Recurrent Circuitry for Balancing Sleep Need and Sleep

1The behavioral hallmarks of sleep are manifold. They include inactivity, reduced responsiveness to external stimuli, rapid reversibility, and homeostatic rebound after sleep loss. Any sleep control system must therefore fulfill a multitude of functions—blocking locomotor activity, gating sensory pathways, inhibiting arousal systems, relieving sleep pressure—and perhaps also directly influence processes germane to a fundamental purpose of sleep, be it metabolic recovery, memory consolidation, or synaptic scaling.

2Surprisingly, given these diverse and widespread manifestations, activity in a tiny minority of two dozen neurons (of a total of ∼100,000 in the brain) suffices to induce sleep in Drosophila. The sleep-promoting neurons send projections to the dorsal fan-shaped body (dFB) of the central complex and act as a feedback controller or homeostat. Their operating principle is remarkably simple: sleep need is encoded in the intrinsic electrical excitability of the sleep-inducing cells, which fluctuates because two potassium conductances, voltage-gated Shaker and the leak channel Sandman, are modulated antagonistically. As sleep pressure builds during waking, the sleep-promoting neurons switch from electrical silence to activity and the animal from wakefulness to restorative sleep. The self-correcting nature of feedback is thus embodied in the biophysics of an excitability switch.

3dFB neurons can be arrested in the electrically silent state by mutating the Rho-GTPase-activating protein Crossveinless-c (Cv-c). The mutation likely prevents the internalization of Sandman that is a prerequisite for flipping the neurons’ sleep-promoting activity back on. cv-c mutants suffer profound insomnia (along with its cognitive consequences) and are unable to sense and/or correct sleep deficits. In contrast to our growing understanding of the sleep-control neurons themselves, however, neither the signals released by them to induce sleep, nor any of their downstream targets, nor the manner in which they regulate these targets have been identified.

4Among the many sleep-regulatory structures in mammals (for reviews, see,, ), a cluster of sleep-active neurons in the ventrolateral preoptic nucleus (VLPO) of the hypothalamus exhibit perhaps the clearest parallels with dFB neurons in flies. VLPO activation is tightly correlated with sleep, and VLPO lesions fracture the sleep-wake cycle, producing insomnia. Like dFB neurons, VLPO neurons modulate their firing rates according to sleep need, with activity peaking at the beginning of recovery sleep. VLPO neurons secrete the inhibitory neuropeptide galanin along with the classical inhibitory transmitter GABA and project to the tuberomamillary nucleus and other arousal centers in the brain stem, which often form reciprocal inhibitory connections with the VLPO. Mutual antagonism between neurons promoting sleep and waking thus creates a bistable flip-flop arrangement. Projections from VLPO neurons to structures other than arousal centers have not been described, leaving open the question of whether sleep-promoting cells can directly control motor or sensory pathways or whether they do so only indirectly by inhibiting arousal systems.

5Here we begin to explore the circuitry downstream of sleep-control neurons in Drosophila. We find that dFB neurons induce sleep via a range of inhibitory transmitters that include the neuropeptide allatostatin-A (AstA). Among the targets of AstA are a group of interneurons of the central complex that we term helicon cells. These neurons are inhibited by sleep-promoting AstA, excited by visual input, permissive for locomotion, and presynaptic to R2 ring neurons of the ellipsoid body, whose activity has been linked to the accumulation of sleep debt. dFB-mediated inhibition of helicon cells may thus account for three cardinal features of sleep: elevated visual thresholds, immobility, and the dissipation of sleep need.

6Sleep-promoting neurons marked by R23E10-GAL4 project their axons to a single dorsal stratum of the fan-shaped body, where they form numerous synaptic release sites revealed by decoration with GFPDSyd-1 (; Figure 1A). Of the at least eight neuropeptides detected in different layers or layer combinations of the fan-shaped body, the distribution of AstA partially overlaps the axons of R23E10-GAL4-positive neurons (Figure 1B), hinting that sleep-promoting neurons may be a source of AstA. To corroborate this notion, we examined AstA immunoreactivity and sleep in carriers of AstAMB10261, a transposon insertion in the 3′ UTR of the AstA locus that disrupts the AstA-RA isoform (Figure S1A). Homozygous carriers of AstAMB10261 lacked detectable AstA staining in the dFB (but retained some AstA immunoreactivity elsewhere in the brain; Figures S1B and S1C) and slept ∼25% less than heterozygous controls (Figures 1C and S1D) because of shorter sleep episodes during the day and night (Figure 1D). This sleep maintenance insomnia is reminiscent of that of cv-c mutants, in whom uncontrolled cell surface expression of Sandman is thought to short-circuit the spike generator of dFB neurons. Like cv-c mutants, AstA mutants exhibited robust free-running circadian rhythms after entrainment (Figure S1E) but failed to compensate homeostatically for a night of mechanical sleep deprivation (Figure 1E). Similar mutant phenotypes are, of course, expected if some of the action potential output of dFB neurons is conveyed by AstA.

7Despite these similarities, the sleep disruptions of AstAMB10261 homozygotes were milder than those of cv-c transheterozygotes, leaving scope for GABAergic or peptidergic cotransmission, as in VLPO neurons, or transmitter heterogeneity within the dFB neuron population. Indeed, the vesicular GABA transporter (vGAT) and several neuropeptides in addition to AstA were translated by R23E10-GAL4-positive neurons (Figure S1G), and spatially restricted RNA-mediated interference (RNAi) with the expression of some of these peptides (e.g., myoinhibiting peptide) implicated them, too, in the regulation of sleep (Figure S1H). While AstA is thus unlikely the only sleep-promoting signal released by dFB neurons, it remains our sole focus here.

8To tie the role of AstA in homeostatic sleep control to dFB neurons, we altered AstA levels selectively in these cells. RNAi knockdown, using either the exquisitely dFB-specific R23E10-GAL4 driver or the somewhat broader 104y-GAL4 line, reduced basal sleep relative to parental controls (Figures 1F, S1D, S1F, and S1I) and eliminated the homeostatic response to sleep deprivation (Figure 1E). dFB-restricted overexpression of a transgene encoding AstA, again with the help of the R23E10-GAL4 and 104y-GAL4 drivers, had the opposite effect; it elevated sleep time (Figure 1G; Figure S1J).

9Knowledge of a neuropeptide secreted by dFB neurons allowed us to search for postsynaptic targets among neurons expressing AstA receptors (AstA-Rs). To pinpoint the relevant receptor type(s), we measured sleep in flies carrying mutant AstA-R alleles. Flies homozygous for AstA-R1MB07922, a transposon insertion in the AstA-R1 locus, exhibited a short-sleeping phenotype that mirrored that of flies lacking AstA, suggesting a match between receptor and ligand: like homozygous AstAMB10261 mutants, homozygous AstA-R1MB07922 mutants lost ∼25% of their daily sleep compared to heterozygous controls (Figures 2A and S2A); as in AstA mutants, the loss in overall sleep time was caused by a shortening of sleep bouts during the day and night (Figure 2B) and accompanied by reduced rebound sleep after a night of enforced sleeplessness (Figure 2C).

10Two GAL4 lines incorporating enhancer modules of the AstA-R1 locus, R22H05-GAL4 and R22H10-GAL4, drive expression in a small number of neurons in the brain. These include a cluster of neuroendocrine cells in the pars intercerebralis and a handful of cells in the central complex (Figure 3A). Here, both R22H05-GAL4 and R22H10-GAL4 label four large interneurons that connect the superior arch to the ellipsoid body via fibers that pass near AstA-immunopositive puncta (Figure 3A) through the R23E10-LexA-positive layer of the dFB (Figure 3B). Because the spiral circular morphology of the individually labeled neurons (see below) resembles the brass instrument, we term these interneurons helicon cells. When expressed in helicon cells, the dendritic marker DenMark localized to the superior arch, while GFPDSyd-1 labeled presynaptic boutons in the bulb and the concentric rings of the ellipsoid body (Figure 3C). dFB neurons may thus gate the flow of signals from dendritic sites in the superior arch to axon terminals in the ellipsoid body.

11Reducing AstA-R1 levels within helicon cells via R22H05-GAL4 or R22H10-GAL4-driven expression of an RNAi transgene decreased the total amount of sleep (Figures 3D and S2B), with an especially pronounced effect on the length of sleep episodes during the afternoon siesta (Figures 3D and S2A–S2C). Although these are some of the expected consequences of rendering dFB targets insensitive to the sleep-promoting effect of AstA, the interpretation of this experiment is ambiguous because the expression domains of both GAL4 lines include neurons of the pars intercerebralis (Figure 3A), which themselves have been implicated in the regulation of sleep. To resolve this ambiguity, we used a third GAL4 driver, R24B11-GAL4, which captured the four helicon cells and a few currently unidentified cells in the dorsal brain but spared the neuroendocrine cells (Figures S3A and S3B). R24B11-GAL4-driven interference with the expression of AstA-R1 recapitulated the sleep phenotypes seen with the R22H05-GAL4 and R22H10-GAL4 lines (Figures 3E, 3F, and S2A–S2D). Because helicon cells are the only neuronal elements in common to all three expression patterns (Figures S3A and S3B), the observed sleep changes must reflect the loss of AstA-R1 from them.

12The consequences for sleep of depleting AstA-R1 from helicon cells differed subtly from those of removing AstA from dFB neurons and also from those of the genomic mutations: whereas RNAi-mediated interference with AstA-R1 expression in helicon cells caused the most profound and consistent sleep loss during the day (Figures 3D, 3E, S2A and S2B), homozygous carriage of the mutant AstA-R1MB07922 allele and diminished allatostatinergic transmission from dFB neurons produced sleep deficits also during the night (Figures 1, 2, S1, and S2). These differences could arise if helicon cells retained some AstA-R1 after knockdown or if dFB neurons controlled AstA-responsive targets in addition to helicon cells. In our view, the dissociation between the daytime and nighttime effects of the cell-specific receptor manipulation and the fact that only a fraction of dFB neuron terminals are found in immediate proximity to helicon cell neurites (Figures 3A and 3B) favor the latter interpretation.

13In whole-cell patch-clamp recordings from head-fixed flies walking or resting on an air-supported trackball, helicon cells (Figure 4A) were found in one of two states: a DOWN state characterized by the near absence of spikes (firing rate < 1 Hz) and an UP state in which the neurons fired persistently, with occasionally metronomic precision, at rates of 16.9 ± 3.6 Hz (Figures 4B and 4C). An average voltage difference of 10.9 ± 2.3 mV (mean ± SEM, n = 10 cells) separated the membrane potential baselines of the two states. Visual stimuli evoked large depolarizations, which, especially in the UP state, released intense flurries of action potentials (Figures 4C and 4D); in 68 of 94 cases (72.3%), these volleys of activity were associated with a locomotor bout (Figure 4E). Spontaneous movements were initiated with approximately 4-fold higher probability when the recorded cell was in the UP rather than in the DOWN state (Figure 4E). Together, these results suggest that helicon cells play a permissive role in visually guided movement.

14AstA-R1 encodes a G-protein-coupled receptor with homology to mammalian galanin receptors, suggesting that sleep-regulatory signals are conserved at the receptor level. Like galanin receptors, AstA-R1 controls the gating of G-protein-coupled potassium channels, whose opening in the presence of AstA is expected to inhibit AstA-R1-positive neurons. Indeed, focal pressure ejection of synthetic AstA from a pipette positioned near helicon cell neurites in the dFB hyperpolarized the cells, whereas the administration of a peptide containing the same amino acid residues in a randomly scrambled sequence, or of AstA to cells depleted of AstA-R1, elicited no response (Figures 4F and 4G).

15To verify that dFB neurons are the physiological source of inhibition, we expressed the ATP-gated cation channel P2X2 under R23E10-LexA control and pressure-ejected 500 μM ATP onto the dendrites of dFB neurons. The simultaneously recorded membrane potentials of helicon cells, which were marked by R22H10-GAL4-driven GFP expression, hyperpolarized as deeply in response to the genetically targeted activation of dFB neurons as they did to the direct delivery of AstA (Figures 4C and 4G). Helicon cells fell silent or fired only sparsely during periods of evoked dFB neuron activity; of four cells found in the UP state before the application of ATP, three switched to the hyperpolarized DOWN state afterward, and all neurons initially in the DOWN state remained (Figure 4C). Although visual stimuli continued to elicit large subthreshold depolarizations in the presence of ATP, spiking responses were attenuated or abolished (Figures 4C and 4D). dFB neurons thus mute the output of helicon cells by pulling their membrane potentials away from action potential threshold.

16In contrast to the profound inhibition of helicon cells, neuroendocrine cells in the pars intercerebralis showed no trace of modulation during artificially evoked dFB neuron activity (Figure 4G), reinforcing our conclusion that helicon cells are the sole dFB targets among the two groups of neurons marked by the R22H05-GAL4 and R22H10-GAL4 drivers.

17If dFB neurons promote rest by inhibiting helicon cells, then reducing the electrical activity of these neurons should, in itself, produce behavioral inactivity. Given the prominent responses of helicon cells to light (Figure 4C), the same manipulation is also expected to raise the threshold for visually evoked locomotor bouts (Figure 4E). To test these expectations, we used R24B11-GAL4 to place an inwardly rectifying potassium channel (Kir2.1) into helicon cells and timed the expression of the conductance with the help of the temperature-sensitive repressor of GAL4, GAL80ts. At the permissive temperature of 18°C, when functional GAL80ts prevented the expression of Kir2.1, locomotion and the percentage of inactive flies startled by 3 min of incubator light were indistinguishable in experimental animals and parental controls (Figures 5A and 5B). At the restrictive temperature of 31°C, when inactivation of GAL80ts allowed the transcription of Kir2.1 in experimental flies, basal and light-induced locomotor activity decreased relative to controls (Figures 5A and 5B).

18To examine whether the direct stimulation of helicon cells could override the inhibitory effect of dFB neurons and extend waking time, we targeted the light-driven actuator CsChrimson under R24B11-GAL4 control to helicon cells and measured locomotion under closed-loop conditions. The movements of individual flies were continuously monitored and used to trigger three pulses of 630-nm light (3 ms at 20 Hz) after > 3 min of inactivity; the pulse triplet was repeated every 30 s until the next movement occurred (Figure 5C). Control flies lacking the obligatory CsChrimson cofactor all-trans retinal, which in adult Drosophila must be supplied from external sources, showed normal levels of sleep despite receiving thousands of light pulses during the 24-hr analysis period (Figures 5D and 5E). The lack of an intrinsically arousing effect of red illumination reflects the minimal intensity of stimulation and the relative insensitivity of Drosophila’s photoreceptors at 630 nm. Rest in retinal-fed flies, in contrast, proved to be sensitive to optical disruption: experimental animals harboring functional CsChrimson in helicon cells maintained elevated levels of activity over a full 24-hr period while being exposed to lower light doses than controls (Figures 5D–5G). Individual behavior, color-encoded in 15-min time bins, illustrates at higher resolution the inverse relationship between locomotor activity and light exposure (Figures 5F and 5G): the more effective the optogenetically evoked helicon cell activity was in keeping an animal awake, the fewer optical stimuli that animal consumed.

19The axonal branches of helicon cells innervate the bulb and concentric rings of the ellipsoid body (Figure 3C), where they lie in close apposition to arborizations of R2 neurons (Figures 6A and 6B). Arrestingly, R2 neurons have been pinpointed as a principal source of sleep pressure. Prolonged periods of R2 neuron activity are thought to contract a sleep debt that is sensed and cleared by dFB neurons. Sleep homeostasis may thus involve an autoregulatory loop in which dFB and R2 neurons are recurrently connected via helicon cells. If helicon cells provide significant excitation to R2 neurons, their inhibition by dFB neurons—whose output will, in turn, reflect the activity history of R2 neurons—could throttle the excitatory drive to R2 neurons, allowing the system to reset during sleep.

20Despite their suggestive anatomical proximity in the ellipsoid body (Figures 6A and 6B), there is presently no functional evidence that helicon cells and R2 neurons are indeed connected and that these connections have the excitatory polarity needed to close the recurrent circuit we envisage. To probe for excitatory synapses from helicon cells to R2 neurons, we monitored the membrane potentials of R2 neurons (which were targeted on the basis of their R48H04-LexA-driven GFP expression; Figure S4) while photostimulating helicon cells (which expressed CsChrimson under the control of R78A01-GAL4, a strong driver recapitulating the R22H05-GAL4 pattern; Figure S3C). Under basal conditions, R2 neurons were showered by large excitatory postsynaptic potentials (EPSPs) that arrived at widely variable rates averaging 9.44 ± 1.41 Hz (mean ± SEM; range, 0.58–21.58 Hz; n = 5 cells) (Figure 6C). Illumination at 630 nm, sustained for 2 s, elevated the mean EPSP frequency by 62% to 15.33 ± 1.73 Hz (Figure 6D); during these barrages of optogenetically stimulated synaptic input, the membrane potential baseline depolarized by 5.30 ± 1.25 mV (mean ± SEM, n = 5 cells), beyond spike threshold (Figure 6C). Because light-evoked depolarizations vanished in the absence of retinal or the presence of 1 μM tetrodotoxin (Figures 6C and 6E), they were caused by action potential-driven transmission from presynaptic helicon cells. Even single pulses of light could elicit reliable spiking of R2 neurons (Figure 6F), attesting to the powerful influence helicon cells exert over the activity of these postsynaptic partners.

21Given the strength of the excitatory connections from helicon cells to R2 neurons, intense helicon cell activity is expected to filter through to R2 neurons and drive the plastic changes thought to represent accumulating sleep pressure, just as prolonged R2 neuron activation does. In our previous optogenetic stimulation experiments (Figure 5), we activated helicon cells minimally, and only when needed, to occlude the rest-promoting effect of dFB neurons without triggering homeostatic compensation. To test whether more forceful activation of the same neurons could induce a homeostatic response, we maintained helicon cells in an optogenetically induced UP state for 24 hr by delivering 3-ms pulses of stimulation light continually at 20 Hz (Figure 7). Because the necessary light exposures exceeded those for minimal stimulation (Figure 5E) by approximately three orders of magnitude, sleep disruptions of presumably visual origin were now also commonly seen in controls (Figure 7). These sleep disruptions obscured the wake-promoting effect of direct helicon cell stimulation (see Figures 5D and 5G for comparison) but were too mild to initiate rebound sleep on their own (Figures 7A and 7B). After the reconstitution of CsChrimson with all-trans retinal, however, which allowed helicon cells to be entrained to the 20-Hz optical stimulus, the sleep control circuitry tipped into rebound mode: experimental flies fell quiescent at the end of photostimulation (Figure 7A) and slept an excess of 7.64 hr relative to controls during the subsequent 24-hr day (Figure 7B).

22Sleep, an organismal phenomenon with many physiological and behavioral facets, is controlled by a handful of neurons with narrowly restricted axonal projections (Figure 1A), creating the apparent paradox of a local action with systemic consequences. A solution to this paradox could take several forms. One possibility is that the sleep-promoting cells do not themselves communicate with a wide range of postsynaptic targets but, rather, act indirectly by inhibiting arousal centers, which provide divergence through their own widespread projections. In other words, sleep would be induced by the widely felt withdrawal of a wake-promoting signal, not a diffusely broadcast command to go to sleep. Our data show that sleep-promoting neurons can directly suppress locomotor activity and blunt visual responses (Figure 4). Sleep-control neurons thus impose sleep via efferent circuits that include direct as well as indirect pathways acting through inhibition of arousal. Their combined effects on these circuits must cause the full spectrum of systemic changes associated with sleep.

23How many efferent circuits are there, and what are the functional relationships among them? Although the restricted expression of AstA-R1RNAi eliminates the electrophysiological response of helicon cells to AstA (Figures 4F and 4G), the same manipulation only incompletely phenocopies the sleep loss seen in homozygous AstA-R1MB07922 mutants (Figures 2A, 3D, 3E, and S2). These results demonstrate a sleep-promoting effect of inhibiting helicon cells, but they also suggest that helicon cells are only one of several dFB outputs used to induce sleep. The potential for specificity in synaptic communication between dFB neurons and their downstream partners raises the possibility that the different behavioral and physiological manifestations of sleep might be separable at the level of dedicated output circuits. In such an arrangement, different efferent channels would gate locomotion or set sensory thresholds, and selective interference with individual channels may dissociate sleep features that are normally grouped.

24In an alternative model, neurons of the central complex, and especially the ellipsoid body, may represent a site where a strategically placed gate can enact many sleep-related changes at once. Large amounts of sensory data from different modalities fan into the ellipsoid body, which uses these data to construct representations of visual space and the animal’s position and orientation within it. These representations then fan out to inform a range of actions, such as the ability to alternate between flight, walking, and climbing; to adjust the speed of locomotion in response to arousing stimuli; to negotiate turns, gaps, and obstacles; and to navigate to memorized locations. The arrangement thus resembles an informational bow tie, with a broad fan of incoming data flowing into a central knot and from there into another broad fan of outgoing motor instructions. In manufacturing, bow tie architectures are advantageous because they allow flexibility in the transformation of raw materials (sensory data) into products (actions) and because they operate economically and efficiently due to the small sizes of their processing cores. However, focused attack on these cores can cause the entire system to shut down. Could sleep-promoting neurons target this vulnerability?

25Among the postsynaptic partners of helicon cells are R2 ring neurons of the ellipsoid body (Figure 6), whose activity generates sleep pressure that is sensed by the dFB. The contours of an autoregulatory loop have thus emerged in which sleep-promoting dFB neurons communicate via helicon cells with R2 neurons, and the activity of these ring neurons is relayed back to dFB neurons (Figure 8). We imagine that, as sleep pressure builds during prolonged R2 neuron firing, activity-dependent plasticity augments the excitatory drive to dFB neurons or instructs them to step up their intrinsic excitability. As a result, dFB neurons switch to the electrically active state and release inhibition. This pushes helicon cells into the hyperpolarized DOWN state (Figure 4), mutes their spiking, and deprives R2 neurons of a powerful source of excitation (Figures 6 and 7). By virtue of this circular arrangement, dFB-derived inhibition can impose intermittent periods of rest on R2 neurons.

26The recurrent R2-dFB-helicon neuron circuit resembles a relaxation oscillator rather than a continuous feedback controller. Continuous feedback would ensure stable operation under a variable load, whereas a relaxation oscillator, such as an electric flasher circuit (Figure 8) or a water clock with a liquid-driven escapement, converts a continuous input signal into a binary output. To this end, the feedback loop contains a switching device that alternates between “fill” and “discharge” modes. A capacitor or reservoir is charged and emptied as its fill level rises to the voltage threshold of a bulb or the opening pressure of a valve (Figure 8). There are unmistakable parallels between these fill and discharge cycles and periods of accumulation and extinction of sleep debt and also between the voltage- or pressure-controlled relief paths of the engineered systems and the switching behavior of dFB neurons, which transition between a silent OFF state when sleep pressure is low (fill mode) and an active ON state when sleep pressure is high (discharge mode). Despite these parallels, many crucial questions remain. They include where precisely along the still unexplored R2-dFB neuron interface sleep debt accrues, in what physical form it is stored, how its accumulation to threshold actuates the dFB switch, and how the accumulated sleep debt is cleared.

27Further information and requests for resources and reagents should be directed to and will be fulfilled by the Lead Contact, Gero Miesenböck (gero.miesenboeck@cncb.ox.ac.uk).

28Drosophila melanogaster strains were grown on media of sucrose, yeast, molasses, and agar, and maintained on a 12 h light:12 h dark schedule at 25°C unless they expressed GAL80ts; in this case the experimental animals and all relevant controls were grown at 18°C. Flies expressing CsChrimson were transferred to food supplemented with 2 mM all-trans retinal in DMSO upon eclosion and reared in darkness thereafter. All studies were performed on male and/or female animals, as indicated below, aged 4–10 days at the beginning of the analysis period.

29The AstAMB10261 and AstA-R1MB07922 mutants carry insertions of MiMIC cassettes. Driver lines R23E10-GAL4, R23E10-LexA, and 104y-GAL4 were used to direct transgene expression to dFB neurons; R22H05-GAL4, R22H10-GAL4, R24B11-GAL4, and R78A01-GAL4 were used to target helicon cells; R58H05-GAL4 and R48H04-LexA provided access to R2 neurons of the ellipsoid body. Effector transgenes encoded membrane-bound fluorescent proteins (UAS-CD8::GFP; lexAop-mCherry); the dendritic and presynaptic markers DenMark and GFPDSyd-1, respectively; a GFP-tagged version of the ribosomal protein mL10a for the cell-specific analysis of polysome-bound transcripts; the ion channels Kir2.1 or P2X2; the optogenetic actuator CsChrimson; or hairpin constructs for RNA-mediated interference with the expression of AstA (transformant 113215KK), its receptor AstA-R1 (transformant 101395KK), and Mip (transformant 106076KK). The UAS-AstA transgene incorporates a codon-optimized synthetic cDNA sequence of 474 bp (Eurofins MWG Operon) in pUAST and was integrated into the genome at a random location (Rainbow Transgenics).

30Female flies were individually inserted into 65-mm glass tubes, loaded into Trikinetics Drosophila Activity Monitors, and housed under 12 h light:12 h dark schedules. Periods of inactivity lasting at least 5 min were classified as sleep. Mechanical sleep deprivation used the SNAP method for 12 h overnight. Sleep lost and regained was calculated for each fly by using the 24-h period preceding deprivation as the baseline. Visual arousal thresholds were estimated by exposing flies every 2 h to a 3-min pulse of incubator light and determining the percentage of sleeping flies awakened.

31Male flies were housed individually in 65-mm glass tubes containing 4% sucrose, 2% agar medium. Locomotor activity was measured in Trikinetics Drosophila Activity Monitors for 10 days in constant darkness. Rhythmicity and period length were analyzed using χ2 tests in the ActogramJ plugin for ImageJ (NIH).

32Female flies were individually inserted into 65-mm glass tubes and loaded into a custom-built array of light-tight chambers, which were each equipped with a high-power LED (Multicomp OSW-4388, 630 nm). The apparatus was operated in a temperature-controlled incubator (Sanyo MIR-154) at 25°C. For movement tracking, the chambers were continuously illuminated from below using low power infrared (850 nm) LEDs and imaged from above with a high-resolution CMOS camera (Thorlabs DCC1545M), using an 8-mm lens (Thorlabs MVL8M23) and a long-pass filter (Thorlabs, FEL800nm) to reject photostimulation light. A virtual instrument written in LabVIEW 9 (National Instruments) extracted real-time position data from video images by subtracting the most recently acquired image from a temporally low-pass filtered background. Non-zero pixels in the difference image indicated that a movement had occurred, with the centroid of the largest cluster of non-zero pixels taken to represent the fly’s new position. To eliminate noise, intensity and size thresholds were applied to pixel clusters in the difference image, and movements < 2.5 mm were discarded. If no movement was detected, the fly was assumed to have remained at its last known location.

33Open-loop stimulation used a continuous train of 3-ms light pulses (∼28 mW/cm2) at 20 Hz. Under closed-loop conditions, blocks of 3 stimulation light pulses (3 ms duration, 20 Hz, ∼28 mW/cm2) were triggered after 3 min of inactivity and repeated every 30 s until the next movement occurred. Periods of inactivity lasting at least 5 min were classified as sleep.

34For each biological replicate, the heads of 400 female flies expressing UAS-EGFP::mL10a in dFB neurons were collected at 4–6 days post-eclosion and homogenized in 500 μl of extraction buffer (pH 7.3) containing 20 mM HEPES, 150 mM KCl, 5 mM MgCl2, 0.5 mM dithiothreitol, 100 μg/ml cycloheximide, 100 U/ml RNaseOUT (Invitrogen), and 1 × cOmplete Protease Inhibitor (Roche). Lysates were centrifuged at 20,000 g for 20 min at 4°C. The supernatants were incubated with 1/8 volume of 10% (v/v) Igepal CA-630 and 1/8 volume of 300 mM 1,2-dihexanoyl-sn-glycero-3-phosphocholine (Avanti Polar Lipids) for 5 min on ice and were then applied to 25 μl Protein G Mag Sepharose beads (GE Healthcare) coated with mouse monoclonal anti-GFP antibody (Htz-GFP-19C8, Memorial Sloan Kettering Monoclonal Antibody Facility). After incubation for 1 h at 4°C, the beads were washed 5 times with 500 μl of wash buffer (pH 7.3) containing 20 mM HEPES, 350 mM KCl, 5 mM MgCl2, 0.5 mM dithiothreitol, 100 μg/ml cycloheximide, 1% (v/v) Igepal CA-630, and 40 U/ml RNaseOUT.

35RNA was isolated from immunoprecipitated ribosomes using the PicoPURE RNA Isolation Kit (Life Technologies) and quantitated with the help of an RNA 6000 Pico Kit (Agilent) on an Agilent Bioanalyzer. Five ng of RNA were reverse-transcribed into cDNA and amplified using the SMART-Seq v4 Ultra Low Input RNA Kit for Sequencing (Clontech). The resulting cDNA was purified using Agencourt AMPure XP beads (Beckman Coulter). Transcript levels were determined by quantitative real-time PCR on a LightCycler 480 system (Roche) using LightCycler 480 Probes Master (Roche) in 5 μl reactions containing 400 nM of each gene-specific primer and ∼5 ng of pre-amplified cDNA. All samples were run in technical triplicates. Relative transcript levels were estimated with the help of the 2-ΔΔCt method, using the geometric mean of the Ct values of three housekeeping genes (Cyp1, Ef1a100E and Rap2l) for normalization.

36Male and female flies with a dorsal cranial window were head-fixed to a custom mount and placed on a spherical treadmill. The treadmill consisted of an air-supported trackball made of extruded styrofoam (13 mm diameter; 50 mg) in a 14 mm tube. An image of a small region of the ball’s surface under 640 nm LED illumination was relayed onto the sensor of an optical mouse (Logitech M-U0017). The sensor was interfaced with a microcontroller board (Arduino Due) based on the Atmel SAM3X CPU and read out in real time using the onboard D/A converter. The resolution of the readout corresponds to 4 mm/s increments in the tangential speed of the trackball.

37The brain was continuously superfused with extracellular solution equilibrated with 95% O2/5% CO2 and containing 103 mM NaCl, 3 mM KCl, 5 mM TES, 8 mM trehalose, 10 mM glucose, 7 mM sucrose, 26 mM NaHCO3, 1 mM NaH2PO4, 1.5 mM CaCl2, 4 mM MgCl2, pH 7.3. Somata of CD8::GFP-labeled helicon cells or R2 neurons were visually targeted with borosilicate glass electrodes (7-13 MΩ). The internal solution contained 140 mM potassium aspartate, 10 mM HEPES, 1 mM KCl, 4 mM MgATP, 0.5 mM Na3GTP, 1 mM EGTA, pH 7.3. Signals were acquired with a Multiclamp 700B amplifier (Molecular Devices), filtered at 6–10 kHz, and digitized at 10–20 kHz using an ITC-18 data acquisition board (InstruTECH) controlled by the Nclamp/NeuroMatic package. Data were analyzed using NeuroMatic software (http://www.neuromatic.thinkrandom.com) and custom procedures in Igor Pro (Wavemetrics).

38For applications of peptides to the FB layer innervated by sleep-control neurons, glass electrodes were filled with 3 mM synthetic AstA (SRPYSFGL-NH2) in extracellular solution or a control peptide containing the same amino acids in a scrambled sequence (GRFSSYLP-NH2). The electrodes were visually guided to the central complex, using GFP-positive neurites as landmarks. The application of a 250 ms pressure pulse (68 kPa; Picospritzer III) resulted in the ejection of ∼40 pl of solution.

39For genetically targeted stimulation of dFB neurons expressing P2X2, a glass electrode containing 500 μM ATP in extracellular solution was positioned unilaterally in the region housing the dendritic fields of these neurons. During periods of stimulation, a 500 ms pressure pulse (68 kPa; Picospritzer III) was applied every 3 s.

40Brains were dissected in PBS (1.86 mM NaH2PO4, 8.41 mM Na2HPO4, 175 mM NaCl) and fixed for 30–45 min in 4% (w/v) paraformaldehyde in PBS at 4°C. For immunostaining, brains were incubated in primary antibodies for 48 h (1:1,000 chicken anti-GFP, Abcam; 1:2 mouse anti-AstA, Developmental Studies Hybridoma Bank, University of Iowa), followed by secondary antibodies for 24 h (1:1,000 anti-chicken antibody conjugated to Alexa Fluor 488; 1:1,000 anti-mouse antibody conjugated to Alexa Fluor 546; both from Invitrogen). Brains containing biocytin-filled neurons were incubated in 1:200 streptavidin conjugated to Alexa Fluor 568 (Invitrogen) in PBS with 0.1% (v/v) Triton X-100 for 48 h. All specimens were mounted in Vectashield (Vector Labs) and imaged on a Leica TCS SP5 confocal microscope.

41Data were analyzed in Prism 6 (GraphPad). Group means were compared by one-way or two-way ANOVA, using repeated-measures designs where appropriate, followed by planned pairwise post hoc analyses using Holm-Šídák’s multiple comparisons test. Where the assumptions of normality or sphericity were violated (as indicated by Shapiro-Wilk and Brown-Forsythe tests, respectively), group means were compared by Mann-Whitney or Kruskal-Wallis tests, the latter followed by Dunn’s multiple comparisons test. Details of statistical analyses are found in figure legends.

42Requests for raw data and instrumentation and analysis code should be directed to the Lead Contact, Gero Miesenböck (gero.miesenboeck@cncb.ox.ac.uk).