Multilevel visuomotor control of locomotion in Drosophila

Vision is critical for the control of locomotion, but the underlying neural mechanisms by which visuomotor circuits contribute to the movement of the body through space are yet not well understood. Locomotion engages multiple control systems, forming distinct interacting "control levels" driven by the activity of distributed and overlapping circuits. Therefore, a comprehensive understanding of the mechanisms underlying locomotion control requires the consideration of all control levels and their necessary coordination. Due to their small size and the wide availability of experimental tools, Drosophila has become an important model system to study this coordination. Traditionally, insect locomotion has been divided into studying either the biomechanics and local control of limbs, or navigation and course control. However, recent developments in tracking techniques, and physiological and genetic tools in Drosophila have prompted researchers to examine multilevel control coordination in flight and walking.


Introduction
All forms of locomotion, including flying, walking, and swimming occur through rhythmic oscillations of an animal's body or appendages. These movements require precise coordination across the body to exert forces on the environment that propel an animal in a goal-directed manner. A prime example of this remarkable and complex coordination is seen during hunting in wild big cats.
In this behavior, they must make split-second decisions about the optimal path for interception, while simultaneously controlling the movement of their legs to attain the desired speed. These animals must also move their tail and adjust the placement of their paws to maintain a stable body posture at high speeds, and their eyes and head must compensate for body movements to ensure a steady gaze for a better detection of the prey and surroundings. This extraordinary control capacity highlights how every step an animal takes is under an intricate regulation by the central nervous system (CNS).
The different aspects of the control of locomotion occur at various spatial and temporal scales and can involve different internal reference frames. Path selection, for example, is performed from the animal's perspective in space and may require integrating multiple sources of information over several seconds. In contrast, gaze stabilization reflexes, such as the vestibulo-ocular reflex (VOR), adjust eye position within a head-reference frame on a timescale of milliseconds. Despite significant research into these different "control levels", the diversity of scales and controllable elements have made it challenging to develop a comprehensive understanding of locomotion in any species. We argue that it is necessary to consider all control levels at play to fully comprehend the mechanisms underlying a single level of locomotion control. This is because locomotion involves constant trade-offs between different levels of control, for example, animals and locomotor systems must balance maneuverability and passive stability [1], which are typically modulated by the environment.
The unpredictable nature of natural environments poses significant challenges to animal locomotion. Terrain heterogeneity, air/water flow turbulence, and unpredicted external events, along with intrinsic factors such as the animal's physiological state, muscle fatigue, or sensorimotor noise, all contribute to locomotion control challenges. To overcome these challenges, nervous systems have evolved feedback systems that help the animal make internal estimations of self-motion and achieve its locomotive goals. Various sensory systems contributed to self-motion estimation, including the vestibular and proprioceptive systems, which convey information about movement and orientation of different body parts, and the visual motion system, which provides feedback on eye, head, and body velocities. Among these sensory systems, visual feedback is particularly important in fine-tuning movement in dynamic conditions. Nonetheless, our understanding of how these self-motion signals are integrated and utilized across control levels to produce the desired effect on locomotion is still limited.
The study of locomotion has been approached through various methods, each focusing on a specific level of control and emphasizing either a behavioral or circuit analysis. Typically, behavioral studies on the local (or spinal) control of locomotion often use preparations that are spatially constrained or partially immobilized [2,3]. Other levels of control, such as action selection or course control, have often been studied under unnaturalistic conditions (such as optomotor responses or flight simulators), or with low resolution for describing movements across the body [4,5]. Studying locomotion from a neural circuit perspective usually requires even larger restrictions or reduced preparations [6,7]. Nonetheless, recent technological advancements in tracking [8*,9*], miniaturization [10], and real-time systems [11,12] are providing opportunities to bridge the gap between these levels of studies and do so in more naturalistic conditions. Insects, particularly due to their small size and rigid bodies, present a promising model for investigating the link between the higher-level goal-directed aspects of locomotion and its lower-level biomechanical control. Furthermore, Drosophila offers a powerful set of experimental tools [13,14**,15**] that allow for a systematic dissection of the underlying distributed neural circuits.
Traditionally, the study of insect locomotion has been categorized into two broad fields: biomechanics and local control of limbs, and the central control of the body's movement through space, such as navigation or course control (Figure 1). This review aims to survey the literature and recent developments in comprehending the systems that underlie the control of locomotion at these two levels in Drosophila. It will particularly focus on recent research investigating visuomotor pathways providing an emerging view of the convergence and interplay between the various levels of control.

Local control of locomotion
The lower-level control of locomotion in insects has been studied for decades, especially in larger insects like locusts, stick insects, and cockroaches (for a more comprehensive review, see [16,17]). Locomotion in both vertebrates and arthropods generally results from coordinated movements of multiple appendages, such as legs or wings. This coordination results in movement patterns that occupy low-dimensional spaces [18*,19], resulting from biomechanical and energetic constraints for stable posture and cost-effective movement [20,21]. These movement patterns are then recruited based on the animal's needs. For instance, accelerating animals reduce the number of ground contacts, as observed in the transitions from wave to tetrapod to tripod gaits in hexapods [22*,23]. A fundamental question has been how this lowdimensional coordination is achieved. Two main nonexclusive hypotheses exist: the first is that coupled central pattern generator networks (CPGs) ensure proper recruitment of motor neurons across limbs and joints [24]; the second is that sensory feedback is structured to establish the necessary coordination [25]. Currently, the most widely accepted hypothesis combines both, with  Multilevel control of locomotion. In natural conditions, animals utilize multiple control systems as they locomote. These "control levels" range from movement selection and course control (under central or "higher level" control) to stepping and posture control (under local or "lower-level" control). Various premotor regions in the invertebrate central nervous system (CNS) have been suggested to play a role in different aspects of locomotion control. This figure illustrates some of the connections between invertebrate CNS areas and levels of locomotion control. the coupling between proprioceptor and GPG circuits playing a crucial role in locomotor coordination [26]. This latter idea has been explored in cats and stick insects, where distributed independent CPGs were found. Sensory information affects their phase of activity, serving as coordinator across joint-specific networks [27,28].
Proprioceptors, found in the limbs of all organisms, coordinate limb movement and force by measuring load or strain [29]. For example, movement and position signals initiate the swing-to-stance transition, while sensory signals from movement and load sensors help reinforce motor activity during stance [30]. Despite their importance, the interaction between proprioceptive signals and motor circuits remains poorly understood, partly due to the context-dependent nature of this interaction [31]. Recent technological advances in Drosophila have enabled in vivo electrophysiological recordings and calcium imaging from genetically identified ventral nerve cord neurons (VNC, the insect analogous of the spinal cord), including during locomotion [32,33,6,34]. These modern experimental systems will enable a better understanding of the context-dependent operation of local sensorimotor circuits for limb control. Combining these approaches with biomechanical models [35] and largescale reconstructions using EM [15**] or tomography [36*], will provide insights into the functional architecture and flexibility of these circuits and how central or descending signals, relaying information about other levels of control, communicate with them.

Descending control of locomotion
Insects and vertebrates without a head can still walk, but struggle with initiation and control of speed and direction [20]. Studies involving lesioning in larger insects have shown that descending neurons in the gnathal ganglion (GNG) are sufficient for initiating walking, while higher brain regions, such as the lateral accessory lobe (LAL), are essential for controlling walking speed and direction [37,38]. Interestingly, partial lesions that preserve the GNG result in prolonged periods of coordinated walking, indicating an inhibitory influence from the rest of the brain on this premotor brain region [38,39]. Conversely, lesions to the GNG greatly reduce spontaneous locomotion [39].  Figure 2c), or multiple neuron populations taking part collectively to adjust movement (Cruz and Chiappe, unpublished data, Fig. 2d). Future studies are necessary to link the genetic diversity of DNs to their orchestration during locomotion. Furthermore, it remains unknown how these diverse descending pathways interact and modulate spinal/VNC circuits.

Central control of locomotion
As important as being able to move in space, it is to do so based on when, where, and how the animal so desires. In this section, we will discuss two aspects of central locomotor control: the selection of specific movements based on a goal and the control of heading or course direction.

Selecting movements
Multiple brain circuits provide input to areas densely innervated by DNs. In insects, anatomical, behavioral, and physiological results have identified the central complex (CC), a brain region involved in motor control and navigation, as a potential locus linking sensation, volition, and action [47,48]. Recent studies have focused on how the CC integrates multimodal information to generate abstract representations of heading, which are maintained by a ring attractor-like network dynamics [49e51], and compares them to an internal goal heading representation [52*]. However, how animals convert these representations into action remains unclear.
In larger insects, CC plays a major role in locomotion control, including fine-tuning motor patterns such as curve walking [53*]. Although the connection between sensory/spatial representations and the premotor functionality of CC is still not well understood, recent anatomical studies reveal a wiring structure that could potentially integrate both functions [54*]. For example, the fan-shaped body (FB) is organized in a 2D grid [54*,55], with different columnar cell types operating as phase-shifted vector representations of the fly's movement direction, and possibly other ongoing motor variables. This anatomical configuration, along with the columnar structure of FB outputs, is proposed to be well-suited for generating detailed, goal-directed selection of motor commands [56]. Additionally, the FB is innervated by many dopaminergic tangential neurons that receive inputs from Mushroom Body (MB) output neurons known to encode valence-related signals, including reward, punishment, behavior goal, or motivation [57]. These dopaminergic neurons have also been implicated in experience-dependent behaviors [58]. Furthermore, different FB tangential neurons encode more broad internal states like locomotion, hunger, sleep, or aggression [59,60]. The organization of this brain structure, combined with an analysis of the programs of gene expression and regulation, suggests an analogy to the vertebrate striatal structure. In both cases, connections between sensory/representational pathways and specific actions are promoted/reinforced by dopaminergic signals coding for valence, locomotion, and internal state information [55]. This hypothesis raises questions about how central complex output neurons modulate command areas/DNs and how this information interacts with other premotor areas to control the animal's course [42].

Course control
The control of the animal's course is generally viewed from the perspective of controlling its rotations and/or translations. Particularly for exploratory movements, these are structured according to a principle of gaze stability [61e63], which involves the animal's ability to maintain the position of its gaze stable, enabling the acquisition of visual and spatial information from the surrounding environment [64,65]. While vision is a primary beneficiary of gaze stabilization, it is also believed to be a significant contributor to its control by complementing nonvisual mechanisms [66]. Experiments that present visual-motion perturbations, independent of animal behavior, have contributed significantly to the understanding of the relationship between vision and gaze control. In both vertebrates and insects, this type of perturbation leads to directed rotations of the eyes, known as an optokinetic response (OKR), or rotations of the head or body, known as optomotor response (OMR) [67e69]. Several control models have proposed that OKR and OMR operate through a feedback control system, whereby visually induced error signals are compensated by head or body turns to reduce gaze and course deviations [70,71]. However, concerns have been raised regarding the sufficiency of OMR mechanisms to guarantee gaze stability under self-generated visualmotion conditions [72,73]. Additionally, these models generally do not consider the interaction between visual signals and the locomotor machinery supporting continuous locomotion.
During locomotion, optic-flow fields, which are selfgenerated coherent shifts in the retinal image, can be used for self-motion estimation [74,75]. This idea is supported by the widespread presence of optic flowsensitive neurons in the animal kingdom [66,76,77]. In flies, optic-flow signals are processed in the lobula plate (LP), a brain region containing wide-field, motion-sensitive tangential cells called LPTCs, which act as matched filters for self-generated optic-flow patterns [78]. Based on their receptive fields, a class of LPTCs called HS cells has been proposed to function as part of the OMR controller for gaze and course control in flies. Indeed, artificial activation of these LPTCs induces ipsilateral head, wing, or leg movements [79,80*,81], while their inactivation leads to contralateral movements [79,82**]. These findings support the role of LPTCs in gaze stabilization. LPTCs are a major output from the LP and provide optic-flow information to higher processing centers in the central brain, as well as to neck motor neurons and descending neurons [83,84]. Interestingly, many neurons in the CC respond to similar patterns of optic-flow fields as LPTCs do, suggesting these visual neurons may convey visual feedback information to CC circuits controlling goaldirected orientation and action selection. However, the mechanisms by which LPTCs interact with other levels of locomotion control remain largely elusive.
Given the accumulated knowledge on gaze and course control circuits (including LPTCs), as well as the controllability offered by virtual reality systems, these visuomotor systems present an opportunity to investigate the mechanisms through which course control circuits interact with other levels of locomotion control.

Multilevel coordination
In the preceding sections, we examined the current understanding of various levels of control involved in fly locomotion. However, during locomotion, these levels are not independent and need to be coordinated. There are two outstanding issues that still limit our comprehensive understanding of locomotion control. The first concerns the maintenance of selected, goal-directed actions via the interaction between descending signals and the local control systems. The second relates to how information about the ongoing locomotion reaches and interacts with the central control systems for corrective actions or the selection of new ones.
Relatively more is known about the first question in the context of fly flight control. In addition to their wings, flies have equilibrium organs called halteres that have evolved from hind wings. During flight, these small organs beat in antiphase with the wings and retain most of the sensory and motor machinery of a wing. Mechanosensory fields consisting of hundreds of campaniform sensilla continuously monitor the movement of the halteres and how this structure is affected by the rotations of the fly, effectively operating as a vestibular-like sensor [85]. These sensory fields then establish strong connections with the wing steering motor neurons, closing the basis for an extremely fast and robust stabilizing reflex. However, this control reflex must be reconciled with the steering commands arising from other levels of flight control. Course control in flight is highly dependent on optic flow, raising the question of where optic-flow-related signals enter this local sensorymotor scheme of wing control.
Previous recordings in quiescent flies showed that there is optic-flow sensitivity in haltere but not in wing muscles [86]. More recently Dickerson and colleagues [87**] have further observed that optic-flow signals are passed into haltere mechanosensory neurons, and that activation of haltere muscles modulates the spike timing of wing motor neurons. This generated a model in which visual input influences wing kinematics during flight, not through direct connections, but rather by manipulating a local reflex loop through their input to the haltere sensorimotor system. The resulting visually mediated changes in haltere kinematics, in turn, influence wing kinematics through the connections between haltere mechanosensory afferents and wing steering motor neurons. With this set of experiments, they showed all the components of a system where slower optic flow-based course control signals coming from the central brain can be coordinated with a fast local sensory-motor loop.
In the context of walking flies, the mechanisms by which visual-based course control interacts with local locomotor control remain less well understood. Cruz and colleagues [88**], Figure 3a) recently employed a behavioral approach to investigate this issue, revealing that during natural explorative locomotion, a tradeoff exists between posture stability reflexes and course/gaze stability. Specifically, when there is uncertainty in leg placement in one step, flies compensate for the momentary decrease in postural stability by adjusting the placement of the leg in the subsequent step. However, this reflexive stabilization causes a deviation in the fly's walking direction, thereby jeopardizing the stability of its gaze and walking course. The authors further demonstrated that visual feedback could tune down the posture-stabilizing reflexes, hence shifting the system towards gaze stability. Therefore, they propose that visual feedback influences locomotion during exploratory walking by dampening local posture control systems. The fast control of these local reflexes suggests that visuomotor circuits in the brain may be orchestrated with the state of stepping from the VNC. All the aforementioned studies emphasize the move, when studying locomotion control, from considering an individual processing level in isolation to viewing it within a coordinated multilevel control system. This type of conceptualization and experimental approach will get us closer to understanding how systems are coordinated to achieve the extraordinary performance of natural locomotion.

Outlook
The coordination between multiple levels of control is essential even for the most seamless locomotor actions. Over the years, scientists have investigated the mechanisms underlying many of these control levels. Two notable examples from insects are the local control of stepping and the visuomotor control of heading and course. However, the interactions between these control levels have only recently received increased focus, driven by emerging technologies in whole-brain physiology, connectomics, and behavioral analysis. Particularly in Drosophila, the implementation of these technologies will facilitate our understanding of the neural processes involved in different aspects of locomotion control.

Declaration of Generative AI and AIassisted technologies in the writing process
During the preparation of this work the authors used ChatGPT in order to improve readability and language. After using this tool/service, the author(s) reviewed and edited the content as needed and take full responsibility for the content of the publication.

Declaration of competing interest
The authors declare the following financial interests/ personal relationships which may be considered as potential competing interests. M. Eugenia Chiappe reports financial support was provided by European Research Council. Tomas Lopes Cruz reports financial support was provided by Foundation for Science and Technology (Portugal). No further comments.

Data availability
Data will be made available on request. In this study, the authors utilized automated transmission electron microscopy to obtain a synapse-resolution dataset of the ventral nerve cord of an adult female Drosophila. Using this dataset, they were able to reconstruct hundreds of sensory and motor neurons responsible for regulating leg or wing movements. The availability of this open-access tool provides a valuable opportunity to comprehend the anatomical basis of local control and to gain insight into the interplay between descending and ascending information with local control loops. 1189-1196. The authors identified a descending neuron type (DNg02) that regulates wing motion during flight. This neuron type exists as a large population that collectively regulates wingbeat amplitude over a large dynamic range. This study demonstrates how a population of descending neurons can work together to control particular aspects of locomotion. The authors of the study discovered that the activity of compass neurons, specifically the EPG neurons, in walking flies reflects the current heading of the flies rather than their intended heading. Through the implementation of short disruptions to the EPG activity, they demonstrated the importance of this heading signal in maintaining stable trajectories. Flies responded to these disruptions by performing corrective turns, which further supports the notion that EPG activity guides locomotion by comparing the heading estimate with a predetermined goal.