Abstract
The neural basis for behavioural evolution is poorly understood. Functional comparisons of homologous neurons may reveal how neural circuitry contributes to behavioural evolution, but homologous neurons cannot be identified and manipulated in most taxa. Here, we compare the function of homologous courtship song neurons by exporting neurogenetic reagents that label identified neurons in Drosophila melanogaster to D. yakuba. We found a conserved role for a cluster of brain neurons that establish a persistent courtship state. In contrast, a descending neuron with conserved electrophysiological properties drives different song types in each species. Our results suggest that song evolved, in part, due to changes in the neural circuitry downstream of this descending neuron. This experimental approach can be generalized to other neural circuits and therefore provides an experimental framework for studying how the nervous system has evolved to generate behavioural diversity.
Main text
Closely related animal species exhibit diverse behaviours, indicating that the nervous system can evolve rapidly to generate new adaptive behaviours. Little is known about the neuronal mechanisms underlying behaviour evolution. The fundamental organization of brains and neural circuits are largely conserved between related animals, suggesting that new behaviours may evolve mostly through modifications of existing neural circuitry1. Functional comparisons of homologous neurons may therefore illustrate how neural circuits evolve to cause behavioural diversification2.
We have explored this problem in species closely related to Drosophila melanogaster, allowing study of neural function underlying diverse behaviours within a taxon with a well-defined phylogenetic history3. Recently, neurons underlying many behaviours have been identified in D. melanogaster4, mainly through progress in targeting genetic reagents to small subsets of cell types5. The amenability of other Drosophila species to genetic manipulation6 provides a rare opportunity to perform functional comparisons of homologous circuits. In this study, we introduced neurogenetic reagents into non-melanogaster Drosophila species to study the function of homologous neurons in species that produce divergent courtship songs.
Courtship song evolution in the D. melanogaster species subgroup
Drosophila species display sophisticated courtship rituals often involving a male chasing the female, dancing around her, singing to her by vibrating one or both wings, waving their sometimes-spotted wings, licking the female, and other behaviours7. Here we focus on singing, which can be systematically quantified more easily than most courtship behaviours8. Females detect courtship song through vibrations of their antennal arista and they mate preferentially with males that sing intact9 courtship song of their own species10. Courtship songs evolve rapidly, presumably as a result of female-choice sexual selection, and every species sings a unique song7.
D. melanogaster courtship song contains two basic elements: trains of pulses (pulse song) and continuous hums (sine song)10. Males of D. yakuba and D. santomea, however, do not sing sine song, but they produce two distinct modes of pulse song: thud song and clack song11. Thud song is generated by unilateral wing vibration; while clack song is generated when males vibrate both wings behind them (Fig. 1a). To infer the evolutionary origins of these song types, we surveyed song in all D. melanogaster subgroup species by simultaneously recording acoustic signals and fly movements during courtship. As reported previously12,13, all of these species except D. orena produce a pulse-like song by unilateral wing vibration (Fig. 1b, Extended Data Fig. 1, and SI Movie 1), suggesting that unilateral pulse song was produced by the common ancestor of the group. We therefore reclassified thud song as pulse song, because it is similar to the ancestral unilateral pulse type. Clack song appears to be an evolutionarily new or elaborated song that evolved in the common ancestor of D. yakuba and D. santomea.
D. yakuba clack song has a higher carrier frequency (Fig. 1c) and is louder (Fig. 1d) than pulse song11,14. Males sing clack song when both the male and female are moving faster than when males sing pulse song (Fig. 1e, f). Additionally, males sing pulse song mostly when they are located directly behind females, whilst they sing clack song across a wide range of distances and positions relative to females (Fig. 1g)11. Consistent with these observations, removing motion signals by providing males with a motionless decapitated female eliminated clack song but not pulse song (Fig. 1h, i). Thus, clack song is a high frequency, high amplitude song generated often during chasing. Pulse song in D. yakuba, by contrast, is quieter and generated when females slow down and allow males to follow them closely.
Neurogenetic reagents from D. melanogaster label homologous neurons in D. yakuba
Several neurons required for D. melanogaster song have been identified4. P1 is a cluster of approximately 20 male-specific neurons per brain hemisphere that integrate multiple sensory stimuli15–17 and whose transient activity triggers a persistent courtship state18,19. Artificial activation of P1 neurons thus causes isolated males to produce many courtship behaviours, including song15,16. pIP10 is a single male-specific descending neuron per hemisphere that projects from the brain to the ventral nervous system (VNS) where it arborizes within the wing neuropil. In D. melanogaster, pIP10 acts as a command-like pathway to drive pulse song15.
We first examined whether the homologous neurons of P1 and pIP10 could be identified in non-melanogaster fly species. Here, we considered three criteria to define neurons as homologs. Homologous neurons (1) should be anatomically similar, (2) should express genetic markers that reflect a similar developmental origin, and (3) may be required to produce similar behaviours.
We first tested a subset of D. melanogaster GAL4 reagents that express in P1 and pIP10 neurons by integrating them into defined landing sites in D. yakuba6, and found that these GAL4 lines usually drove similar global expression patterns in both species (Extended Data Fig. 2). Since GAL4 reagents often drive expression in many unrelated neurons, we adopted the split-GAL4 strategy5 to identify reagents that labeled targeted neurons more cleanly. Because both the P1 and pIP10 neurons express the male-specific isoform of the sex-determination transcription factor-encoding gene fruitless (fru)15, we also generated fru expressing reagents in D. yakuba by replacing the first exon of the male-specific fru isoform with GAL4, GAL4 activating domain (AD), and DNA-binding domain (DBD) via CRISPR/Cas9-mediated homology dependent repair (HDR)20 (Extended Data Fig. 3a-c).
We screened two large D. melanogaster GAL4 driver line collections21,22 and identified split-GAL4 combinations that labeled P1 (GMR071G01-AD n VT054805-DBD, VT059450-AD n VT054805-DBD) and pIP10 (VT040556-AD n VT043047-DBD) with little extraneous expression (Extended Data Fig. 3d, f). In D. yakuba, we tested five and seven split-GAL4 combinations for P1 and pIP10 respectively. In all cases, we identified neurons with projection patterns similar to the targeted neurons (Extended Data Fig. 3e, g). We used multiple relatively clean P1 and pIP10 reagents for further behavioural analysis. Among them, the P1 reagent R71G01-AD n R15A01-DBD and the pIP10 reagent VT040346-AD n VT040556-DBD labeled male-specific neurons with the expected projection patterns almost exclusively. We show below that these labeled neurons also participate in producing courtship song in D. yakuba. Thus, based on criteria of anatomical similarity, expression of the same genetic markers (inferred because the male-specific neurons are labeled with the same GAL4 lines), and behavioural phenotypes, the labeled D. yakuba neurons appear to represent homologs of P1 and pIP10. We exploited these reagents to explore the circuitry changes contributing to song evolution.
P1 neurons drive a persistent courtship state in both D. melanogaster and D. yakuba
We first tested whether P1 neurons have a conserved role in the two species. We expressed the red-shifted channel rhodopsin CsChrimson23 in D. melanogaster and D. yakuba P1 neurons and exposed isolated males to red light. Consistent with previous reports15–19, we found that optogenetic activation of P1 neurons in D. melanogaster triggered multiple courtship behaviours, including both pulse and sine song (Extended Data Fig. 4a). Transient optogenetic activation of P1 neurons in D. yakuba caused isolated males to perform extended bouts of courtship behaviour, including abdomen quivering, wing rowing, wing scissoring, and song that consisted mainly of clacks (SI Movie 2 and Extended Data Fig. 4b). Since D. yakuba males produce pulse song mostly when they move close to females, and a moving object triggers P1-activated D. melanogaster males to court vigorously17,24, we provided optogenetically activated D. yakuba males with a recently anesthetized male D. yakuba and found that P1-activated males then produced large quantities of pulse song (Extended Data Fig. 4b). Thus, P1 neurons have retained a conserved role in eliciting a persistent courtship state in both species. In addition, since activation of P1 neurons in D. yakuba males never caused production of sine song, the neural connections downstream of P1 likely evolved to cause loss of sine song.
pIP10, the pulse song command neuron in D. melanogaster, is required for clack but not pulse song in D. yakuba
To address the role of neurons that specifically drive courtship song, we examined pIP10 function in both species. pIP10 inhibition in D. melanogaster was previously shown to reduce wing extension during courtship15. We found that pIP10 inhibition in D. melanogaster using our new split-GAL4 line caused almost complete elimination of pulse song and a small reduction in sine song produced during normal courtship (Fig. 2a). In contrast, pIP10 inhibition in D. yakuba eliminated clack song consistently across different split-GAL4 drivers and neuronal inhibitors (Fig. 2b and Extended Data Fig. 5a-c). In addition, in some treatments, pIP10 inhibition resulted in a quantitative reduction of pulse song (Extended Data Fig. 5a, c). Therefore, pIP10 is essential for pulse song production in D. melanogaster and for clack song production in D. yakuba. pIP10 also contributes to quantitative levels of sine song in D. melanogaster and pulse song in D. yakuba. It appears that pIP10 has switched its role in D. yakuba, from a descending neuron required primarily for pulse song to a neuron required primarily for clack song. These evolutionary changes likely occurred in the common ancestor of D. yakuba and D. santomea, because inhibiting pIP10 activity in D. santomea, using a non-sparse GAL4 line that labels pIP10, also blocked clack song but not pulse song during courtship (Extended Data Fig. 5d).
Activation of pIP10 drives clack(-like) and pulse song in an intensity dependent manner in both species
To further explore the role of pIP10 in D. yakuba, we optogenetically activated pIP10 neurons expressing CsChrimson. In both intact and decapitated males, pIP10 activation drove clack and pulse song in an intensity dependent manner (Fig. 2c, d, Extended Data Fig. 6a, b, and SI Movie 3); low levels of pIP10 activation drove only clack song, but higher levels drove both clack and pulse song (Extended Data Fig. 6c, d). Sine song was never elicited. Artificially activated song had an inter-clack interval within the wildtype range (~115-140 ms)11,14 at low levels of light intensity, but an abnormally short inter-clack interval at higher activation levels (Extended Data Fig. 6e). Thus, the lower activation levels may more closely mimic natural pIP10 activity levels than the higher activation levels in D. yakuba.
The observation that pIP10 can drive both clack and pulse song in D. yakuba led us to re-examine the effect of activating pIP10 in D. melanogaster. Throughout most of the light intensity range, pIP10 activation drove mainly pulse song with normal carrier frequency (mode = ~200 Hz) (Extended Data Fig. 6f, g). Sine song was elicited rarely and the probability of sine song production increased with increasing light intensities (Extended Data Fig. 6h). Activation of pIP10 in D. melanogaster at very low light levels induced flies to produce a pulse song with a high carrier frequency (Extended Data Fig. 6f, g). We performed simultaneous recording of acoustic signals and fly movements and found that most, if not all, of these high frequency pulses are generated by double-wing vibrations, mimicking the wing posture of clack song (SI Movie 4). Similarly, optogenetic activation of P1 in D. melanogaster induced both normal pulses and pulses resembling clack song in carrier frequency (Extended Data Fig. 4c, d) and wing posture (SI Movie 5).
Overall, these observations are consistent with our inactivation experiments that revealed that pIP10 is required for clack song and natural levels of pulse song in D. yakuba and required for pulse song and natural levels of sine song in D. melanogaster. In addition, the induction of clack-like song by low levels of pIP10 stimulation in D. melanogaster suggests that the common ancestor of the D. melanogaster subgroup species possessed neural circuitry capable of producing clack song and that a circuit change in the common ancestor of D. yakuba and D. santomea allowed production of clack song more readily in these species. This evolutionary change was probably driven by female-choice sexual selection because D. yakuba males whose clack song is suppressed, by expression of Kir2.1 in pIP10 neurons, had significantly lower copulation success than control males (Extended Data Fig. 5e-g).
Conserved physiological properties of pIP10 in D. melanogaster and D. yakuba
In both species, clack(-like) song is associated with low pIP10 activation levels. However, D. melanogaster produced clack-like song only in a narrow range of low activation levels, whereas D. yakuba produced clack song across a wide range of activation levels (Fig. 2e). Thus, the threshold between production of clack and pulse song has shifted between these species.
One hypothesis to explain these observations is that D. melanogaster pIP10 may exhibit higher excitability than D. yakuba, which would allow this neuron to more readily reach an activity level sufficient to drive pulse song. We therefore expressed CsChrimson in pIP10 neurons and assayed the responses of these neurons to light stimuli via ex vivo whole-cell patch-clamp recordings. Both species displayed increasing spiking rates with increasing levels of light stimulation, but we observed no statistically significant species-differences in spiking pattern at any illumination level (Fig. 3a, b) and the neurons responded very similarly in the illumination range corresponding to that used in the behavioural experiments (yellow range in Fig. 3b). We also found no differences in other electrophysiological properties, including responses to depolarizing current, resting membrane potential, spike threshold, spike amplitude, and afterhyperpolarization amplitude (Fig. 3c-i). Thus, while it is possible that there are species differences in electrophysiological properties that could not be measured at the soma, the results suggest that pIP10 electrophysiological properties are conserved between these species and cannot account for the differences in song type elicited by pIP10 activation. We also found that the onset of song in response to light stimuli is similar in both species (Fig. 2e) and red light penetrates the cuticle of both species similarly (Extended Data Fig. 7a). Thus, there are no technical differences in the ability to activate these neurons in vivo that can explain the species behavioural differences exhibited in response to pIP10 activation. These results suggest that the circuitry downstream of pIP10 has evolved to produce differential responses to similar pIP10 activity in these two species. However, we cannot exclude that species-specific differences in pIP10 neurotransmitter release, which we did not assay, may play a role in the behavioural differences elicited by similar levels of pIP10 activity.
Quantitative differences in pIP10 anatomy
One explanation for the species-specific difference in the song circuit response to pIP10 activity could be that pIP10 may synapse more extensively onto a conserved set of downstream neurons in D. melanogaster than in D. yakuba, thus resulting in an increased sensitivity to pIP10 activity in D. melanogaster. We do not yet know the synaptic partners of pIP10, so we cannot be sure that pIP10 connects with the same neurons in each species, but we can use the existing reagents to examine pIP10 arborization patterns to estimate the total number of synaptic connections in different regions.
Although the gross pIP10 projection patterns were similar in the two species (Fig. 4a, b, and SI Movie 6), we identified multiple species-specific differences (Fig. 4c-h). In the VNS, D. yakuba pIP10 displayed denser arbors than D. melanogaster pIP10 in several regions, particularly at the base of the mesothoracic triangle (red region 9 in Fig. 4f, h) and at the posterior-most descending projections in the T3 neuropil (blue region 12, magenta region 13 in Fig. 4f, h). Synaptotagmin staining, which marks pre-synaptic axons25, is observed in all of the pIP10 projections in the VNS15 (Fig. 4a and SI Movie 7), suggesting that pIP10 is largely, if not exclusively, presynaptic in the VNS. While the song circuit is incompletely known, all identified VNS song neurons co-localize with the mesothoracic triangle9,15. Thus, pIP10 pre-synaptic arbors are less dense in the region containing the song circuitry in D. melanogaster compared to D. yakuba, which suggests that total synaptic output alone cannot explain the species-specific effects of pIP10 activation. It is possible that the relative strength of pIP10 connectivity to downstream neurons has shifted between species, perhaps biasing the song circuit toward production of clack song in D. yakuba. Testing this hypothesis will require synaptic-level reconstruction of the song circuit in both species.
In addition to the differences found in the VNS, we also identified differences in pIP10 brain arborization patterns (Fig. 4). For example, D. melanogaster pIP10 extends arbors into the posterior portion of the SEZ that are not observed in D. yakuba (magenta region 7 in Fig. 4e, g). Conversely, D. yakuba displayed dense arbors extending laterally into the SEZ, whereas D. melanogaster displayed few projections in the same area15 (yellow region 4 in Fig. 4c-e, g). Synaptotagmin staining indicates that these SEZ arbors provide synaptic output in the brain (Fig. 4a and SI Movie 7). Thus, pIP10 anatomy has evolved in several ways in the brain as well, but further studies will be required to determine the evolutionary causes and functional consequences of these anatomical differences.
Potential evolutionary antecedents of clack song in D. melanogaster
The observation that artificial activation of either P1 or pIP10 can, under certain conditions, elicit clack-like song in D. melanogaster males led us to re-examine the wildtype song of these species. Male D. melanogaster mainly produce low frequency (< 250 Hz) pulses, but they also sometimes produce high frequency (> 250 Hz) pulses (Extended Data Fig. 8a)26. Lower frequency pulses (150-250Hz) are associated with large wing extension angles (~60°-90°) and higher frequency pulses (250-500Hz) are sometimes associated with smaller wing extension angles (~20°-40°) and sometimes with larger wing extension angles26 (Extended Data Fig. 8d). We observed a negative correlation between wing angle and carrier frequency specifically for higher frequency pulses produced with shallow wing angles in D. melanogaster (Extended Data Fig. 8d) and this correlation is also observed in D. simulans and D. mauritiana (Extended Data Fig. 8e, f), suggesting that this reflects a conserved mechanism for generating high frequency pulses. We therefore searched more deeply for high frequency pulses generated by wings at acute angles and found, in ~1% of all pulse events, that D. melanogaster males generate short trains of high frequency pulses (mode = 298 Hz) without obvious extension of either wing (SI Movie 8). These song events are similar to the clack-like song elicited by P1 and pIP10 activation and suggest that the evolutionary antecedents of D. yakuba clack song can be observed rarely in other species. Thus, the evolution of abundant clack song in D. yakuba and D. santomea may reflect cooption of existing neural circuitry in the common ancestor of the D. melanogaster subgroup species through quantitative changes in VNS circuitry that increases the probability that pIP10 activity drives clack song.
Discussion
Our results reveal five novel findings about the neural basis for the evolution of Drosophila courtship song. First, P1 neurons have retained a conserved function in establishing a persistent courtship state in D. melanogaster and D. yakuba. Since P1 activation results in species-specific songs, the sine song circuitry downstream of P1 has evolved between these species. Second, pIP10, a command neuron driving pulse song in D. melanogaster became largely dispensable for pulse song but essential for clack song in D. yakuba. Third, clack (-like) song and pulse song can be elicited by pIP10 activation in an intensity dependent manner in both species. This observation, together with our observation of rare D. melanogaster clack-like song, suggests that the common ancestor of these species possessed neural circuitry that could produce clacklike song. Fourth, differential responses to pIP10 activity, resulting in the production of mainly pulse song in D. melanogaster and clack song in D. yakuba, likely arose due to differences in neural circuitry downstream of pIP10. Finally, pIP10 neural anatomy has evolved both qualitatively and quantitatively, raising new questions about descending neuron evolution, structure and function.
It is curious that pIP10 drives mainly different songs in each species, when both species produce the apparently conserved pulse song. A similar observation has been reported in swim central pattern generator neurons of sea slugs, where homologous neurons play different roles in the production of homologous behaviours27. However, in both species pIP10 elicits primarily the louder song type in the context of males chasing females: pulse song in D. melanogaster8 and clack song in D. yakuba (Fig. 1d-f). In both species, when females slow down and allow males to follow them closely, they produce a quieter song, sine song in D. melanogaster8 and pulse song in D. yakuba (Fig. 1d-f). Thus, the song types driven by increasing levels of pIP10 activity correlate with similar behavioural contexts in the two species (Fig. 5). This suggests that pIP10 receives similar inputs in both species that reflect information about the behavioural context and that downstream circuity has changed so as to elicit divergent songs that are appropriate to the social context.
In both species, pIP10 activation drives different song types in an intensity dependent manner. These observations are consistent with findings that minor differences in the activity of descending pathways can drive different patterns of song circuit activity in insects29. The availability of a descending pathway that can access different song circuit activity through minor modifications may have facilitated the rapid evolution of courtship behaviours in response to female choice sexual selection. In theory, evolution of the excitability of pIP10 could have driven evolution of new song types, but in this case, it has not. Instead, our data suggest that changes in the circuitry downstream of the descending inputs has evolved to generate the diverged patterns of courtship song in response to similar social cues.
D. melanogaster provides a powerful system for dissecting the neural circuitry underlying behaviour30 and the genus Drosophila contains over 1,500 species displaying divergent adaptive behaviours31. Our functional comparative approach illustrates how we can leverage tools developed in D. melanogaster to study the homologous neural circuitry underlying the evolution of many behaviours across this genus32.
Author Contributions
Y.D., J.L.L., and D.L.S. designed the study and wrote the manuscript. Y.D. generated the D. yakuba and D. santomea genetic reagents with contributions from J.C. and D.L.S. J.L.L. and B.J.D. generated the D. melanogaster split-GAL4 lines. Y.D. and J.L.L. performed behavioural experiments. J.L.L. performed morphological quantification and electrophysiological experiments. B.A. built the behavioural rig and G.J.B. developed the k-means based song classification algorithm. M.X. performed fly husbandry and manual annotation of courtship. B.J.D. contributed to project discussion and manuscript editing.
Acknowledgements
We thank Elizabeth Kim for extensive assistance in the laboratory, Kari Close, Christina Christoforou, and Gudrun Ihrke in the Janelia Project Technical Resources team for the assistance in dissection, histological preparation, and confocal imaging, Hideo Otsuna for assistance with brain registration and VVD Viewer, and Emily Robitschek and Jaime Cervantes for help in animal tracking using ilastik. Gerry Rubin and Heather Dionne for sharing plasmids, Artyom Kopp for providing the D. orena strain, and Mala Murthy for helpful comments on the manuscript.