) relatively unchanged. Here, we find that the wing angle of attack is sufficiently altered that both lift and drag forces contribute significantly to the corrective torque. Collectively, these results indicate that the correction strategy under large perturbations also involves wing pitch modulation.

These results raised an important question: why did fruit flies implement this strategy during these large perturbations? In principle, a fruit fly could have achieved the same corrective torque in this regime by simply increasing the modulation of its front stroke angle. However, such a modulation would have altered the total lift generated, which is essential for keeping aloft. This observation suggests that the flies might have adopted a strategy aimed at maximizing lift. Alternatively, the shift in strategy may have been driven by energetic considerations, such as minimizing power consumption. To explore these possibilities further, we developed a simulation-based model (see Quasi-steady aerodynamic calculation and Modeling the pitch correction response in Method) that allowed us to analyze the interplay between lift, power, and torque production. The model parameterizes wing strokes using variations in the backstroke pitch angle ( Backstroke Pitch Angle) and front stroke angle modulation ( Front Stroke Angle).

To investigate whether the fly was attempting to maximize its lift, we created a lift production heat map and overlaid equal-lift contours (dashed white lines and color bar in Fig 2N). Superimposed on this heat map are equal pitch torque contours (solid colored lines, ranging from yellow to red, representing dimensionless torques of , , , , , and ). The black point represents a stably hovering fly, experiencing zero aerodynamic torques and producing a total lift force sufficient to support its weight. The stars indicate the optimal adjustments to the Backstroke Pitch Angle and Front Stroke Angle needed to generate a specific torque while maximizing lift. Our analysis revealed that implementing this strategy would have entailed initially adjusting the backstroke pitch angle to achieve the desired dimensionless pitch torque. Once the fly reached its wing pitch rotational limit, here set at 40°, it would have resorted to changing its stroke angle degree of freedom (see also Fig H in S1 Appendix). This strategy, however, differed from what we observed in the experiments (Fig 2I), thereby ruling out the hypothesis that the fly’s behavior was driven by lift maximization.

Next, we plotted a power consumption heat map along with equal-power contours (dashed white lines and color bar in Fig 2O) to investigate whether the fly aimed to minimize its energy expenditure. Once again, we overlaid equal pitch torque isolines onto the heat map (white dashed lines). The black point represents a stably hovering fly, experiencing zero aerodynamic torques and producing a total lift force sufficient to support its weight. The stars indicate the optimal adjustments to the Backstroke Pitch Angle and Front Stroke Angle required to generate a specific torque while minimizing power consumption. Under the power minimization hypothesis, we found that the front stroke angle was initially adjusted to generate small dimensionless pitch torques, as this required the least power. For larger dimensionless pitch torques, both the backstroke pitch angle and the front stroke angle were utilized, as this combination minimized power expenditure along the pitch torque isolines.

To better illustrate this trend, we simulated a fly correcting for a pitch perturbation of a given magnitude and extracted the Backstroke Pitch Angle and Front Stroke Angle under the constraint of power minimization. Specifically, we calculated the pitching torque for each body pitch angle, , required to return the fly to its original orientation within a time scale of two wing beats (). Using the power consumption landscape from Fig 2O, we identified the optimal adjustments to the Backstroke Pitch Angle and Front Stroke Angle and plotted them in Fig 2P and 2Q, respectively. These model-derived plots closely resemble their experimental counterparts (Fig 2I and 2J). Notably, we observed a sigmoidal change in the Backstroke Pitch Angle and a linear increase in the Front Stroke Angle. The resulting picture suggests that the fly may employ a drag-based correction mechanism to minimize power consumption during large perturbations. Although other factors, including mechanical constraints, proprioceptive feedback, and indirect flight muscle dynamics, may contribute to the observed responses, they do not readily explain the systematic pattern of wing kinematic modulation (see “Alternative mechanisms for the observed kinematic strategy” in S1 Text in S1 Appendix).

The tergopleural muscle affects the fly’s ability to implement the large perturbation stabilization reflex

These results, in conjunction with the data from Fig 1F and 1G, suggest that the tp1 muscle plays a critical role in generating the wing kinematic changes underlying the minimal power consumption correction strategy in the large perturbation regime. To further investigate the role of the tp1 muscle in implementing these strategies, we analyzed the wing kinematic changes resulting from optogenetic inhibition and activation of the tp1 motoneuron in both small and large perturbation regimes. We focused our analysis on the normalized changes in backstroke wing pitch and front stroke angle, as these are the key contributors to the corrective torques in both strategies.

We show the results of optogenetic inhibition on these wing responses in Fig 3A – 3D. For small perturbations (Fig 3A and 3B), optogenetic inhibition of the tp1 motoneuron (pink and dark blue) did not significantly alter the wing pitch and wing stroke angles from those observed in control (black) experiments. On the other hand, we found that for large perturbations (Fig 3C and 3D), the backstroke wing pitch response of the tp1-inhibited flies (pink and dark blue) was significantly smaller than that observed in the control experiments (black). Once again, we did not observe any statistically significant differences between the peak responses in the silenced and control groups for the maximum change in normalized front stroke angle. We did, however, observe a consistent decrease in the population average response over the entire correction maneuver. These data indicated that during normal correction maneuvers, the tp1 muscle did not alter the wing kinematics in the small perturbation regime. In the large perturbation regime, this muscle primarily controlled changes to the backstroke wing pitch angle and may have modulated the wing stroke angle as well. These wing stroke changes resulted in a decrease in the proportion of tp1-inhibited flies that corrected more than 50% of body deflection amplitude at 40 ms (Fig 3E) compared to the control group.

thumbnail

Fig 3. The tp1 muscle is involved in implementing the wing pitch response.

(A, B) Left: normalized change in backstroke wing pitch () and front stroke () response for control and tp1-inhibited flies during small pitch perturbations. Right: maximum change in normalized wing pitch and wing stroke responses. (C, D) Left: backstroke wing pitch () and front stroke () responses for control and tp1-inhibited flies during large pitch perturbations. Right: maximum change in normalized wing pitch and wing stroke responses. (E) Proportion of flies correcting more than 50% of the control flies’ body deflection amplitude at 40 ms. (F, G) Left: normalized change in backstroke wing pitch () and front stroke () response for control and tp1-activated flies during small pitch perturbations. Right: maximum change in normalized wing pitch and wing stroke responses. (H, I) Left: backstroke wing pitch () and front stroke () response for control and tp1-activated flies during large pitch perturbations. Right: maximum change in normalized wing pitch and wing stroke responses. (J) Proportion of flies over-correcting more than 10% of the control flies’ body deflection amplitude at 40 ms. Statistical significance for (A)–(D) and (F)–(I) was determined via the Kruskal–Wallis test with Dunn’s post hoc multiple comparisons. Statistical significance for (E) and (J) was determined via the proportional test with Bonferroni correction (***p < .001; **p < .01; *p < .05). Full statistical results are provided in Table A in S1 Appendix. Sample sizes for optogenetic inhibition with small perturbation: control n = 12, tp1 n = 24, tp1-alt n = 15; with large perturbation: control n = 15, tp1 n = 26, tp1-alt n = 19. Sample sizes for optogenetic activation with small perturbation: control n = 16, tp1 n = 17, tp1-alt n = 11; with large perturbation: control n = 19, tp1 n = 16, tp1-alt n = 11. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.22642575.

https://doi.org/10.1371/journal.pbio.3003565.g003

Next, we investigated the effects of optogenetic activation on the wing responses (Fig 3F – 3J). We found that in the small perturbation regime (Fig 3F and 3G), optogenetic activation of the tp1 motoneuron resulted in significant changes to the maximum response of the normalized backstroke wing pitch angle (Fig 3F). For the maximum change in normalized front stroke angle, we observed slight increases during activation. While the tp1-alt line showed a statistically significant increase, and the tp1 line exhibited a similar but nonsignificant trend, we note that these results were obtained under optogenetic activation, which imposes a nonphysiological level of tp1 activity, particularly in the small perturbation regime, where tp1 may not normally be strongly engaged. Taken together, these results suggest that tp1 primarily contributes to pitch modulation under physiological conditions, while the stroke angle changes observed during activation likely reflect alterations in the normal temporal coordination of muscle recruitment within the reflex

In the large perturbation regime (Fig 3H and 3I), we observed no significant changes in either the normalized backstroke wing pitch or front stroke angle responses. The optogenetic activation of the tp1 motoneuron in the small perturbation regime indicated that additional pitch torques were generated via changes in the wing pitch angle and the wing stroke angle, which could lead to an over-correction. To test this conjecture, we counted the proportion of flies within each group that corrected beyond the 10th percentile of the control group body pitch angle at 40 ms (Fig 3J). We observed a significantly higher proportion of tp1 motoneuron-activated flies that over-corrected (also see Fig JA in S1 Appendix). The data for the large pitch perturbations were consistent with the idea that the tp1 muscle was already active in the large perturbation regime; since the tp1 muscle would already have been active in this regime, optogenetic activation did not affect the wing kinematics. Consequently, we also did not observe a significant difference in the normalized body pitch angle response when the tp1 motoneuron was activated (Fig JB in S1 Appendix).

Tergopleural muscle controls the rest angle parameter in the torsional spring model for the wing hinge

The wing hinge of the fly is a complicated joint that uses coordinated actions of the sclerite processes at the wing base, as modulated by various wing steering muscles, to generate the wing motions. Prior studies, based largely on static morphological observations across a range of dipteran species [22,29,30,32,40–43], have provided a mechanistic perspective on how wing steering muscles influence wing motion. For example, the tergopleural (tp) muscles were shown to produce small relative movements between the scutum and the pleural wing process, leading to a rotation of the first axillary sclerite. This rotation was hypothesized to determine the positioning of the radial stop on the pleural wing process, and consequently to affect the angle of attack of the wing [29].

Given the complexity of the wing hinge, here, we simplify our analysis by examining the effects of the tp muscles within a simple torsional spring framework. While this model has not been directly tested due to the challenges of measuring wing hinge stiffness and damping during free flight, it nevertheless provides a useful conceptual simplification that captures several aspects of wing stroke dynamics. For example, previous studies have shown that the dynamics of the wing pitch angle can be effectively described using a damped torsional spring model [44,45]. This approach captures the essential features of wing pitch dynamics by treating the wing hinge as a torsional spring resisting wing rotations induced by aerodynamic and gravitational torques (Fig 4A). The torsional spring torque, , is expressed as , where k is the torsional spring stiffness, c is the damping factor, and is the rest angle. Although this model is highly simplified and does not fully capture the detailed mechanics of the wing hinge, it offers a more intuitive perspective on how wing rotation might be modulated by the muscle activity through changes in the spring parameters. Here, we applied the torsional spring model [44,45] to investigate the changes in spring parameters underlying the observed wing pitch kinematics in the control group. Using a similar approach outlined in a previous study [45], we utilized the body and wing kinematics (excluding wing pitch) for each wingbeat along with a set of optimized spring parameters to best replicate the experimental wing pitch dynamics (Fig 4B and 4C). We then repeated this procedure on a stroke-by-stroke basis to examine changes in the spring parameters throughout large perturbation correction maneuvers. This fitting procedure revealed that the torsional spring rest angle (Fig 4D) and the damping factor (Fig 4E) changed significantly during the correction sequence. The torsional stiffness showed only minor changes (Fig 4F). The same parameter changes were observed across the entire control population (Fig 4G – 4I), demonstrating a robust and consistent corrective mechanism captured by the torsional spring model.

thumbnail

Fig 4. The tp1 muscle modulates the rest angle in a torsional spring model for wing pitch control.

(A) Schematic of the torsional spring model. (B) Example from a single fly showing the torsional model fit (solid curve) to pre-maneuver wing kinematics (dashed curve); inset shows the relationship between wing stroke and wing pitch. (C) Example from a single fly showing the torsional model fit (solid curve) wing kinematics associated with maximum change in wing pitch response (dashed curve); inset shows the relationship between wing stroke and wing pitch. (D) Changes in the torsional spring rest angle for the same fly shown in (B) and (C) during large perturbations; the shaded region marks the perturbation period, with black and gray circles indicating the rest angle values corresponding to the conditions in (B) and (C), respectively. (E) Changes in the damping factor for the same fly shown in (B) and (C) during large perturbations. (F) Changes in the stiffness constant for the same fly shown in (B) and (C) during large perturbations. (G–I) Maximum changes in rest angle (G), damping factor (H), and stiffness constant (I) as a function of body deflection amplitude for all control flies (n = 164). (J–L) Wing pitch profiles as functions of stroke cycle, comparing reference spring parameters (k = 50 pNm/deg, c = 10 fNm/deg/s, deg) with modified deg (J), c = 20 fNm/(deg s–1) (K), and k = 20 pNm/deg (L). The insets illustrate the relationship between wing stroke and wing pitch. (M–O) Combined effects of modifying two spring parameters on wing pitch profiles, with reference parameters in black. Note that the wing pitch profile and inset for (M) closely resemble those for the max response in (C). (P–U) Left: normalized changes in torsional spring parameters in response to pitch perturbations. Right: maximum change in normalized torsional spring parameters: (P–R) Comparison of control and tp1-inhibited flies during large perturbations. (S–U) Comparison of control and tp1-activated flies during small perturbations. Statistical significance for (P)-(U) is determined via the Kruskal–Wallis test with Dunn’s post hoc multiple comparisons. (***p < .001; **p < .01; *p < .05). Sample sizes for optogenetic inhibition with large perturbation: control n = 15, tp1 n = 26, tp1-alt n = 19. Sample sizes for optogenetic activation with small perturbation: control, n = 16, tp1 n = 17, tp1-alt n = 11. The data underlying this figure can be found in https://doi.org/10.5281/zenodo.22642575.

https://doi.org/10.1371/journal.pbio.3003565.g004

To provide intuition for how changes in the various spring parameters influenced the wing pitch profile, we conducted simplified quasi-steady aerodynamic simulations (see Torsional spring model fitting and simulation in Method) with parameters that qualitatively replicated real wing pitch kinematics, . In addition to examining the wing pitch profile over the course of the stroke cycle, we also present an inset showing wing pitch versus wing stroke. This loop features a prominent hump at the top, representing the overshoot in wing pitch angle toward 180° during the transition from the front stroke to the back stroke, as well as a similar overshoot at the lower right toward 0° during the transition from the back stroke to the front stroke. When the torsional spring rest angle () was reduced to –30° (Fig 4J), the top overshoot disappeared, and the bottom overshoot deepened. On the other hand, increasing the damping factor (c) to (Fig 4K) reduced both overshoots. Finally, decreasing the torsional stiffness (k) by 20 pNm/deg (Fig 4L) amplified both the top and bottom overshoots. These observations showed that reproducing the experimentally observed loop required modifying at least two spring parameters. Among the three possible combinations of two spring parameters shown in Fig 4M–4O, lowering the rest angle () and increasing the damping factor (c) (Fig 4M) successfully recreated the experimental loop for large perturbations (Fig 4K). Specifically, these two parameters acted in concert to decrease the top overshoot while maintaining the overshoot at the bottom right. In contrast, the other pairwise combinations (Fig 4N and 4O) resulted in more pronounced overshoots or other loop distortions.

Having identified the relevant spring parameter changes, we investigated their effect when the tp1 motoneuron was inhibited during a large perturbation correction. We found that reproducing the observed wing pitch angle dynamics requires changes in both the rest angle and the damping constant, consistent with Fig 4J. The change in normalized torsional spring rest angle () was significantly smaller when the tp1 motoneuron was inhibited compared to controls (red versus black curves in Fig 4P). While the tp1 manipulation showed a significant change in damping, this effect was not consistently observed across manipulations. In contrast, we observed no significant differences between the silenced and control groups for the normalized changes in spring stiffness () (pink and dark blue versus black curves in Fig 4R). These results indicate that the tp1 muscle primarily modulates the torsional spring rest angle (), while additional changes in damping may arise from other components of the wing hinge system that also contribute to shaping the backstroke wing pitch dynamics.

Next, we examined the impact of optogenetic activation on torsional spring parameters in the small perturbation regime. We found that activating the tp1 motoneuron significantly altered the normalized torsional spring rest angle () (Fig 4S). As with the inhibition experiments, however, the normalized changes in damping factor () and stiffness constant () remained unchanged relative to the control group. These findings further supported the idea that the tp1 muscle modulated the rest angle of the wing hinge.

The tergopleural muscle is modulated by a proportional gain term in a wing pitch PI controller

Finally, to gain an understanding of how the tp1 muscle activity was temporally modulated to produce the observed time course of wing pitch changes during large perturbations, we leveraged a control-theoretic framework established in previous work [6,7]. Past studies have shown that pitch stabilization in fruit flies could be effectively modeled using a PI controller. In this framework, the controller receives the rate of change of the body pitch angle () as input, which is sensed by the halteres, the mechanosensory organs that function as vibratory gyroscopes. The controller then outputs the front wing stroke angle at time t, computed as a linear combination of the body pitch angle () and pitch velocity () at a previous time [7], as illustrated in the orange block of Fig 5A:

(1)

where , , and represent the proportional gain, integral gain, and time delay/reflex latency, respectively. The proportional and integral gains determine the relative contributions of angular velocity and displacement to the corrective response, while the time delay corresponds to neural processing and actuation delay in the reflex pathway. In Eq. (1), the body pitch angular displacement () is defined relative to the fly’s pre-perturbation attitude and can be obtained by integrating the angular velocity signal (). This controller model could also be formulated as a proportional–derivative (PD) controller, with angular displacement as the input. However, previous studies have established that halteres are primarily sensitive to angular velocity rather than angular position, and therefore we adopted the PI convention established in the literature [4,6,7]. That said, there is no direct evidence that the angular pitch velocity signal from the haltere is integrated to produce a neural representation of angular displacement; in fact, the fly’s pitch orientation may instead be measured via other sense organs such as the ocelli [46,47], vision [48–52], or the antennae (or legs) [13,53–56].

thumbnail

Fig 5. The proportional gain in the pitch stabilization controller modulates tp1 activity.

(A) Gain-scheduling Proportional-integral (PI) controller model for rapid flight stabilization in response to the pitch perturbation. The rate of change of body pitch angle () is measured by mechanosensory organs and processed through two parallel controllers: one for wing stroke modulation (orange) and another for wing pitch modulation (pink). In each controller containing the PI motif, the input is subjected to a time delay ( = 5 ms, see “Sensitivity Analysis of in the control-theoretic model” in S1 Text in S1 Appendix) and proportional and integral gains and , respectively. We modeled the observed “preflex" response in wing pitch by incorporating an additional zero-delay integral component (depicted in a lighter shade of pink) into the wing pitch controller. The outputs of these controllers are summed to determine corrective changes in wing kinematics, generating the corrective torque to stabilize the fly. Additionally, the controllers dynamically adjust their gains based on a threshold in the change in body pitch angle, . (B, C) Change in wing stroke angle over time for simulated flies (black lines, PI controller) fitted to the experimental data (black circles representing the population average, with the gray shaded region indicating the SEM) for small (B) and large (C) mechanical perturbation. Gray dashed and solid lines in (C) denote the contributions from the P term (proportional), and the I term (integral), respectively. The light blue line in (B) denotes the PI controller model fit with parameters from small perturbation. (D, E) Change in wing pitch angle over time for simulated flies (black lines) fitted to the experimental data (black circles representing the population average, with the gray shaded region indicating the SEM) for small (D) and large (E) perturbations. The solid black line, dashed gray line, and solid gray line in (E) represent the combined passive and active controller model fit, the individual passive model fit, and the active model fit, respectively. (F, G) Change in body pitch velocity over time for simulated flies (black lines) fitted to the experimental data (black circles representing the population average, with the gray shaded region indicating the SEM) for small (F) and large (G) perturbation. (H, I) Summary statistics for wing stroke controller parameters—integral gain (, H) and proportional gain (, I) showing the meanSEM for small (light blue) and large (blue) perturbation. (J, K) Summary statistics for wing pitch active controller parameters—integral gain () and proportional gain () showing the meanSEM. (L) Summary statistics for wing pitch passive controller parameter: integral gain () showing the meanSEM. Sample sizes for small perturbation: control (control) n = 28, optogenetic activation (tp1 act) n = 37, optogenetic inhibition (tp1 inh) n = 27. Sample sizes for large perturbation: control n = 61, tp1 act n = 43, tp1 inh n = 53. The data underlying this figure can be found at https://doi.org/10.5281/zenodo.22642575.

https://doi.org/10.1371/journal.pbio.3003565.g005

We found that the PI model in Eq. (1) effectively captured the wing stroke response for the small perturbation stabilization reflex (Fig 5B). The same controller gains, however, failed to accurately reproduce experimental wing stroke data for large perturbations (light blue curve, Fig 5C). This discrepancy suggests that a fixed-gain controller is insufficient and that a gain-scheduling approach, where the controller gains depend on the perturbation regime, may better capture the observed dynamics (Fig 5A, bottom). Once the controller gain is adjusted to account for this difference in regime, we find that the wing stroke response is well captured by the controller (black line, Fig 5C).

To account for the changes in the wing pitch angles, we introduce an additional control block (pink block in Fig 5A)—which regulates wing pitch based on the rate of change of the body pitch angle. For small pitch perturbations (light blue pathway in Fig 5A), this block is inactive (Fig 5D). For the large perturbation regime, this controller features a PI architecture with a time delay, consistent with that for stroke control. Interestingly, careful inspection of the wing pitch modulation shows an instantaneous response component as indicated by a decrease in wing pitch that is nearly coincident with the onset of the mechanical perturbation. This observation suggests the possibility of a passive, mechanically mediated adjustment to the wing pitch—often referred to as a ‘preflex’ [57,58]—that occurs without the latency of neural processing. To capture this effect, we incorporate an additional zero-delay integral term (light pink block in Fig 5a). The combined controller can be expressed as:

(2)

where , , and represent the proportional gain, integral gain, and time delay of the active wing pitch controller, respectively. The last term in Eq. (2) with integral gain, represents the zero-delay (putatively passive) component. We find that this wing pitch controller, which combines active (gray solid line) and passive (gray dashed line) blocks, accurately captures the wing pitch response in the large perturbation regime (black line in Fig 5E).

To further validate the robustness of this combined controller framework, we performed closed-loop simulations of a fly experiencing a pitch perturbation and implementing a correction maneuver guided by these controllers. Here, we use a linearized form of the body pitch dynamics along with quasi-steady aerodynamic calculations to estimate the torques generated by the wing strokes. The simulation results for the body pitch velocities as a function of time for the two perturbation regimes are shown in Fig 5F and 5G. We find quantitative agreement between the experimental measurements (data points) and simulation results (black lines), particularly in the early part of the correction maneuver, further confirming the accuracy of these controllers.

This combined controller model provides a useful framework for understanding how the underlying reflex circuit modulates the tp1 muscles, which are critical for implementing the large perturbation response. Specifically, we next investigated how the wing stroke, the active wing pitch, and the low-latency (passive) wing pitch controller fits are influenced by optogenetic manipulations of the tp1 motoneuron (Fig 5H–5L). For each condition, we fit the wing stroke responses to the controller model and extract the corresponding controller gains. Consistently, across all experimental conditions—including genetic control, and optogenetic activation and inhibition—we observed a systematic decrease in the fitted controller gains, and as we transition from the small perturbation regime (light blue data) to the large perturbation regime (purple data). This trend indicates that the controller has two distinct operating modes. We first focus on the orange block of Fig 5A, corresponding to the wing stroke controller. We find that optogenetic inhibition of the tp1 motoneuron only had minor effects on the integral gain in both perturbation regimes. In contrast, tp1 activation had a large effect on the integral gain in the small perturbation regime (light blue arrow in Fig 5H). We observed only minor effects due to inhibition and activation on the proportional gain (Fig 5I). These data indicate that activation of the tp1 muscle boosts the effect of the integral gain portion of the controller for wing stroke modulation.

Next, we focus on the pink block in Fig 5A, corresponding to the wing pitch controller. We observed that optogenetic inhibition and activation of the tp1 motoneuron had only minor effects on the active block controller’s integral gain across both perturbation regimes (Fig 5J). In contrast, while tp1 inhibition did not affect the active block controller’s proportional gain in the small perturbation regime, it significantly reduced this gain in the large perturbation regime (light purple arrow in Fig 5K). Conversely, tp1 activation in the small perturbation regime markedly increased the magnitude of the active block proportional gain (light blue arrow in Fig 5K), while having only a marginal effect in the large perturbation regime. Since both silencing and activation of the tp1 motoneuron alter the active block proportional gain, these results suggest that the output of the proportional gain may be modulating the tp1 motoneuron, which in turn drives changes to the wing pitch.

Finally, we note that tp1 inhibition had minimal impact on the passive block integral gain across both perturbation regimes. Interestingly, tp1 activation led to an increase in passive integral gain during the small perturbation regime. We speculate that this shift may arise from changes in thoracic or wing mechanics induced by the unnatural activation of the tp1 motoneuron in the small perturbation regime.

Discussion

In this study, we demonstrated that Drosophila exhibit strength-dependent, compensatory strategies to recover from pitch disturbances. Through optogenetic manipulation of its motor neuron, we found that the tp1 muscle is recruited in response to large, but not small, perturbations, apparently playing a role in modulating wing pitch angle. Motivated by this insight, we extended an existing control-theoretic framework [4–7] to more accurately capture the fly’s corrective responses and found that the wing pitch dynamics are modulated by the proportional component of a PI controller. Using a simplified model of wing motion in which the base of the wing is represented as a torsion spring that deforms under the instantaneous aerodynamic torque [44,45], we found that tp1 activity was correlated with adjustments to the rest angle for wing pitch. Together, these results suggest that neuromuscular recruitment during aerial correction maps onto distinct components of canonical feedback control structures.

Interestingly, we also found that the correction strategies observed in our experiments could be closely reproduced by simulating wing kinematic adjustments under a power minimization principle. Although speculative, this finding suggests that energetic considerations may act as an additional constraint in shaping the neuromuscular control of flight. Our data do not demonstrate that this reflex pathway evolved explicitly to minimize energetic cost, but the correspondence between observed responses and power-minimizing predictions is consistent with a broader pattern in locomotor systems, where control strategies often converge on energetically favorable solutions [59–64]. Given that flapping flight is among the most energetically costly modes of locomotion, it remains plausible that energy efficiency places meaningful constraints on how pitch correction is implemented.

Our torsional spring model provides a useful and analytically tractable framework for understanding how tp1 activity might influence wing pitch. However, it necessarily abstracts away the substantial mechanical complexity of the wing hinge, which has been the focus of decades of biomechanical study. Such studies have shown that the primary rotational joint of the wing is formed by the second axillary sclerite (ax2) and its ventral articulation with the pleural wing process (pwp), which serves as the fulcrum for wing motion [25,29,32,40,41,43]. Distally, ax2 is firmly connected to the wing’s main structural vein (the radius), making its orientation critical for determining the angle of attack. Boettiger and Furshpan (1952) first proposed that consistent changes in angle of attack throughout the stroke result from a phase difference between the anterior notal wing process, which actuates ax2 anterior to its articulation with the pwp, and the scutellar lever arm, which actuates ax2 posterior to this articulation. In this view, the overall pattern of wing rotation is mechanically embedded in the hinge exoskeleton by the indirect power muscles, with fine-scale adjustments mediated by the direct steering muscles [23,25]. Although a resilin-rich flexion joint at the wing base has been identified that allows for passive modulation of stroke amplitude under inertial load at stroke reversal [25,65], no evidence has been found for any structure that on its own could function as a torsional spring specifically regulating wing pitch. In this light, the torsional spring framework should be regarded as capturing a relatively simple emergent behavior of a very complex joint. The potential role of tp1 within this framework remains unresolved. In Drosophila, genetic silencing experiments have suggested a contribution to male courtship song rather than to flight directly [26]. Anatomically, tp1 does not insert on any of the four axillary sclerites of the wing but instead makes broad contact with the mesoscutum near the convergence of the transverse and notopleural ridges [30]. From its morphology, it seems unlikely that tp1 directly sets the wing pitch rest angle. Nonetheless, given the overall complexity of the wing hinge, it is possible that tp1 contributes indirectly [22]—for example, by subtly deforming the mesoscutum and thereby modifying the tension among all the wing hinge sclerites and the muscles attached to them. Such alterations within the hinge might modify wing pitch kinematics in a manner consistent with the rest angle adjustments posited by the simplified model, thus shaping higher-order control dynamics.

Another important open question arising from our study concerns the functional role of the tp1 muscle during unperturbed, free flight. Specifically, does optogenetic manipulation of tp1 motoneurons alter wing pitch modulation in the absence of external perturbations, or does tp1 primarily act to amplify the effects of other muscles? To investigate this question, we conducted free flight optogenetic experiments (Fig L in S1 Appendix). Inhibition of tp1 produced no statistically significant changes in body or wing kinematics, suggesting that tp1 is not required for baseline flight control. Conversely, optogenetic activation led to a modest decrease in vertical body velocity and subtle front stroke changes, but had no significant effect on wing pitch angle. Together with our perturbation experiments, these results suggest that the contribution of tp1 becomes more pronounced when stronger corrective responses are required. In particular, manipulating tp1 has relatively modest effects during unperturbed flight and following small perturbations, whereas its influence on pitch stabilization is more apparent following large perturbations. We therefore propose that tp1 functions primarily as a modulatory or amplifying component within a broader steering muscle motor program, rather than independently generating the major wing pitch rotations associated with pitch correction. Previous morphological and functional studies [22,23,25,29] highlight the iv (also known as hg) muscles—which attach to the fourth axillary sclerite—as strong candidates for directly driving wing pitch rotations. Future investigations that simultaneously manipulate tp1 and iv muscles may clarify whether these muscles act cooperatively to control wing kinematics, particularly during high-demand corrective maneuvers.

Our control-theoretic analysis revealed a reduction in stroke controller gains during large perturbations (Fig 5H–5I). While this reduction in gain could reflect changes in neural circuitry, it may also arise from nonlinearities in muscle power output. Prior work has shown that muscles in other animals often operate near the steepest portion of the power–phase activation curve, where small shifts in activation timing yield large changes in mechanical output [66–73]. This steep regime enables precise control via phase modulation, particularly during unsteady maneuvers. Large perturbations, however, may shift muscle activation away from this optimal phase range into a saturated regime, where output becomes less sensitive to timing. This shift flattens the phase–power relationship, effectively reducing the system’s responsiveness and lowering controller gains. Thus, phase-dependent muscle nonlinearity provides a plausible mechanistic basis for the observed drop in control gain during extreme maneuvers.

While muscle-level nonlinearities may explain the reduction in stroke controller gains, the onset of the wing pitch response likely originates from distinct neural mechanisms. One plausible mechanism involves coincidence detection among primary sensory afferents—especially those from the halteres, which serve as gyroscopic sensors of body rotation [35,37,74]. These afferents may require temporally aligned activation across multiple sensilla to drive downstream interneurons, resulting in a dependence on perturbation strength. For example, stronger perturbations could either recruit more haltere afferents or increase the synchrony of their spiking activity, either of which could increase the likelihood of reaching the excitation threshold for tp1 motoneuron activation. Such a mechanism would function as a neural switch to activate pitch corrections when perturbations are large. This mechanism could explain the absence of tp1 engagement during steady or mildly perturbed flight and its selective activation during larger maneuvers. Future electrophysiological studies will be essential to determine whether such nonlinear integration underlies the sensorimotor transformation driving wing pitch control. Recent work has begun to explore the underlying neural pathways involved in flight control and haltere feedback, laying the groundwork for probing these mechanisms in more detail [38,75,76].

A key distinction between the present results and prior work [26] lies in the mode and timescale of neural manipulation. O’Sullivan and colleagues employed chronic genetic silencing of tpn, which resulted in a severe loss of flight capacity, likely reflecting impairments in behaviors requiring sustained or high-force motor output, such as flight initiation and takeoff. In contrast, our experiments use acute optogenetic inhibition during ongoing flight, allowing us to isolate the immediate contribution of tp1 to pitch stabilization without inducing developmental or circuit-level compensation. Furthermore, tp1 and tp2 muscles are not exclusively innervated by their corresponding motoneurons, but also receive input from the tpn motoneuron, which stochastically innervates both muscles [26]. Consequently, silencing tpn does not selectively eliminate tp1 function. Instead, the phenotype reported by O’Sullivan and colleagues is best interpreted as reflecting a broader disruption of the flight motor circuit, rather than the specific loss of tp1-mediated control.

Taken together, these findings support a specific role for tp1 in modulating wing pitch during perturbation responses. We note that this study does not include nonretinal-fed control experiments, which would provide an additional control for potential light-induced effects independent of opsin activation. While the bidirectional and specific nature of the observed kinematic changes argues against nonspecific illumination effects, we cannot fully exclude such contributions. Future experiments incorporating nonretinal-fed controls and additional genetic background controls will be important to further validate the specificity of the observed tp1-dependent effects.

While the main text focuses on the functional characterization of tp1, we also performed similar analyses of tp2 and tpn, which are presented in Figs N, O, P, Q, and R in S1 Appendix. Although this study primarily examines pitch stabilization under symmetric perturbations, tp1 and tp2 are bilateral muscles and could, in principle, contribute to yaw or roll control through asymmetric activation. Previous roll perturbation experiments [4] have shown that flies modulate wing pitch angle to generate corrective torques, and studies of yaw control [5] similarly indicate that flies actively adjust wing pitch asymmetrically to produce differential aerodynamic forces. Given that the tp muscles regulate wing pitch, we hypothesize that they may also contribute to yaw and roll control in these contexts. Testing this possibility will require experiments specifically designed to probe left–right asymmetries in muscle activation.

Collectively, this study revealed a functional role for the tp1 indirect wing steering muscle during flight stabilization that is distinct from its role during courtship [27,77]. In particular, our conjecture that the tp1 muscle enhances the response of other steering muscles is quite intuitive and consistent with our everyday experience. As anyone who has ever tripped while walking will recall, many muscles beyond those directly controlling the legs are activated during the recovery. Our novel findings shed light on how Drosophila are able to leverage the beautiful interplay between such indirect and direct steering muscles to produce their emergent elegant corrective responses to a midair trip.

Methods

Fly stocks and fly handling

Flies used for optomechanical experiments were reared in the dark at room temperature on 0.4 mM retinal food (Media Facility, HHMI Janelia Research Campus). Wild-type files used for mechanical perturbation were raised at room temperature on a standard fly medium made from yeast, agar, and sucrose with a 12-hour light/12-hour dark cycle. Female flies, 4–6 days after eclosion, were used for all flight experiments. A full list of Drosophila melanogaster stocks used in this paper is given in Table 1.

Immunohistochemistry

To visualize the nervous system, we dissected Drosophila CNS in PBS using fine forceps and placed them in 4% paraformaldehyde solution (in PBS) for 15 min at room temperature. Subsequently, we washed the CNS in PBST three times, 15 min each, before incubating the CNS in normal goat serum (NGS) for 20 min. We stained the tissue overnight at 4 °C with 1:10 mouse anti-nc82 (DSHB) and 1:1000 rabbit anti-GFP (Invitrogen no. A6455) in PBST. Then, we washed the CNS in PBST and applied a secondary antibody stain consisting of 1:250 goat anti-mouse AlexaFluor 633 (Thermo Fisher no. A21052) and 1:250 goat anti-rabbit AlexaFluor 488 (Thermo Fisher no. A11034) in PBST overnight at 4 °C. We then removed the CNS from the solution and placed them in PBS for 30 min on an oscillator at room temperature. Afterward, we mounted the CNS on a glass microscope slide in Vectashield (Vector Labs H-1000-10), covered with one #1.5, 22 22 mm coverslip, and sealed with nail polish. We imaged the CNS at 20x magnification on a Zeiss LSM 880 confocal microscope at 1 M depth resolution. Final images are presented as maximum z-projections over relevant depths.

To visualize the muscles and their innervations, flies were rinsed in 95% ethanol before they were then frozen for 3−4 min at −20 °C in a mold containing Tissue-Tek O.C.T. compound (VWR Chemicals). The frozen thoraxes were then hemisected along the midline and transferred into 4% paraformaldehyde in PBS for 30 mins. Samples were incubated with 1:10 mouse anti-nc82 (DSHB), 1:1000 rabbit anti-GFP (Invitrogen no. A6455), and Alexa 568 phalloidin (Thermo Fisher Scientific, 1:50) at 4 °C for 3−4 days. Then, we washed the CNS in PBST and applied a secondary antibody stain consisting of 1:250 goat anti-mouse AlexaFluor 633 (Thermo Fisher no. A21052) and 1:250 goat anti-rabbit AlexaFluor 488 (Thermo Fisher no. A11034) in PBST overnight at 4 °C for 5−6 days. We mounted the hemisected thoraxes 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 the CNS at 20x magnification on a Zeiss LSM 880 confocal microscope at 1 M depth resolution. Final images are presented as maximum z-projections over relevant depths.

Fly preparation

Individual flies were anesthetized on a custom-made cold plate, regulated by Peltier devices, with temperatures ranging from C to C. Subsequently, a ferromagnetic pin, measuring between 1.4 and 2 mm in length and 0.15 mm in diameter, was affixed to the flies’ notum along their sagittal plane using UV glue. The added mass of the pins is comparable to the natural mass variation of the flies and negligibly impacts the off-diagonal components of the fly’s inertia tensor. Experiments conducted with unpinned flies in free flight demonstrated that the addition of a pin did not qualitatively affect the flies’ flight kinematics. For detailed calculations, refer to [4,6].

High-speed videography

Experiments were conducted using 10–25 flies per genotype, following the preparation procedure described above. Flies were introduced into a custom-built cubic flight chamber (13 cm per side) and allowed to fly freely. Three orthogonally arranged high-speed cameras (Phantom V7.1), positioned around the center of the chamber, captured flight at 8,000 frames per second with a resolution of 704 704 pixels. The side length of the filming volume was either 2.2 or 4.3 cm, so the imaging volume occupied only a small region of the chamber, making the probability of repeatedly sampling the same individual across trials low. Accordingly, each recorded event was treated as an independent sample for analysis. Illumination was provided by near-infrared light-emitting diodes (850 ± 30 nm; Osram Platinum Dragon). Entry of flies into the imaging volume was detected using an optical trigger system [39], consisting of split and expanded beams from a 5-mW, 633-nm HeNe laser (Thorlabs, HRR050), attenuated by a neutral density filter (optical density 2.0; Thorlabs, NE20A), and detected by two photodiodes (Thorlabs, FDS100). Prior to each experimental session, the cameras were calibrated using the easyWand protocol [78].

Optomechanical perturbation experiments

For each optomechanical perturbation experiment, a cohort of 10–25 flies, each affixed with a pin to their notum, was released into the flight chamber as previously described, remaining therein for approximately 12 hours. To apply midair optomechanical perturbations, we used the optical trigger circuit outlined earlier, which drove a 45-ms duration voltage pulse to an LED driver (Thorlabs, LEDD1B). This LED driver then supplied a 1-A current to either a 625 nm red LED (Thorlabs, M625L4) or a 565 nm green LED (Thorlabs, M565L3) for optogenetic excitation (CsChrimson) or inhibition (GtACR1) experiments, respectively. Both red and green LED sources were equipped with a collimating attachment (Thorlabs, COP2-A) to produce a 50-mm-diameter beam profile. The resulting intensities were 731 and 316 W/mm2 for the red and green LEDs, respectively. The cross-sectional area of this beam was sufficiently large to ensure that the light source would unavoidably intersect a fly anywhere within the filming volume of all three cameras. Additionally, the collimation ensured uniform stimulus intensity irrespective of the fly’s position within the filming volume. To prevent external light contamination during these experiments, the entire flight apparatus was enclosed by blackout curtains. Given that flies are unlikely to initiate flight bouts in total darkness, a dim, blue fluorescent light bulb illuminated the arena during experimental trials. This protocol was also used for optogenetics experiments.

Following a 10 ms delay, the trigger initiated a 7 ms magnetic field pulse by delivering a rapid current pulse to a pair of Helmholtz coils positioned on the top and bottom faces of the flight chamber. The placement of these coils generated a roughly uniform vertical magnetic field in the center of the filming volume, activated upon a fly’s entry into this region of the flight chamber. Typical magnetic field strengths were on the order of 10–2 T. The magnetic field from the coils acts on the magnetic moment of the ferromagnetic pin glued to the fly, consequently inducing a moment about the fly’s pitch axis. To generate pitch perturbations of different magnitudes, the voltage source supplied to the Helmholtz coils was varied from 15V to 28V. Further details of this procedure are described in [4–7]. Simultaneously, the high-speed cameras recorded flight activity at 8,000 frames per second, encompassing periods before, during, and after the optogenetic stimulation and application of the magnetic field.

Flight data selection and kinematic extraction

From the data collected in both optogenetic and optomechanical perturbation experiments, we initially filtered the fly movies to ensure that each contained a fly within the field of view of all three cameras within the time range of ms, facilitating further analysis. In addition, we focused our attention only on videos where there was no evidence of a fly performing volitional maneuvers before the onset of optogenetic stimulation and mechanical perturbation. Given that our optogenetic activation/inhibition, lasting 45 ms, commences at t = 0 ms and mechanical perturbation begins at t = 10 ms, we imposed additional criteria aimed at videos where the mechanical perturbation primarily influences the body’s pitch axis. This criterion was applied to isolate corrective maneuvers related to a single rotational degree of freedom.

To extract kinematic data from the three high-speed camera views, we used a custom 3D hull reconstruction algorithm detailed in [39]. This algorithm allows us to extract 12 kinematics variables describing the orientation of a fly—the 3D position of the fly center of mass and three sets of Euler angles describing the fly body, left and right-wing- at each time point. In the event of encountering an occlusion that prevents a direct extraction of a particular kinematics variable, we estimate the missing data values by using a cubic spline interpolation. As the wing response to our perturbation is symmetric, we select one of the wings with the best-estimated wing span and wing chord vector from the wing visual hull for further analysis to constrain wing angle estimation errors due to occlusion. In our analyses, similar to [7], we filtered the raw body kinematics with a 100 Hz low-pass filter. We smoothed the raw wing kinematics using the Savitzky–Golay method. In particular, we used a 7th-order polynomial with a window size of 21 frames (2.625 ms) to smooth the wing stroke angle; for the wing deviation and rotation angles, we used a 5th-order polynomial and a window size of 11 frames (1.375 ms).

To examine the effect of optogenetic manipulation of tps motoneuron on body kinematics, as showcased in Figs 1G, 1H, 3A, 3D, 3G, and 3J), we aligned the body pitch angle time series to a standardized time interval [0,40 ms] by uniformly resampling the time series via a spline fit (MATLAB’s smoothing spline function spaps.m). The resampled time series were subtracted from their baseline, defined as the average body pitch angle from 0 ms to 10 ms. Subsequently, the body pitch kinematics were normalized by the maximum deflection amplitude before the population responses were quantified via the average of the time series.

To average wingbeat kinematics across flies—as in Fig 2B, 2D, 2F, and 2H—we isolated wingbeats from time series data using stroke angle maxima as reference points. Segmented wingbeat kinematics were then aligned to a standardized wingbeat cycle time by uniformly resampling Euler angle values via a spline fit (MATLAB’s smoothing spline function spaps.m). For extracting the maximal wing response to pitch perturbation, depicted in Fig 2I, and 2J, we identified the stroke angle minimum and wing pitch angle maximum within each wingbeat. The maximal wing responses were then located from the time series of wing features subtracted from the baseline estimate, defined as the averaged wing feature over the pre-perturbation period (0 ms to 10 ms).

To evaluate the population-wide wing response to pitch perturbation, normalized by the perturbation strength, as depicted in Figs 2I, 2J, 3B, 3C, 3E, 3F, 3H, 3I, 3K, and 3L, we employed the same approach as outlined above, identifying the front stroke angle and backstroke wing pitch angle response from the baseline. Subsequently, the wing responses were normalized by the maximum deflection amplitude before employing a windowed average estimate with a window size of 2.5 ms.

Backstroke wing pitch response curve fitting

To model our backstroke wing pitch response, to the strength of deflection amplitude , we considered both a linear fit, and a sigmoidal fit, . We compared the two models based on Akaike Information Criterion (AIC) [79–81], , where MSE is the mean squared error from our least square fitting, m is the number of parameters and n is the number of data points. AIC provides a mean for model selection based on the relative goodness of fit of different models while accounting for the number of parameters. We selected the sigmoidal model based on its lower AIC score (Sigmoid model AIC: 686.427 < Linear model AIC: 694.114).

Quasi-steady aerodynamic calculation

To estimate the aerodynamic forces and torques generated by the wing kinematics observed in our experiment, we utilize a quasi-steady aerodynamic model developed in previous studies, validated with scaled mechanical models [82–84]. In particular, we calculated the total instantaneous quasi-steady aerodynamic force on a wing:

(3)

where FL and FD are the lift and drag components of the translational force , and Frot is the rotational force. Similar to [7,85], we did not include the contribution from the added-mass effect owing to the large error in estimating the second derivative of the angle of attack from experimental data. Their full expressions are:

(4)

where is the air density; S is the wing area; is the velocity of the wing tip; is the nondimensionalized second moment of wing area; is the wing angle of attack; CL and CD are the lift and drag coefficients; Crot the is rotational force coefficient; is the angular velocity of the wing about its spanwise axis; c is the mean wing chord length; R is the wing span length; and are the nondimensionalized span and chord; and , , and are unit vectors in the directions of the lift, drag, and wing surface normal. The CL and CD coefficients are given by:

(5)

where , , and are dimensionless parameters best fit for Drosophila flight. We summarized all the values used in our computation in Table D in S1 Appendix.

To calculate the aerodynamic torque, Taero produced by a flapping wing, we approximate the wing center of pressure to be 70% along the wing span [86]:

(6)

where R is the wing span length, is the wing span unit vector, Faero is the total aerodynamic force produced by the wing, and is the vector from the fly center of mass to the wing hinge. Similar to [6,7], as we are primarily concerned about the pitching torque, we express the wing hinge vector to be purely along the fly’s long body axis, .

To determine the power required to generate a particular set of desired wing kinematics, we estimate the power necessary to overcome both aerodynamic Taero,w and inertial torques Tinertial,w in the wing frame of reference:

(7)

where is the angular velocity vector of the wing in the wing frame of reference. Similar to [87], we assume that the cost for negative power is negligible and that the effect of elastic storage is minimal.

Modeling the pitch correction response

To model the selection strategy for pitch correction described in the main text, we utilized representative pre-maneuver wingbeat kinematics and corrective wingbeat kinematics to sample possible wing modulations during a corrective response to pitch perturbations. The transition between pre-maneuver and corrective wing responses was modeled using linear interpolation:

(8)

where controls the interpolation for wing stroke and wing deviation as they are strongly coupled, and controls the interpolation for wing pitch . As the deviation angle response is always correlated to the wing stroke response, we use the same as the wing stroke. Using this interpolation approach, we densely sampled pitch torques and lift forces across the range of wing kinematics observed in our experiments. For simplicity, we assumed that a fly could correct a pitch perturbation and return to its initial orientation by generating a corrective pitch torque, , over two wingbeats. To generate a specific corrective torque, , we identified an isoline that contains a series of wing modulations, , that achieved . We then traversed along the isoline by progressively adjusting the modulations, changing both front stroke and backstroke wing pitch, until we identified a combination, , that satisfies the lift maximization or power minimization constraints. This procedure was repeated over various pitch torques to explore the corresponding wing modulations under the two constraints.

Torsional spring model fitting and simulation

We followed the approach outlined in [45] to estimate the torsional spring parameters best fitting the experimentally observed wing pitch angle. Briefly, for each wingbeat, the body angular orientation , body center of mass velocity , wing stroke , wing deviation kinematics, and an initial condition for the wing pitch angle () were combined with a set of torsional spring parameters () to solve the wing pitch angle’s equation of motion,

(9)

where I1,1 and I1,2 are entries of the wing moment of inertia tensor, is the first component of the sum of aerodynamic and spring torques, , , and are the wing angular velocity vector component in the wing frame of reference, is the first component of the body angular acceleration in the wing reference frame, is the first component of the wing angular acceleration in the wing reference frame. The equation was solved using MATLAB’s ODE solver ode45. To identify the optimal spring parameters, we performed a grid search over the torsional spring parameter space, minimizing the root mean squared error (RMSE) between the calculated and experimentally measured wing pitch angles.

To examine the influence of spring parameters on wing pitch kinematics (Fig 4J–4O), we assumed a body angular orientation of , zero body center of mass velocity, and simplified wing kinematics: with for the wing stroke and for the wing deviation. Using these assumptions, we varied the spring parameters while solving Eq. 9 to compute the resulting wing pitch kinematics.

To examine the influence of tp1 motoneuron on the spring parameters,—as in Fig 4P–4U, we extracted the spring parameters for each wing beat. The spring parameters response was then subtracted from the baseline estimate, defined as the averaged spring parameters over the pre-perturbation period (0 ms to 10 ms). The spring parameters were then normalized by the maximum deflection amplitude before employing a windowed average estimate with a window size of 2.5 ms.

Control theory model fitting

For the control theory model fitting, we combined the data across all genetic lines for each condition (control, optogenetic activation, and inhibition), as the inter-line variance was negligible. We estimated controller transfer functions and hence the controller parameters using MATLAB’s tfest command, with body pitch velocity as the input and wing stroke and pitch profiles as the outputs. For tfestOptions, we applied the Levenberg-Marquardt least-squares method with ‘InitializeMethod’ set to ’all’, ‘EnforceStability’ set to ‘true’, and ‘InitialCondition’ set to ‘Estimate’. For the assessment of fitting quality, we used:

(10)

where yexp is the experimental data (wing stroke/pitch profiles), ymodel,sm is the respective model fit smoothened with a moving average filter of window length of 2.6 ms, is the mean of the experimental data,yexp, and is the L2 norm.

For modeling wing kinematics and fly mechanics, we adopted a similar approach to our previous study [7]. We approximate the output of the wing kinematics block, the pitch torque (Tpitch), as a linear combination of the differential wing stroke () and wing pitch angle () as follows:

(11)

where the coefficients are and . The negative sign of aligns with the observed decrease in differential wing pitch. For the linearized flight dynamics, we assume:

(12)

where Ipitch is the pitch moment of inertia, Cfriction is the coefficient of pitch rotational drag (Table D in S1 Appendix). , , and are the body pitch angle, velocity, and acceleration, respectively. H represents the Heaviside step function, which is nonzero only during the duration of the applied perturbation, with t0 = 10 ms and Tpulse = 7 ms. The first fitting parameter K is analogous to a spring constant that captures the relationship between body pitch angle and torque that can arise, for example, due to an offset between the center of lift and center of mass [4,5]. The second fitting parameter, A, captures the scaling that relates changes in wing kinematic output to changes in pitch torque depending on the perturbation strength and the fly lines used.

This linearized model leads to the following transfer functions for the plant (P(s)) and composite PI controller (C(s)) as follows:

Here, P(s) is the transfer function from Tpitch to the angular velocity. where s is the Laplace transform complex frequency and represents the controller time delay (5 ms for control, tp1; 4 ms for tp2 and tpn), and are the proportional and integral gain parameters of the respective controller. Strictly speaking, the integral feedback need not be computed by integrating angular velocity, but may instead arise from a direct measurement of pitch angle from the compound eyes [48–52], antennae [13,53–56], or ocelli [46,47]. Using MATLAB’s feedback command, we generated the closed-loop transfer function G(s) = P(s)/(1 + P(s)C(s)). where G(s) maps pitch perturbation, B to pitch angular velocities. The mechanical perturbation, B is modeled as a resistance-inductance (RL) circuit current profile with a time constant of 0.5 ms, where the perturbation is applied starting at 10 ms and terminated at 17 ms.

We employed a linear model with zero intercept for the wing kinematics (i.e., the estimation of aerodynamic pitch torque Tpitch), using the simulated profiles of pitch torque and lift force as described in “Modeling the Pitch Correction Response" as follows:

(13)

with p1 = 0.006629 and p2 = −0.002207.

Acknowledgments

We thank Anne von Philipsborn for providing us with UAS-Chrimson; VT22025-p65.ADZ VT29310-GAL4.DBD for the tp1-SG experiments. We are grateful to Brad Dickerson’s lab for sharing the muscle hemisection protocol. We would also like to thank Marie Suver’s lab for sharing the CNS dissection and imaging protocol.

References

  1. 1. Muijres FT, Elzinga MJ, Melis JM, Dickinson MH. Flies evade looming targets by executing rapid visually directed banked turns. Science. 2014;344(6180):172–7.
  2. 2. Taha HE, Kiani M, Hedrick TL, Greeter JSM. Vibrational control: a hidden stabilization mechanism in insect flight. Sci Robot. 2020;5(46):eabb1502. pmid:32999048
  3. 3. Chang S, Wang ZJ. Predicting fruit fly’s sensing rate with insect flight simulations. Proc Natl Acad Sci U S A. 2014;111(31):11246–51. pmid:25049376
  4. 4. Beatus T, Guckenheimer JM, Cohen I. Controlling roll perturbations in fruit flies. J R Soc Interface. 2015;12(105):20150075. pmid:25762650
  5. 5. Ristroph L, Bergou AJ, Ristroph G, Coumes K, Berman GJ, Guckenheimer J, et al. Discovering the flight autostabilizer of fruit flies by inducing aerial stumbles. Proc Natl Acad Sci U S A. 2010;107(11):4820–4. pmid:20194789
  6. 6. Whitehead SC, Beatus T, Canale L, Cohen I. Pitch perfect: how fruit flies control their body pitch angle. J Exp Biol. 2015;218(Pt 21):3508–19. pmid:26385332
  7. 7. Whitehead SC, Leone S, Lindsay T, Meiselman MR, Cowan NJ, Dickinson MH, et al. Neuromuscular embodiment of feedback control elements in Drosophila flight. Sci Adv. 2022;8(50):eabo7461. pmid:36516241
  8. 8. Mussells Pires P, Zhang L, Parache V, Abbott LF, Maimon G. Converting an allocentric goal into an egocentric steering signal. Nature. 2024;626(8000):808–18. pmid:38326612
  9. 9. Peterka RJ. Sensory integration for human balance control. Handb Clin Neurol. 2018;159:27–42. pmid:30482320
  10. 10. Stöckl AL, Kihlström K, Chandler S, Sponberg S. Comparative system identification of flower tracking performance in three hawkmoth species reveals adaptations for dim light vision. Philos Trans R Soc Lond B Biol Sci. 2017;372(1717):20160078. pmid:28193822
  11. 11. Sutton EE, Demir A, Stamper SA, Fortune ES, Cowan NJ. Dynamic modulation of visual and electrosensory gains for locomotor control. J R Soc Interface. 2016;13(118):20160057. pmid:27170650
  12. 12. Sponberg S, Dyhr JP, Hall RW, Daniel TL. Luminance-dependent visual processing enables moth flight in low light. Science. 2015;348(6240):1245–8. pmid:26068850
  13. 13. Fuller SB, Straw AD, Peek MY, Murray RM, Dickinson MH. Flying Drosophila stabilize their vision-based velocity controller by sensing wind with their antennae. Proc Natl Acad Sci U S A. 2014;111(13):E1182–91. pmid:24639532
  14. 14. Schnell B, Weir PT, Roth E, Fairhall AL, Dickinson MH. Cellular mechanisms for integral feedback in visually guided behavior. Proc Natl Acad Sci U S A. 2014;111(15):5700–5. pmid:24706794
  15. 15. Dyhr JP, Morgansen KA, Daniel TL, Cowan NJ. Flexible strategies for flight control: an active role for the abdomen. J Exp Biol. 2013;216(Pt 9):1523–36. pmid:23596279
  16. 16. Sefati S, Neveln ID, Roth E, Mitchell TRT, Snyder JB, Maciver MA, et al. Mutually opposing forces during locomotion can eliminate the tradeoff between maneuverability and stability. Proc Natl Acad Sci U S A. 2013;110(47):18798–803. pmid:24191034
  17. 17. Roth E, Reiser MB, Dickinson MH, Cowan NJ. A task-level model for optomotor yaw regulation in Drosophila melanogaster: a frequency-domain system identification approach. In: 2012 IEEE 51st IEEE Conference on Decision and Control (CDC); 2012. pp. 3721–6. https://doi.org/10.1109/cdc.2012.6426231
    • 18. Jusuk Lee, Sponberg SN, Loh OY, Lamperski AG, Full RJ, Cowan NJ. Templates and anchors for antenna-based wall following in cockroaches and robots. IEEE Trans Robot. 2008;24(1):130–43.
    • 19. Lockhart DB, Ting LH. Optimal sensorimotor transformations for balance. Nat Neurosci. 2007;10(10):1329–36. pmid:17873869
    • 20. Madhav MS, Cowan NJ. The synergy between neuroscience and control theory: the nervous system as inspiration for hard control challenges. Annu Rev Control Robot Auton Syst. 2020;3(1):243–67.
    • 21. Li L, Nagy M, Amichay G, Wu R, Wang W, Deussen O, et al. Reverse engineering the control law for schooling in zebrafish using virtual reality. Sci Robot. 2025;10(101):eadq6784. pmid:40305578
    • 22. Dickinson MH, Tu MS. The function of dipteran flight muscle. Comp Biochem Physiol A Physiol. 1997;116(3):223–38.
    • 23. Lindsay T, Sustar A, Dickinson M. The function and organization of the motor system controlling flight maneuvers in flies. Curr Biol. 2017;27(3):345–58. pmid:28132816
    • 24. Trimarchi JR, Schneiderman AM. The motor neurons innervating the direct flight muscles of Drosophila melanogaster are morphologically specialized. J Comp Neurol. 1994;340(3):427–43. pmid:8188860
    • 25. Melis JM, Siwanowicz I, Dickinson MH. Machine learning reveals the control mechanics of an insect wing hinge. Nature. 2024;628(8009):795–803. pmid:38632396
    • 26. O’Sullivan A, Lindsay T, Prudnikova A, Erdi B, Dickinson M, von Philipsborn AC. Multifunctional wing motor control of song and flight. Curr Biol. 2018;28(17):2705-2717.e4. pmid:30146152
    • 27. Ehrhardt E, Whitehead SC, Namiki S, Minegishi R, Siwanowicz I, Feng K, et al. Single-cell type analysis of wing premotor circuits in the ventral nerve cord of Drosophila melanogaster. bioRxiv. 2023.
      • 28. Lesser E, Azevedo AW, Phelps JS, Elabbady L, Cook A, Syed DS, et al. Synaptic architecture of leg and wing premotor control networks in Drosophila. Nature. 2024;631(8020):369–77. pmid:38926579
      • 29. Miyan JA, Ewing AW. How Diptera move their wings: a re-examination of the wing base articulation and muscle systems concerned with flight. Philos Trans R Soc Lond B Biol Sci. 1985;311(1150):271–302.
      • 30. Wisser A, Nachtigall W. Functional-morphological investigations on the flight muscles and their insertion points in the blowfly Calliphora erythrocephala (Insecta, Diptera). Zoomorphology. 1984;104(3):188–95.
      • 31. Wisser A, Nachtigall W. Biomechanical aspects of the wing joints in flies, especially in Calliphora erythrocephala. In: Constructional morphology and evolution. Springer Berlin Heidelberg; 1991. pp. 193–207. https://doi.org/10.1007/978-3-642-76156-0_14
        • 32. Walker SM, Schwyn DA, Mokso R, Wicklein M, Müller T, Doube M, et al. In vivo time-resolved microtomography reveals the mechanics of the blowfly flight motor. PLoS Biol. 2014;12(3):e1001823. pmid:24667677
        • 33. Mokso R, Schwyn DA, Walker SM, Doube M, Wicklein M, Müller T, et al. Four-dimensional in vivo x-ray microscopy with projection-guided gating. Sci Rep. 2015;5(1):8727.
        • 34. Azevedo A, Lesser E, Phelps JS, Mark B, Elabbady L, Kuroda S, et al. Connectomic reconstruction of a female Drosophila ventral nerve cord. Nature. 2024;631(8020):360–8. pmid:38926570
        • 35. Mohren TL, Daniel TL, Eberle AL, Reinhall PG, Fox JL. Coriolis and centrifugal forces drive haltere deformations and influence spike timing. J R Soc Interface. 2019;16(153):20190035. pmid:31014202
        • 36. Nalbach G, Hengstenberg R. The halteres of the blowfly Calliphora: II. Three-dimensional organization of compensatory reactions to real and simulated rotations. J Comp Physiol A. 1994;175:695–708.
        • 37. Pringle JWS. The gyroscopic mechanism of the halteres of Diptera. Philos Trans R Soc Lond B Biol Sci. 1948;233(602):347–84.
        • 38. Dickerson BH, de Souza AM, Huda A, Dickinson MH. Flies regulate wing motion via active control of a dual-function gyroscope. Curr Biol. 2019;29(20):3517-3524.e3. pmid:31607538
        • 39. Ristroph L, Berman GJ, Bergou AJ, Wang ZJ, Cohen I. Automated hull reconstruction motion tracking (HRMT) applied to sideways maneuvers of free-flying insects. J Exp Biol. 2009;212(Pt 9):1324–35. pmid:19376953
        • 40. Boettiger EG, Furshpan E. The mechanics of flight movements in Diptera. Biol Bull. 1952;102(3):200–11.
        • 41. Pringle JWS. Insect flight Cambridge monographs in experimental biology No. 9; 1957.
          • 42. Ennos AR. A Comparative study of the flight mechanism of Diptera. J Exp Biol. 1987;127(1):355–72.
          • 43. Deora T, Gundiah N, Sane SP. Mechanics of the thorax in flies. J Exp Biol. 2017;220(Pt 8):1382–95. pmid:28424311
          • 44. Bergou AJ, Xu S, Wang ZJ. Passive wing pitch reversal in insect flight. J Fluid Mech. 2007;591:321–37.
          • 45. Beatus T, Cohen I. Wing-pitch modulation in maneuvering fruit flies is explained by an interplay between aerodynamics and a torsional spring. Phys Rev E Stat Nonlin Soft Matter Phys. 2015;92(2):022712. pmid:26382437
          • 46. Parsons MM, Krapp HG, Laughlin SB. Sensor fusion in identified visual interneurons. Curr Biol. 2010;20(7):624–8. pmid:20303270
          • 47. Krapp HG. Ocelli. Curr Biol. 2009;19(11):R435–7.
          • 48. Kim AJ, Fenk LM, Lyu C, Maimon G. Quantitative predictions orchestrate visual signaling in Drosophila. Cell. 2017;168(1–2):280-294.e12. pmid:28065412
          • 49. Suver MP, Huda A, Iwasaki N, Safarik S, Dickinson MH. An array of descending visual interneurons encoding self-motion in Drosophila. J Neurosci. 2016;36(46):11768–80. pmid:27852783
          • 50. Sugiura H, Dickinson MH. The generation of forces and moments during visual-evoked steering maneuvers in flying Drosophila. PLoS One. 2009;4(3):e4883. pmid:19300507
          • 51. Erginkaya M, Cruz T, Brotas M, Marques A, Steck K, Nern A, et al. A competitive disinhibitory network for robust optic flow processing in Drosophila. Nat Neurosci. 2025;28(6):1241–55. pmid:40312577
          • 52. Theobald JC, Ringach DL, Frye MA. Dynamics of optomotor responses in Drosophila to perturbations in optic flow. J Exp Biol. 2010;213(Pt 8):1366–75. pmid:20348349
          • 53. Budick SA, Reiser MB, Dickinson MH. The role of visual and mechanosensory cues in structuring forward flight in Drosophila melanogaster. J Exp Biol. 2007;210(Pt 23):4092–103. pmid:18025010
          • 54. Mamiya A, Dickinson MH. Antennal mechanosensory neurons mediate wing motor reflexes in flying Drosophila. J Neurosci. 2015;35(20):7977–91. pmid:25995481
          • 55. Mamiya A, Straw AD, Tómasson E, Dickinson MH. Active and passive antennal movements during visually guided steering in flying Drosophila. J Neurosci. 2011;31(18):6900–14. pmid:21543620
          • 56. Mills KM, Cowan NJ, Suver MP. Multisensory integration for active mechanosensation in Drosophila flight. bioRxiv. 2025:2025–06.
            • 57. Jindrich DL, Full RJ. Dynamic stabilization of rapid hexapedal locomotion. J Exp Biol. 2002;205(Pt 18):2803–23. pmid:12177146
            • 58. Proctor J, Holmes P. Reflexes and preflexes: on the role of sensory feedback on rhythmic patterns in insect locomotion. Biol Cybern. 2010;102(6):513–31. pmid:20358220
            • 59. Margaria R. Sulla fisiologia e specialmente sul consumo energetico della marcia e della corsa a varie velocità ed inclinazioni del terreno. Dott. Giovanni Bardi; 1938.
              • 60. Griffin TM, Kram R, Wickler SJ, Hoyt DF. Biomechanical and energetic determinants of the walk-trot transition in horses. J Exp Biol. 2004;207(Pt 24):4215–23. pmid:15531642
              • 61. Brown GL, Seethapathi N, Srinivasan M. A unified energy-optimality criterion predicts human navigation paths and speeds. Proc Natl Acad Sci U S A. 2021;118(29):e2020327118. pmid:34266945
              • 62. Donelan JM, Kram R, Kuo AD. Mechanical and metabolic determinants of the preferred step width in human walking. Proc Biol Sci. 2001;268(1480):1985–92. pmid:11571044
              • 63. Ralston HJ. Energy-speed relation and optimal speed during level walking. Int Z Angew Physiol. 1958;17(4):277–83. pmid:13610523
              • 64. Hoyt DF, Taylor CR. Gait and the energetics of locomotion in horses. Nature. 1981;292(5820):239–40.
              • 65. Lerch S, Zuber R, Gehring N, Wang Y, Eckel B, Klass K-D, et al. Resilin matrix distribution, variability and function in Drosophila. BMC Biol. 2020;18(1):195. pmid:33317537
              • 66. Sponberg S, Daniel TL. Abdicating power for control: a precision timing strategy to modulate function of flight power muscles. Proc Biol Sci. 2012;279(1744):3958–66. pmid:22833272
              • 67. Tu MS, Daniel TL. Submaximal power output from the dorsolongitudinal flight muscles of the hawkmoth Manduca sexta. J Exp Biol. 2004;207(Pt 26):4651–62. pmid:15579560
              • 68. Josephson RK. Mechanical power output from striated muscle during cyclic contraction. J Exp Biol. 1985;114(1):493–512.
              • 69. Ahn AN, Full RJ. A motor and a brake: two leg extensor muscles acting at the same joint manage energy differently in a running insect. J Exp Biol. 2002;205(Pt 3):379–89. pmid:11854374
              • 70. Marsh RL, Olson JM. Power output of scallop adductor muscle during contractions replicating the in vivo mechanical cycle. J Exp Biol. 1994;193:139–56. pmid:7964397
              • 71. Ettema GJ. Mechanical efficiency and efficiency of storage and release of series elastic energy in skeletal muscle during stretch-shorten cycles. J Exp Biol. 1996;199(Pt 9):1983–97. pmid:8831144
              • 72. D’Août K, Curtin NA, Williams TL, Aerts P. Mechanical properties of red and white swimming muscles as a function of the position along the body of the eel Anguilla anguilla. J Exp Biol. 2001;204(Pt 13):2221–30. pmid:11507106
              • 73. Girgenrath M, Marsh RL. Power output of sound-producing muscles in the tree frogs Hyla versicolor and Hyla chrysoscelis. J Exp Biol. 1999;202(Pt 22):3225–37. pmid:10539971
              • 74. Yarger AM, Fox JL. Single mechanosensory neurons encode lateral displacements using precise spike timing and thresholds. Proc Biol Sci. 2018;285(1887):20181759. pmid:30232160
              • 75. Verbe A, Lea KM, Fox JL, Dickerson BH. Flies tune the activity of their multifunctional gyroscope. Curr Biol. 2024;34(16):3644-3653.e3. pmid:39053466
              • 76. Dhawan S, Huang Z, Dickerson BH. Neural connectivity of a computational map for fly flight control. bioRxiv. 2025:2025–05.
                • 77. Von Philipsborn AC, Liu T, Jai YY, Masser C, Bidaye SS, Dickson BJ. Neuronal control of Drosophila courtship song. Neuron. 2011;69(3):509–22.
                • 78. Theriault DH, Fuller NW, Jackson BE, Bluhm E, Evangelista D, Wu Z, et al. A protocol and calibration method for accurate multi-camera field videography. J Exp Biol. 2014;217(Pt 11):1843–8. pmid:24577444
                • 79. Akaike H. Factor analysis and AIC. Psychometrika. 1987;52(3):317–32.
                • 80. Anderson DR, Burnham KP, White GC. Comparison of Akaike information criterion and consistent Akaike information criterion for model selection and statistical inference from capture-recapture studies. J Appl Stat. 1998;25(2):263–82.
                • 81. Bozdogan H. Model selection and Akaike’s Information Criterion (AIC): the general theory and its analytical extensions. Psychometrika. 1987;52(3):345–70.
                • 82. Sane SP, Dickinson MH. The aerodynamic effects of wing rotation and a revised quasi-steady model of flapping flight. J Exp Biol. 2002;205(Pt 8):1087–96. pmid:11919268
                • 83. Whitney JP, Wood RJ. Aeromechanics of passive rotation in flapping flight. J Fluid Mech. 2010;660:197–220.
                • 84. Dickinson MH, Lehmann FO, Sane SP. Wing rotation and the aerodynamic basis of insect flight. Science. 1999;284(5422):1954–60. pmid:10373107
                • 85. Fry SN, Sayaman R, Dickinson MH. The aerodynamics of hovering flight in Drosophila. J Exp Biol. 2005;208(Pt 12):2303–18. pmid:15939772
                • 86. Birch JM, Dickinson MH. Spanwise flow and the attachment of the leading-edge vortex on insect wings. Nature. 2001;412(6848):729–33. pmid:11507639
                • 87. Berman GJ, Wang ZJ. Energy-minimizing kinematics in hovering insect flight. J Fluid Mech. 2007;582:153–68.
                Read Entire Article

                         

                        

                Start the new Vibrations with a Medbed Franchise today!  

                Protect your whole family with Quantum Orgo-Life® devices

                  Advertising by Adpathway