Flexible circuit mechanisms for context-dependent song sequencing

Abstract

Sequenced behaviours, including locomotion, reaching and vocalization, are patterned differently in different contexts, enabling animals to adjust to their environments. How contextual information shapes neural activity to flexibly alter the patterning of actions is not fully understood. Previous work has indicated that this could be achieved via parallel motor circuits, with differing sensitivities to context^1,2. Here we demonstrate that a single pathway operates in two regimes dependent on recent sensory history. We leverage the Drosophila song production system³ to investigate the role of several neuron types^4-7 in song patterning near versus far from the female fly. Male flies sing 'simple' trains of only one mode far from the female fly but complex song sequences comprising alternations between modes when near her. We find that ventral nerve cord (VNC) circuits are shaped by mutual inhibition and rebound excitability⁸ between nodes driving the two song modes. Brief sensory input to a direct brain-to-VNC excitatory pathway drives simple song far from the female, whereas prolonged input enables complex song production via simultaneous recruitment of functional disinhibition of VNC circuitry. Thus, female proximity unlocks motor circuit dynamics in the correct context. We construct a compact circuit model to demonstrate that the identified mechanisms suffice to replicate natural song dynamics. These results highlight how canonical circuit motifs^8,9 can be combined to enable circuit flexibility required for dynamic communication.

Fig. 1. Context-dependent differences in song sequencing in D. melanogaster.

a, Drosophila male courtship song is structured into bouts comprising two main modes: ‘pulse’ (p) and ‘sine’ (s). We focus on song bout patterning, although the duration, amplitude and spectral modulation of pulse and sine trains constitute other sources of song variability^, (Extended Data Fig. 1c). b, Song bouts consist of either simple pulse or sine trains, or complex sequences involving continuous alternations between modes. c, Population-averaged probability (median ± median absolute deviation from the median) of wild-type males singing simple pulse, simple sine or complex bouts at a given mfDist. The grey vertical line indicates the distance threshold of 4 mm used to define far and near song bouts. d, The distribution of song sequence types differs far versus near the female. Complex p are complex bouts starting in pulse mode. Complex s are complex bouts starting in sine mode. Both far from and near the female, simple pulse bouts constituted the majority of all bouts (more than 95% and around 55%, respectively), followed by complex ‘ps...’ bouts near the female (around 30%). Simple sine bouts constituted the minority of bouts at all distances. e, P(female location) during the production of simple pulse (red, right half) versus complex bouts (purple, left half) in male-centric coordinates (male at origin), averaged across recordings. mfAngle is the angle of the female thorax relative to the body axis of the male. Complex bouts are more likely to be produced when females are close and in front of the male. f,g, Average mfDist (f) and mfAngle (g) during simple and complex song bouts. h, Average duration of simple and complex song bouts. For c–h, n = 20 wild-type males (biological replicates) courting wild-type females (see Supplementary Table 2 for genotypes). For d, f–h, central mark indicates the median; the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively.Whiskers extend to 1.5 times the interquartile range away from the box edges. For f–h, Wilcoxon rank-sum test for equal medians. *P < 0.05, **P < 0.01, ***P < 0.001, NS, not significant. Source data

Fig. 2. Reciprocal interactions between pulse-producing and sine-producing neurons in the presence of a female.

a, Broad-range optogenetic stimulation of song neurons pIP10 and TN1 (see Methods). One block is 15 trials for 8 s each. Neuron schematic in a was adapted from ref. , Elsevier, and ref. , Elsevier, under a Creative Commons licence CC BY 4.0. b, Song production per trial and time-resolved song probabilities across trials following optogenetic activation of pIP10 neurons in a solitary male. Responses are shown for 3 out of 20 randomized stimulus blocks. Pulse and sine probability for the third example stimulus block, averaged across n = 20 recordings (bottom). Rebound sine is the production of sine song immediately following pulse song production. Opto stim, optogenetic stimulation. c–e,h–j, Average song probabilities (b) for all stimulus blocks (distinct stimuli per pair of rows); activation of pIP10 (c–e) or TN1 (h–j) in solitary males (c,h), males paired with a wild-type female (d,i) and decapitated solitary males (e,j). f, Rebound sine probability following activation of pIP10 neurons (highest irradiance level only) in solitary, female-paired or headless males. Female presence promotes complex bout (pulse followed by rebound sine) generation following pIP10 activation. g,l, Average duration of song bouts generated via activation of pIP10 (g) or TN1 (l) neurons in solitary, female-paired or headless males. Female presence promotes longer song bouts following activation of either neuron type. k, Rebound pulse probability following activation of TN1 neurons (highest irradiance level only) in solitary, female-paired or headless solitary males. Female presence promotes complex bout generation (sine followed by rebound pulse) following TN1 activation. m, Simplified circuit model of song pathway; female cues ‘unlock’ complex bout generation via modulation of post-inhibitory (post-inh) rebound excitability (exc) in pulse-driving and sine-driving neurons of the ventral nerve cord (VNC). Disinh., disinhibition. Pir, post-inhibitory rebound. For c–g, n = 20 (solitary), 20 (with female) and 10 (headless) males (biological replicates). For h–l, n = 23 (solitary), 28 (with female) and 10 (headless) males (biological replicates). For f,g,k,l, Wilcoxon rank-sum test for equal medians. *P < 0.05, **P < 0.01 and ***P < 0.001. Central mark indicates the median; the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively.Whiskers extend to 1.5 times the interquartile range away from the box edges. Source data

Fig. 3. Investigation of rebound dynamics among Dsx⁺ TN1 neurons of the VNC.

a, Two-photon calcium imaging from VNC Dsx⁺ TN1 neurons combined with optogenetic activation of pIP10 descending neurons (see Methods for details; see Supplementary Table 2 for genotypes). The numbers of pulse-related and sine-related subpopulations are according to refs. ^,. Schematic in a was created using BioRender (https://biorender.com). b, TN1 neurons show diverse calcium response dynamics following pIP10 activation (123 TN1 neurons across 3 biological replicate flies). Each row shows the normalized calcium response (dF/F) of a single soma, averaged over seven trials (and each trial contained four stimulus presentations), and responses are sorted by their correlation (corr) with the optogenetic stimulus. The optogenetic stimulus pattern was chosen to produce pulse song followed by rebound sine (see Fig. 2c) in solitary males. c, Pairwise correlation of trial-averaged activity between any two TN1 neurons from one hemisphere in one male (to avoid any cross-hemisphere effects due to, for example, wing choice), following pIP10 activation. Examples of strong anti-correlation and correlation are shown on the right (I/II; P < 1 × 10⁻⁵⁰), where light-red boxes indicate stimulus intervals. d, Time-averaged fluorescence of the calcium indicator GCaMP6s expressed in Dsx⁺ TN1 neurons. The red and blue regions of interest correspond to anti-correlated or correlated pairs shown in c. a.u., arbitrary units. Schematic in d was adapted from ref. , Elsevier. e, Distribution of calcium response correlation coefficients (computed per hemisphere and per male) across TN1 recordings in n = 3 males. Colours are as in c. Owing to the near-perfect anti-correlation observed for some neuron pairs, we assume that the intermediary neurons of the rebound circuit (Fig. 2m) are mutually inhibitory, in addition to providing inhibition to song-generating neurons. For c,e, ‘i’ and ‘j’ denote indices to pairs of recorded TN1 neurons. Source data

Fig. 4. Acute female sensory cues promote complex song bout generation.

a,b,d, Pulse and sine song probabilities following optogenetic activation of P1a (a), pC2 (b), or both pIP10 and P1a neurons (d), in solitary male flies (n = 17, 16 and 16 biological replicates; genotypes are available in Supplementary Table 2). Schematic in a was adapted from ref. , eLife Sciences, under a Creative Commons licence CC CY 4.0. Schematic in b was adapted from ref. , Elsevier. c, Peak song probability per optogenetic stimulus duration for pC2 neurons (25 μW mm⁻²). e,f, Peak rebound sine probability (e) and average bout duration (f) for optogenetic activation (25 and 205 μW mm⁻²) of pC2, pIP10 or P1a neurons in solitary males or males paired with a wild-type female (data shown in a,b,d; Fig. 2c,d). g–i, Song amount (g), proportion of simple pulse bouts (h) and song complexity (mean number of pulse–sine or sine–pulse alternations) (i) in pC2 > TNT males paired with wild-type females. j, Automated tap detection (green; see Methods) and mfDist (black) with a 4-mm threshold for far or near context (grey horizontal dashed line) from an example recording (left). Male locations during tap and no tap events, in female (f)-centric coordinates (recording is the same as on the left) (right). k, Examples of simple (top) and complex (bottom) pulse bouts along with detected taps (green). l, Average tap rate (see Methods) before and during simple and complex pulse-leading bouts, driven by activation of pIP10 in males paired with a wild-type female (Fig. 2d; n = 18 biological replicates). m, To fit generalized linear models (GLMs) predicting pulse bout type (simple versus complex), we used movement features or P1 rate (shades of cyan; see Methods) over the 5 s preceding the end of the first pulse train. fmAngle, female–male angle; mfAngle, male–female angle. n, GLM relative deviance reduction for features predicting bout type (m). Input features are ranked by their predictive power (n = 51 model fits on random subsets of data from n = 18 biological replicates; see Methods). fFV, female forward velocity; fLS, female lateral speed; fRS, female rotational speed; mFV, male forward velocity; mLS, male lateral speed; mRS, male rotational speed. o, GLM filters for the four most predictive features in n. p, Updated model of the brain circuitry involved in male song sequencing. In e,f,h,i,l,n, Wilcoxon rank-sum test for equal medians. *P < 0.05, **P < 0.01 and ***P < 0.001. For g, **P < 0.01, two-sample Kolmogorov–Smirnoff test for equal distributions. For g–i, n = 21 and n = 17 biological replicates for experimental and control groups. For g,k,m, red and blue indicate pulse and sine song, respectively. For e–i,l,n, central mark indicates the median; the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. Whiskers extend to 1.5 times the interquartile range away from the box edges. Source data

Fig. 5. Neural circuit model of context-dependent song patterning.

a, Circuit model for male song patterning far from and near a female. mfDist (top), the only input to the model, enters the circuit via the pC2 node (see Methods), which drives the pulse pathway. Strong input (near the female) additionally disinhibits the VNC rebound circuit, enabling complex song production (alternating activity of the pulse and sine nodes). Grey indicates nodes becoming inactive at far or near conditions. Here, bout termination mainly relies on increases in mfDist (Extended Data Fig. 6a,b), consistent with ref. . b, Spiking neuronal network of four nodes (pC2, inh, p and s) representing the key computational features of the circuit in a, disinhibition, rebound excitability and mutual inhibition, fit to wild-type courtship data (see Methods). Model simulations with brief and weak (top) or long and strong (bottom) input to pC2 (corresponding to mfDist = 4.2 and 1.5 mm) result in either simple (‘p’) or complex (‘psp...’) song outputs. c, Song statistics for genetic algorithm fits of the model in b to song data at far (top) or near (bottom) distance (see Methods; experimental distributions shown in Extended Data Fig. 7k). The model reproduces bout statistics of courting wild-type flies (see Fig. 1d). d,e, Average mfDist (d) or population-averaged probability (mean ± mean absolute deviation from the mean) at a given mfDist (e) of simulated simple pulse, simple sine or complex bouts (models as in c) matches observations in courting wild-type flies (see Fig. 1c,f). Vertical grey line in e separates near and far contexts. f, Fit error (genetic algorithm objective function) for the full model versus models with individual computational features knocked out (see Methods), or disinhibition replaced with an excitatory motif (‘exc modulation’; see Methods; Extended Data Fig. 7a–c). For c–f, n = 24 (c–e) and n = 93 (f) genetic algorithm model fits to song (400 and 200 s each for c–f) randomly chosen from n = 20 wild-type recordings (biological replicates). For d,e, Wilcoxon rank-sum test for equal medians. *P < 0.05, **P < 0.01 and ***P < 0.001. For c,d,f, central mark indicates the median; the bottom and top edges of the box indicate the 25th and 75th percentiles, respectively. Whiskers extend to 1.5 times the interquartile range away from the box edges. Source data

Extended Data Fig. 1. Context-dependence of song sequencing in Drosophila melanogaster males (supplement to Fig. 1).

a, Male behaviors during courtship (modified from, including aspects from^,), with those focused in the present study highlighted in the grey box. b, Cumulative fraction of simple pulse (red), simple sine (blue), or complex (purple) bouts over time in recording for n = 20 wild-type male-female pairs (biological replicates). c, Distribution of pulse train duration, sine train duration, and the number of pulse-sine alternations in male song of n = 20 wild-type male-female pairs (biological replicates). d Chamber for behavioral experiments. Male courtship song was recorded using 16 microphones (green) tiling the chamber floor. Female (magenta) and male (blue) fly pose and tracks were estimated using SLEAP. e Cumulative fraction of far (brown) and near (yellow) bouts over time in recording for n = 20 wild-type male-female pairs (biological replicates). f, Population-averaged probability to sing simple pulse, simple sine, or complex bouts relative to male forward velocity (mFV). Color code as in (a). g, Distribution of mFV near (yellow) and far (brown) from the female. h, The majority (91%) of final bouts (the last song bout prior to copulation) occur within 0.6 seconds preceding copulation. i, The majority (59%) of bouts immediately preceding copulation are complex (n = 23 wild-type pairs with copulation within a 20 minute recording). Song bouts are aligned to bout end. Time-resolved probability of pulse (red) and sine song (blue) (shown below song traces) rises prior to copulation. Black curve at the bottom shows the fraction of males that sing both pulse and sine song in the time prior to copulation. 80% of males sang both song modes within the final 1.5 seconds of song before copulation, suggesting complex bouts facilitate mating. b,e, mean ± mean absolute deviation from the mean. b,c,e,f,g, n = 20 recordings of male-female pairs (biological replicates). Source data

Extended Data Fig. 2. Reciprocal interactions between pulse- and sine-producing neurons (supplement to Fig. 2).

a, Example raw song responses drawn from n = 20 solitary pIP10 > CsChrimson males (biological replicates) with a single type of optogenetic stimulus (205uW/mm2 on for 2s per 8s trial). For every recording, five out of 15 trials were randomly chosen for display. Numbers on y-axis indicate recording. Color code: red - pulse song, blue - sine song, grey - silence, pink - optogenetic stimulus. b, Example raw song responses drawn from n = 23 solitary TN1 > CsChrimson males (biological replicates) to the same stimulus type shown in a). c, Example raw song responses frawn from n = 20 pIP10 > CsChrimson males (biological replicates), paired with a wild-type female, to the same stimulus type shown in a). d, Example raw song responses drawn from n = 28 TN1 > CsChrimson males (biological replicates), paired with a wild-type female, to the same stimulus type shown in a). e, Population-averaged song responses of n = 18 pIP10 > CsChrimson males (biological replicates) paired with a wild-type female as shown in Fig. 2d, but split into instances during which male and female were far or near (as quantified in Fig. 4e,f). f, Population-averaged song responses of n = 15 TN1 > CsChrimson males (biological replicates) paired with a wild-type female as shown in Fig. 2i, but split into instances during which male and female were far or near (as quantified in Fig. 4e,f). g, Population-averaged song responses of n = 10 pIP10 > CsChrimson males (biological replicates) paired with a wild-type male. h, Population-averaged song responses of n = 12 TN1 > CsChrimson males (biological replicates) paired with a wild-type male. i, Example raw song responses drawn from n = 9 solitary headless pIP10 > CsChrimson males (biological replicates) to the same stimulus type shown in a). j, Example raw song responses drawn from n = 10 solitary headless TN1 > CsChrimson males (biological replicates) to the same stimulus type shown in a). Source data

Extended Data Fig. 3. Post-inhibitory rebound dynamics in the VNC (supplement to Fig. 3).

a, Anti-correlation between calcium responses of TN1 neuron pairs persists across trials. While Fig. 3c shows the correlation between trial-averaged calcium responses of TN1 neuron pairs in one fly, here we show the correlation between TN1 pairs for individual trials (7 trails, each trial consisting of four optogenetic stimulus presentations and a pause, as shown in Fig. 3b), only for pairs with trial-averaged anticorrelation coefficient below −0.8 (n = 17). b, Standard deviation (SD) across trials of the correlation coefficients shown in a). The majority of anti-correlated TN1 pairs are consistent across trials. c, Song of a mutant male systematically lacking I_h, courting a wild-type female. d, Overall sine probability (fraction of time spent singing sine song in a 30-minute recording) for two different strains of I_h mutants (mutant A, $I_h^{03055}$, and mutant B, $I_h^{01485}$; see Supplementary Table 2) and wild-type males. e, Proportion of simple pulse bouts in song of I_h mutants and wild-type males, produced far from (>4mm) a wild-type female. f, Proportion of simple pulse bouts in song of I_h mutants and wild-type males, produced near (<4mm) a wild-type female. g, Song of males with TN1-specific downregulation of I_h or Rdl (GABA-A receptors). h, Mean number of pulse-sine alternations in song of males with TN1-specific downregulation of I_h or Rdl (GABA-A receptors), and genetic controls (see Supplementary Table 2), produced near (<4mm) a wild-type female. i, Proportion of complex bouts with leading pulse mode, in song of males with TN1-specific downregulation of I_h or Rdl (GABA-A receptors), and genetic controls (see Supplementary Table 2), produced near (<4mm) a wild-type female. h-i, n = 17 for TN1 > I_h, n = 20 for TN1 > Rdl, n = 15 for genetic controls (all biological replicates). The effect of I_h reduction was modest, possibly because neurons other than TN1 contribute to sine song production, because rebound excitability in TN1 neurons arises from a degenerate set of ion channels that are robust to small perturbations (via knockdown) of I_h, or because a reduction of I_h channels maintains some rebound excitability through increased channel conductance at stronger hyperpolarization. d-f, n = 7 for mutant A, $I_h^{03055}$, n = 9 for mutant B, $I_h^{01485}$, and n = 20 for wild-type males (biological replicates). d-f,h,i Wilcoxon rank-sum test for equal medians; *P < 0.05, **P < 0.01, ***P < 0.001; NS, not significant. Source data

Extended Data Fig. 4. Neurons targeted in genetic driver lines (supplement to Fig. 4).

a, Male brain expressing CsChrimson.mVenus via GMR42B01 ∩ Dsx (green). Neuropil is labeled with nc82 (magenta). The intersection labels pC2 neurons (circled in blue), as well as 6 pCd-like neurons (circled in yellow). These pCd-like neurons do not express Fru (see panel c), and therefore constitute a different subset of neurons than the pCd neurons contributing to persistent male arousal downstream of P1a neurons in. Broad-range optogenetic activation in males using the genetic driver for pCd neurons from produces no song (data not shown). b, Zoom of the boxed area in a), showing that the pC2 population labeled in the intersection consists of both pC2l neurons (solid circle) and pC2m neurons (dashed circle). c, Male brain expressing CsChrimson.mVenus via GMR42B01 ∩ Fru (green). Neuropil is labeled with nc82 (magenta). d, Male brain and VNC expressing CsChrimson.mVenus (green) via both P1a and pIP10 drivers. Neuropil is labeled with nc82 (magenta). Cell bodies of P1a are circled in red. Cell bodies of pIP10 are circled in cyan (see Supplementary Table 2 for full genotypes). Source data

Extended Data Fig. 5. Sensory feedback, disinhibition, and P1a neuron priming in the generation of complex song bouts (supplement to Fig. 4).

a, Example raw song responses drawn from n = 17 solitary P1a > CsChrimson males (biological replicates) to a single type of optogenetic stimulus (205uW/mm2 on for 2s per 8s trial). For every recording, five out of 15 trials were randomly chosen for display. Numbers on y-axis indicate recording. Color code: red - pulse song, blue - sine song, grey - silence, pink - stimulus. b, Z-scored maximal wing angle (top) and probability to sing (bottom) of solitary males around optogenetic activation of P1a for three different stimuli (25 and 205 uW/mm2 for 250 ms and 2s, respectively, during 8s trials in n = 17 biological replicates, and 205 uW/mm2 for 10s during 100s trials in n = 20 biological replicates). Each line in the top row corresponds to the mean across trials. c, Example raw song responses drawn from n = 16 solitary pC2 > CsChrimson males (biological replicates) to the same stimulus type shown in a). d, Peak probability of two types of pulse song termed Pfast and Pslow (orange and red) and sine song (blue) as a function of stimulus duration for intermediate-irradiance activation (25uW/mm2) of pC2 or pIP10 in solitary males, or pIP10 in males far or near from a wild-type female (n = 16/20/20 biological replicates). e, Example raw song responses drawn from n = 16 solitary P1a-pIP10 > CsChrimson males (biological replicates) to the same stimulus type shown in a). f, Left: P1 neurons constitute a male-specific subset of pC1 neurons^,. Top right: disinhibitory circuit motif (an inhibitory ‘F1’ follower neuron inhibiting another ‘F2’ follower neuron) postsynaptic to an excitatory (cholinergic) neuron of the pC1a subset, identified in public female connectome data, using FlyWire^,. Bottom right: Number of GABAergic disinhibitory motifs postsynaptic to neurons of the pC1 subtypes a-d, detected in the female connectome. g, Output neuropils of F2 follower neurons for all disinhibitory motifs in (f), sorted by the number of output synapses. The majority of output synapses target the anterior ventrolateral protocerebrum (AVLP), the anterior optic tubercle (AOTU), and the posterior ventrolateral protocerebrum (PVLP). h, Two-photon calcium imaging from GABAergic (Gad1+) brain neurons combined with optogenetic activation of P1a brain neurons (see Supplementary Methods for details; see Supplementary Table 2 for genotypes). Schematic in h was created using BioRender (https://biorender.com). i Example Gad1 calcium responses for two regions of interest (ROIs) showing activity locked to stimulation (‘opto stim’) of P1a (‘F1 ROI’) or suppressed activity during F1 activity (‘F2 ROI’), as expected for neurons forming a disinhibitory motif postsynaptic to P1a (schematic at top). j, Anatomical distribution along the dorsal-ventral (D-V) axis of (n = 262) F1 and (n = 75) F2 follower ROIs (see i) recorded in two hemispheres, but collapsed to the left/right hemisphere respectively for visualization. k, Anatomical distribution of the F1 and F2 follower ROIs shown in j, across a sagittal slice of the brain. l, Tap-detector model performance. (Top) Example of non-tap (left) and tap (right) events. Green arrows indicate the position of male foreleg tarsi. (Bottom) Receiver operator characteristic (ROC) curve for model after 100 epochs of training (orange points). Each point corresponds to a different tap probability threshold. Area under the ROC curve (AUC) is used as an evaluation metric - an ideal model would have an AUC of 1. Performance of a null model (gray diagonal line) is included for comparison. m, Average tap rate before and during simple and complex pulse bouts, for n = 20 wild-type male-female pairs (biological replicates; analog to Fig. 4l). n, A generalized linear model (GLM) to predict complex vs. simple pulse bout production based on the history of sensory features prior to the end of the first pulse train in each (ps... complex or p simple) bout in n = 51 random samples from n = 20 biological replicate recordings of wild-type male-female pairs (analog to Fig. 4n,o). Sensory features are ranked by their predictive power, and GLM filters are shown for the four most predictive features. o, To test for effects of persistent male arousal on optogenetically driven song, males were primed (allowed to court a virgin wild-type female) for 5 minutes preceding optogenetic activation. p, Song probabilities for optogenetic activation of pIP10 neurons in solitary males that were primed. n = 19 biological replicates. q, Comparison of peak rebound sine probability for optogenetic activation at intermediate and strong irradiance (25 and 205uW/mm2) of pIP10 in primed, solitary, female-paired, or P1a-coactivated males. r, Comparison of peak pulse probability for optogenetic activation at lowest irradiance (1uW/mm2) of pIP10 in groups identical to those in (q). s, Example raw song responses drawn from n = 20 solitary pIP10 > CsChrimson males (biological replicates) to the same stimulus type shown in a). Males were primed (allowed to court a virgin wild-type female, to induce male courtship state) for five minutes prior to the start of the optogenetic stimulus protocol. t, Population-averaged song responses of n = 20 primed solitary TN1 > CsChrimson males (biological replicates). u, Raw song responses of n = 20 primed solitary TN1 > CsChrimson males (biological replicates) to the same stimulus type shown in a). j,k, n = 4 biological replicate animals. p,r, Simple pulse song was induced in a fraction of primed males even for the weakest levels of activation, in contrast to males subject to identical stimulation without priming, suggesting that male arousal modulates the excitability of pIP10 neurons at the timescale of minutes but without promoting complex song (compare Fig. 2c). q,r, n = 19/20/20/16 biological replicates for activation of pIP10 in primed, solitary, female-paired, or P1a-coactivated males. m,q,r, Wilcoxon rank-sum test for equal medians; *P < 0.05, **P < 0.01, ***P < 0.001, 1; NS, not significant. Source data

Extended Data Fig. 6. Neural activity dynamics driving simple and complex bouts in the song circuit model (supplement to Fig. 5).

a, (top) Z-scored male-female distance (mfDist) from wild-type courtship data (which served as input to the model) triggered around the time of simple (p,s) or complex (ps...,sp...) bout start in simulations of the song circuit model. Each line is the z-scored mfDist averaged across bouts for one simulation (lines were smoothed for visualization, using a uniform filter of 44.4 ms length). Every simulation uses song randomly chosen from all wild-type recordings (such that the chosen song contained a minimum of 10% of bouts at mfDist ≥ 4 mm, and the fit error / objective function value was below 0.1). For all bout types, mfDist decreases around the time of bout start. (Bottom) Instantaneous spike rate of the ‘pC2’ (green) and ‘inh’ (black) nodes of the circuit model around the time of bout onset. Distinctly timed release from inh-mediated inhibition in combination with distinct levels of pC2-mediated excitation drives different bout types. b, Dynamics of z-scored mfDist (top) and instantaneous spike rate of the pC2 and inh nodes in the circuit model at the time of bout termination, for complex bouts ending in pulse (left) or sine mode (right). In both cases, bout termination is accompanied by increases in mfDist and a resulting reduction in pC2-mediated excitation of the pulse and sine node. a-b, n = 24 model fits to song (400 seconds each) randomly chosen from n = 20 wild-type recordings (biological replicates). Source data

Extended Data Fig. 7. Testing and expanding the neural circuit model of context-dependent song patterning (supplement to Fig. 5).

a, Song circuit model with default disinhibitory modulation of the pulse/sine rebound circuit (left) and quasi-equivalent excitatory modulation (right), with simulated responses of the respective four nodes to a ‘near’ input. b, ‘Song’ statistics of the model with excitatory modulation (a, compare Fig. 5f). In contrast to the default model with disinhibition, the excitatory model exclusively produces pulse bouts (simple and complex). c, Population-averaged probability of simulated simple pulse (red), simple sine (blue), or complex (purple) bouts at a given male-female distance (mfDist) in the model with excitatory modulation (a) matches the relationship between distance and song types observed in courting wild-type flies for simple pulse and complex bouts, but not for simple sine bouts (compare with Fig. 1c). d, Triggering pIP10 activation on sine song in males courting a wild-type female strongly increases bout duration and complexity compared to controls with yoked activation (that is, identical stimulus statistics as in the closed loop condition but uncorrelated to the control male’s song). Song shown from an example recording. e, Song and stimulus probability around the onset of male sine song, for closed loop (top) and yoked (bottom) activation of pIP10 during the recording shown in a). f, Population level comparison of four song features between closed-loop (CL) and yoked (OL) activation of pIP10 (n = 9 biological replicates): the fraction of trains belonging to complex bouts, the median number of sine-pulse or pulse-sine alternations in complex bouts, the median duration of complex bouts, and the median sine train duration within complex bouts. To show that all effects extend beyond generation of a single rebound sine, only ‘psp...’ bouts were considered for these analyses. Wilcoxon rank-sum test for equal medians; *P < 0.05; **P < 0.01; ***P < 0.001. g, Amount of simple pulse song bouts produced during courtship of a female, in recordings of males with tonically hyperpolarized pIP10 neurons (top) and males with blocked chemical synapses in pIP10 neurons (bottom; via expression of inward-rectifying potassium channels in pIP10 neurons, VT040556 > kir, and via expression of tetanus toxin light chain / TNT in pIP10 neurons, VT040556 > TNT; filled box plots), compared to two genetic controls (blank box plots). See Supplementary Table 2 for genotypes. Under both manipulations, the amount of simple pulse bouts was strongly reduced in male song. h, Amount of simple and complex sine song bouts produced near a female, in recordings of males with tonically hyperpolarized pIP10 neurons (top) and males with blocked chemical synapses in pIP10 neurons (bottom; same manipulations as in g), compared to two genetic controls (blank box plots). The amount of bouts with leading sine was increased in males with blocked chemical synapses in pIP10, but unaffected in males with tonically hyperpolarized pIP10 compared to controls. i Z-scored male forward velocity (mFV) around the start of simple pulse (p), simple sine (s), or complex bouts with leading pulse (ps..) or sine (sp..), for solitary males with optogenetic activation of P1a and intact vision (n = 17 biological replicates; same as Fig. 4a), or blind males with simultaneous activation of P1a and pIP10 (n = 16 biological replicates; same as Fig. 4d). Only song bouts outside the stimulus interval (persistent song) are included here. At onset, bouts with leading pulse or sine show increases and decreases in mFV. j, Circuit model to explain the findings in d,e: chemical synapses from pIP10 onto the VNC pulse node explain the reduction in simple pulse bouts with kir and TNT expression in pIP10. Gap junctions (electrical synapses) between pIP10 and the inhibitory interneuron node of the pulse pathway facilitate simple and complex sine bouts with blocked chemical synapses in pIP10, by transforming pIP10 activity through the electric synapses into inhibition onto sine driving neurons, leading to rebound sine bouts after termination of pIP10 activity. k, Song bout statistics at far and near distances, for n = 24 200 second segments randomly drawn from wild-type courtship data (n = 20 biological replicates) that were used to fit the model shown in Fig. 5a–c. b,c, n = 93 genetic algorithm fits to experimental song data randomly chosen from n = 20 biological replicates. f-h, Wilcoxon rank-sum test for equal medians; *P < 0.05, **P < 0.01, ***P < 0.001. g,h, n = 15/13/13 biological replicates for VT040556 > kir and the two genetic controls, n = 16/18/16 biological replicates for VT040556 > TNT and the two genetic controls. Source data

Extended Data Fig. 8. Spike-frequency adaptation in pIP10 neurons (supplement to Fig. 5).

a, In vivo patch-clamp electrophysiology of descending neuron pIP10. Action potentials (spikes) were observed both during injection of positive current and following injection of negative current (post-inhibitory rebound spikes). b, For each current stimulus amplitude, instantaneous spike rates were defined as the inverse of each inter-spike interval for all successive pairs of spikes observed within the trial. Shown is the median (per stimulus) instantaneous spike rate following negative or during positive current injection. c, Instantaneous spike rate of the first ten spike pairs in each trial, normalized by the spike rate of the first pair. Normalized spike rate decreases for successive spikes, indicative of spike-frequency adaptation. Source data