Singing on the fly: sensorimotor integration and acoustic communication in Drosophila

Abstract

The capacity to communicate via acoustic signals is prevalent across the animal kingdom, from insects to humans. What are the neural circuit mechanisms that underlie this ability? New methods for behavioral analysis along with an unparalleled genetic toolkit have recently opened up studies of acoustic communication in the fruit fly, Drosophila melanogaster. Its nervous system comprises roughly 100,000 neurons, yet flies are able to both produce and process time-varying sounds during courtship. Just as with more complex animals, sensory feedback plays an important role in shaping communication between the sexes. Here, we review recent work in Drosophila that has laid the foundation for solving the mechanisms by which sensory information dynamically modulates behavior.

Introduction

Communication is important for quality of life, and in many cases, survival. Unsurprisingly, animals have evolved numerous strategies for exchanging information with members of their own species, and some of the most elaborate are designed to attract a mate. Whether considering the complicated nest-building of the bowerbird [1], the multi-step mating dance of the jumping spider [2], or the aerial acrobatics of the hummingbird [3], males often go to extraordinary lengths to prove their suitability to a female. One such mating signal, common to many species, is song. In general, males produce the acoustic cue, while females silently arbitrate mating decisions [4–6] — although there are notable examples of both sexes vocalizing (e.g. in flies [7], songbirds [8], and mice [9]). Courtship songs range in complexity, from the stereotyped and repetitive chirp of the cricket [10] to the highly variable and multisyllabic song of the Bengalese finch [11]. Regardless of intricacy, males must produce these songs robustly and reliably in order to compete for a mate. How does the male nervous system produce the patterns present in song and how are these patterns processed by the female to drive mate choice? Solving these questions will reveal how neural circuits, and the computations they perform, mediate social interactions.

Studies of the mechanisms underlying either song production or perception have mostly focused on a small number of non-genetic model systems [12–14]. Recently, however, Drosophila melanogaster has emerged as a strong genetic model system for studies of acoustic communication. This is largely due to the development of (1) computational methods to automate the collection and analysis of sizable behavioral datasets and (2) an impressive genetic toolkit for targeting genes and neurons that play a role in sexually dimorphic behaviors. Such advances have revealed that males pattern their songs in accordance with the dynamics of female behaviour — demonstrating that (as with humans) Drosophila acoustic communication relies heavily on sensorimotor integration [15^••]. Highly specific genetic manipulations have also begun to uncover the circuits and mechanisms that underlie acoustic behaviors. Here, we review these recent findings, their contributions to our understanding of acoustic communication, and the questions that remain unanswered.

Sensorimotor integration and the patterning of Drosophila song

During courtship, male fruit flies chase females and generate courtship songs via wing vibration (Figure 1a); these songs are important for courtship to proceed to copulation [16]. While fly song has been investigated for more than fifty years [17], new methods have permitted a more thorough statistical analysis of Drosophila song patterns. These included optimization of hardware to detect the softest elements of song, parallelization of recordings on multiple microphones (facilitating high-throughput data collection), and development of software to automatically segment recorded song into its constituent elements [18] (Figure 2a). Males structure their songs into bouts (much like songbirds [4]), and most bouts consist of alternations between two modes, pulse and sine (Figure 1b). High-throughput analytic methods identified previously unobserved patterns in song (e.g. steady increases in sine mode frequency within song bouts), eliminated previously reported spurious song patterns (e.g. KH cycles — [19]), and have proved useful for beginning to map the genes [20] and neural circuits [21] underlying song production.

Figure 1.

(a) Schematic of Drosophila melanogaster acoustic communication. Central nervous systems are indicated in gray (male) and magenta (female) shading. (b) 15s example of fly song (middle) segmented into pulse (red) and sine (blue) modes (song bout structure is indicated), with accompanying spectrogram (bottom). A short song excerpt (top) highlights pulse amplitude and inter-pulse interval (IPI). (c) Left, schematic of fly movement features. Right, summary of the influence of movement features on song bout patterning, as revealed by GLM analysis. Modified with permission from [15^••].

Figure 2.

(a) The stages of automated song segmentation. Modified with permission from [18]. (b) Drawing of setup for simultaneous tracking of flies and recording of song on multiple microphones. Modified with permission from [15^••]. (c) Drawing of ‘FlyMAD’ apparatus for closed-loop thermogenetic activation of freely behaving flies: the fly is tracked throughout a 9 cm arena and galvanometer mirrors are adjusted to target the IR laser (red line) to the fly’s current location. Modified with permission from [32^••]. (d) Sketch of fly-on-the-ball setup combining visual feedback and optogenetic neural activation. As the tethered fly walks, an optical sensor records ball movement and adjusts the image displayed accordingly. Optogenetic stimulation is targeted to the brain with an optical fiber. Modified with permission from [33^••]. (e) Top, Schematic of ‘Flyatar’ apparatus wherein a remotely actuated fly dummy is used to elicit courtship behavior from the male. Bottom, three example chases identified by an automated behavior classifier. Modified with permission from [40^••].

Males can sing hundreds of bouts, each lasting from ~50 ms to >30 s, before a mating decision occurs [15^••,18]. Because courtship is innate, Drosophila song production was long considered a fixed action pattern — a behavioral sequence that is invariant to external factors once initiated [22]. This assumption was recently tested with a large dataset of tracked movements of male and female flies and simultaneously recorded male courtship song from an array of microphones in a large chamber (Figure 2b). Generalized linear models that took as inputs the parameterized motion tracks of the flies effectively predicted the patterning of male song: from bout initiation, through switches between song modes, to the termination of each bout (Figure 1c). This approach revealed that males are most sensitive to how far away the female is (her distance) and how fast she moves (her forward speed), and that this information sculpts song structure in real time. In other words, song emerges through a sequence of rapid sensorimotor transformations. More recent studies have now shown that males not only modulate song mode, but also the amplitude of acoustic signals, to compensate for changes in female distance, a behavior previously only documented in humans and songbirds [23^•]. Together, these findings have revealed a new level of complexity in Drosophila social communication, likely arising from an equally intricate neural circuit.

The Drosophila song pathway

Two transcription factor-encoding genes, doublesex (dsx) and fruitless (fru), are known to specify the majority of sex-specific circuitry in the fruit fly (see also review in this issue: Auer and Benton) [24,25]. Optogenetic or thermogenetic activation of either fru or dsx neurons in males leads to song production or wing extension (often used as a proxy for song production), even in the absence of a female [26–28]. Of the ~2000 neurons expressing doublesex and/or fruitless, which are part of the song pathway? Intersectional genetic techniques have given researchers the ability to divide these neurons into subsets [29], and four of these neural classes (P1, pIP10, dPR1, and vPR6) have been causally linked to song production [28] (Figure 3). On the basis of their anatomy and the phenotypes generated via activation, it was postulated that the roughly 20 P1 neurons in the brain serve as ‘command neurons’ [30], triggering both sine and pulse song [15^••] through downstream effectors, including the bilateral descending neurons, pIP10 [28] and P2b [31]. pIP10 in turn likely synapses onto dPR1, and both of these neurons innervate the mesothoracic ganglion of the ventral nerve cord (VNC) — where motor neurons driving the wing muscles reside. The vPR6 cluster, also located within the mesothoracic ganglion, is a likely downstream target of pIP10 and dPR1 and a putative component of the song central pattern generator: varying the amount of activation to vPR6 neurons causes a change in the rate at which song pulses are produced, a parameter known as the inter-pulse-interval (Figure 1b).

Figure 3.

Summary diagram of components of the song production (male, gray outline) and perception (female, magenta outline) pathways. Identified neurons and regions related to song production or perception in the brain (green) and ventral nerve cord (purple) are shown (JO — Johnston’s organ, AMMC — antennal mechanosensory and motor center, VLP — ventrolateral protocerebrum, MNS — motor neurons). Dashed arrows represent connections that are supported in the literature but for which the neural pathways remain unknown. For simplification, details of male song perception and putative male forward locomotor pathways are not shown.

This putative hierarchy has been supported by a number of subsequent studies. For example, artificial and acute activation of P1 — with either thermogenetic [32^••] or optogenetic [33^••] techniques (Figure 2c) — leads to wing extension that outlasts the activation period. Conversely, wing extension elicited through pIP10 activation ends immediately after the activation period. Another cluster of ~35 neurons in the male brain (pC2l) expresses dsx and not fru — activation of these neurons in tethered males generates wing vibration on only the contralateral side, while unilateral P1 activation can drive wing extension on either the left or right side of the male [34^••]. These data collectively indicate that P1 neurons are the most upstream in the song pathway. However, because low-level activation of P1 (in combination with male olfactory cues) promotes aggression rather than courtship, P1 neurons likely have context-dependent functions in changing behavioral state [35^•].

Downstream of the pathway outlined above, which motor neurons and muscles are involved in song production and patterning? Wing movements (and thereby song production) are controlled by two sets of thoracic muscles, the indirect and direct flight muscles (Figure 3). Of the direct muscles, hg1 (a dsx-expressing muscle) is uniquely enlarged in males, and the motor neurons that innervate this muscle are specifically required for production of sine song, whereas another motor neuron, ps1, is specifically required for aspects of pulse song production [21]. The indirect flight muscles appear to control the amplitude, rather than the pattern, of song [23^•]. These data suggest that different neural circuits coordinate the production of sine versus pulse song, and that switching between these circuits, required to generate song bouts (Figure 1b), occurs at the level of the mesothoracic ganglion.

Vision and song patterning

Song pathway ‘command neurons’ should integrate courtship-relevant sensory signals to change behavioral state. Recent studies using the calcium indicator GCaMP to measure neural activity have shown that P1 neurons are active during visually induced chasing [34^••], and respond to auditory [36] and pheromonal cues [34^••,37,38]. However, increasing evidence suggests that vision is also of critical importance. For example, activating a subset of P1 neurons, or activating the majority of P1 neurons below threshold, is only able to drive wing extension when males are also presented with a visual stimulus — even if this stimulus is a square on a computer screen [34^••], a rubber band [39] or a piece of wax [32^••] (Figure 2d). In one study, the authors used an actuated magnet to simulate the female fly [40^••] (Figure 2e). Males readily chased this ‘flyatar,’ and extended their wings, despite the absence of female pheromones (although coating the magnet with pheromones extended chase duration).

From these results, it is tempting to conclude that P1 represents the interface between sensory and motor pathways. However, new evidence suggests that the situation is more complicated. Constitutive activation at different levels of the song pathway (Figure 3) reveals that some visual information intersects the pathway downstream of P1 and pIP10 neurons, most likely in the VNC [23^•]. In addition, other pathways of visual influence are indirect. Male self-motion (which is driven primarily by a visual estimate of female speed) strongly influences song patterning [15^••]. Remarkably, this remains true even for artificially activated song produced in the absence of a female, indicating a neural link between the locomotor and song circuits. Thus, some sensory signals influence the decision to initiate song at the level of the P1 neurons while others modulate song patterning (e.g., switching between song modes or changing song amplitude) within the VNC or via interactions with, as of yet unmapped, locomotor circuits.

Despite the importance of vision for male song production, the visual neurons involved remain unknown. Neurons in the identified elementary motion detection (EMD) and looming pathways [41] affect the male’s ability to follow the female while singing, but silencing these neurons has little impact on the modulation of, for example, song amplitude with female distance [23^•]. Candidates for neurons that carry distance information include small target motion detectors, so far only identified in larger insects [42]. Furthermore, while visual cues from the female influence male speed — and consequently song patterning — interpreting those cues requires knowledge of self-motion, as with visual processing during flight [43]. Once these two motor circuits (singing and walking) are mapped, it will be critical to determine how the visual and self-motor information streams are combined, and how contradictory cues are reconciled in order to correctly pattern song.

The influence of song on females

Males go to great lengths to produce songs matched to the dynamics of female motion — what aspects of song do females care about and how do they extract this information? One challenge to investigating female song perception is that, unlike male song, female receptivity is difficult to quantify. For example, the use of song playback assays (placing a wingless male with a female while playing synthetic or recorded song through a speaker and assessing copulation rates [44]), has led to conflicting results regarding which features of song females prefer [45–47]. More recently, with the advent of tracking software, experimenters have begun to use female speed to estimate song responses throughout courtship [15^••,48^••,49]. Such studies found that females either slow down or speed up in response to both sine and pulse song, dependent on their receptivity state [15^••]. In addition, the average duration (or length) of song bouts (Figure 1b), extending over ~80 s, is the strongest predictor of female speed [48^••]. Thus, as with male song patterning, female song processing can also be thought of as a continuous series of sensorimotor transformations.

Auditory processing begins in the Johnston’s organ (JO): a collection of ~480 primary sensory neurons activated in response to deflections of the auditory receiver, a structure known as the arista [50,51]. Song signals are processed by two of the five subsets of JO neurons [52] — these neurons project to the antennal mechanosensory and motor center (AMMC), and AMMC neurons in turn project largely to the ventrolateral protocerebrum (VLP) [53] (Figure 3). Whole cell patch clamp recordings of a subset of AMMC or VLP neurons (in immobilized females listening to artificial song) revealed surprisingly simple and stereotyped responses across neurons [48^••,54,55]. These similarities facilitated the development of computational models to predict neural responses to natural song, and decoding these responses accurately predicted female speed [48^••]. Whether these neurons are required for the female’s behavioral response to song remains to be determined, but neural silencing of two distinct AMMC neural types (not sampled in [48^••]) reduced copulation rates [56^•]. GCaMP imaging of those neurons’ responses suggested they are modestly tuned to the melanogaster conspecific inter-pulse interval (IPI). Thus neural tuning for short and long timescale song features appears to coexist within the AMMC and VLP. Short timescale features, like the IPI, likely indicate species identity, whereas long timescale features, like bout duration, may represent an individual male’s fitness (e.g., his ability to follow the female).

Which downstream neural circuits are responsible for extracting song information, integrating it over time, and producing a behavioral change? Although females do not produce the Fruitless protein, they generate a female-specific isoform of Doublesex, and doublesex-expressing neurons appear to be critical for regulating female courtship behaviors [57]. New genetic tools have identified two female-specific neural clusters, PC1 and PCd, which dramatically increase the probability of copulation when activated in female files, and produce a corresponding decrease in receptivity when silenced [58^••]. The PC1 neurons also appear to respond (assessed via GCaMP imaging in immobilized flies) to both auditory and pheromonal stimuli, indicating they are involved in multimodal integration, akin to P1 neurons in the male. Interestingly, PC1 (female) and P1 (male) neurons are located in the same brain region and P1 neurons in the male also respond to auditory stimuli — in particular, they are tuned for conspecific pulse song parameters [36]. These data lead to the tantalizing hypothesis that similar neural architectures are responsible for song production in the male and perception in the female. This common neural elements hypothesis has been proposed in other systems [59,60] and would explain the apparent vestigial song production circuit in the female [26].

Conclusions

Drosophila melanogaster has emerged as a premier model for studies of sensorimotor integration during communication. Genetic and computational tools have revealed new complexity in courtship behavior (e.g., Drosophila male song causes changes in female locomotion, which in turn affect the patterning of song) and have identified components of both the song production and perception pathways. But significant work remains for a comprehensive understanding of the underlying neural circuit mechanisms. Ongoing work to map and characterize neural activity in the Drosophila visual [61] and locomotor [62] pathways will facilitate solving how sensory information (related to female movements) modulates activity within the song pathway of males. However, to date, only a limited number of song neurons have been identified — other neurons may have been missed in previous studies because either (1) they do not express the transcription factors fruitless or doublesex or (2) they do not elicit a behavioral phenotype when activated in isolation. Widening the search with new genetic tools that sparsely tile all neurons [63] may solve the prior, but activating smaller subsets of neurons may make it more unlikely to observe a behavioral phenotype. Thus, in vivo imaging and recording will certainly be required to characterize both song production pathways and female song processing pathways. Ideally, mapping neural circuits underlying sensorimotor integration should take place in behaving animals, but these experiments are challenging to execute [64]. One major obstacle for studying song patterning neurons is that accessing the ventral nerve cord for calcium imaging or whole-cell patch clamp electrophysiology necessitates opening up the thorax, which consequently prevents wing motion (and song production). As with other systems [65], it should be possible use fictive behavior as a proxy for song, but the development of new methods for non-invasive neural recording would certainly facilitate studies of sensorimotor integration. In addition, better measures of female behavior are required to interpret activity in her song processing pathways. It is now possible to automatically identify subtle behaviors in freely moving flies [66,67^•]. These tools are well suited to finding a robust indicator of female receptivity — ideally one that can be observed in tethered flies during neural recordings. Despite the remaining challenges, the studies reviewed here have laid the groundwork for a complete dissection of the circuits and mechanisms underlying this example of complex social communication.