A tripartite flip-flop sleep circuit switches sleep states

Inhibitory sleep-active neurons depolarize at sleep onset to shut down the activity of wakefulness circuits. Wake-active arousal neurons in turn suppress inhibitory sleep-active neurons, thus forming a bipartite flip-flop switch. However, how sleep states are switched is unclear because neural circuits that directly depolarize inhibitory sleep-active neurons are not understood. Using optogenetics, we solved the presynaptic circuit for depolarization of the sleep-active RIS neuron in C. elegans. Surprisingly, we found that the PVC forward command interneuron, which is known to control wake behavior, is a major activator of RIS. The PVCs are inhibited by reverse command interneurons, which are stimulated by arousing cues. This suggests a model for sleep switch operation in which declining arousal increases activation of PVC, thus triggering activation of RIS. Depolarization of RIS in turn promotes the activation of PVC, thus forming a positive feedback loop for all-or-none sleep induction. The flip-flop sleep switch in C. elegans thus is tripartite and requires excitatory sleep-promoting neurons activated by wakefulness that act as an amplifier that translates reduced arousal into the depolarization of an inhibitory sleep-active neuron. A tripartite flip-flop switch likely also underlies sleep state switching in other animals including in mammals.


Introduction
optogenetic inhibition of the presynaptic neurons on RIS activation. We used ArchT, which 161 hyperpolarizes neurons by pumping protons out of the cell [36,37]. As before, we 162 specifically illuminated each presynaptic neuron class and quantified RIS activation using 163 calcium imaging. Before lethargus, inhibition of AVJ and PVC led to an inhibition of RIS,164 whereas inhibition of the other neurons tested had no acute significant effect on RIS. 165 During lethargus, only optogenetic inhibition of PVC led to significant RIS inhibition,166 whereas there was no effect seen for the other neurons ( Figure 1C, Figure S2A). While 167 inhibition of RIM had no significant net effect on RIS, in roughly one quarter of the 168 measured trials, the inhibition of RIM led to RIS inhibition before and during lethargus 169 (Figure S2B). 170 171 Our optogenetic analysis revealed a complex set of presynaptic inputs for regulation of RIS 172 activity. The optogenetic gain-of-function experiments suggest that CEP, PVC, and SDQL 173 present the most potent presynaptic activators of RIS. The capacity of PVC to activate RIS 174 is strongly increased during lethargus, indicating that this neuron is involved in the 175 lethargus-specific activation of RIS. The optogenetic loss-of-function experiments suggest 176 that PVC is an essential presynaptic activator of RIS during lethargus, but additional 177 neurons might contribute to RIS activation during lethargus or other conditions. RIM has 178 the potential to both inhibit and activate RIS. While the activation could be direct, RIM is 179 known to be an inhibitor of PVC through activation of the reverse command interneurons 180 AVA/AVD/AVE and could thus perhaps inhibit RIS indirectly [38][39][40]. Consistent with 181 the lack of net effects of RIM on RIS, RIM ablation does not change spontaneous sleep 182 amounts (Figure S2C). Therefore, the majority of synaptic inputs into RIS is activating and 183 inhibition appears to be predominantly indirect. The CEP, URY, and SDQL neurons 184 present sensory receptors and might play a role in activating RIS in response to stimulation 185 [41,42]. Because of the strong and lethargus-specific effects of the PVC neurons on RIS 186 activation we focused our analysis on these key neurons. 187 188 As the PVC neurons are crucial activators of RIS we tested their role in inducing sleep 189 behavior. We used a strain that ablated PVC and other command interneurons by 190 expressing the pro-apoptosis regulator ICE from the nmr-1 promotor and measured sleep 191 and RIS activation [43]. Command interneuron ablation reduced sleep bouts during 192 lethargus by about 76% (Figure S3A). The command interneuron-ablated worms also 193 generally moved much slower (Figure S3B) and RIS activation was reduced by 63% 194 (Figure S3C). Quiescence bouts did not occur at the beginning of the lethargus phase as 195 defined by cessation of feeding, and were only observed around the middle of the lethargus 196 phase ( Figure S3D). An independently generated strain that ablates command interneurons 197 using egl-1 expression caused a reduction of sleep by 81% ( Figure S2A). Because the 198 command interneurons are controlled by glutamatergic signaling we also tested mutants in 199 which this type of signaling is impaired and also observed impaired sleep consistent with 200 ablating command interneurons ( Figure S4). 201

202
To be able to more specifically manipulate PVC and to test the effects of PVC inhibition 203 on behavior without affecting the other command interneurons, we used a more specific 204 promoter for expression in PVC, which had been identified from single-cell RNA 205 sequencing data [44,45](Jonathan Packer, personal communication). There was no gene 206 in the available datasets that was expressed only in the cluster of cells containing PVC, but 207 the previously uncharacterized gene zk673.11 was expressed strongly in PVC and in only 208 a few other neurons excluding other command interneurons and therefore we used the 209 promoter of this gene for PVC expression (Figure S5). Hyperpolarization of PVC through 210 activation of ArchT led to an acute inhibition of RIS, an increase in locomotion, and a 211 reduction of sleep (Figure 1D). These experiments show that RIS is controlled by a 212 command interneuron circuit. The major command interneuron activator of RIS during L1 213 lethargus are the PVC forward command interneurons (Figure 1E). While PVC has 214 previously been shown to promote forward locomotion [46], its inhibition leads to increased 215 mobility implying that the reduction of locomotion after simultaneous ablation of most 216 command interneurons stems from the reverse command interneurons that promote 217 mobility, whereas PVC appears to play a predominant role in dampening motion behaviors 218 through activation of RIS. 219 220 RIS and PVC activate each other forming a positive feedback loop 221 222 PVC presents a major activator of RIS, but how a forward command interneuron can cause 223 massive activation of the RIS neuron during sleep bouts is not clear. We hence tested how 224 optogenetic RIS activation affects PVC activity. We selectively activated RIS using 225 ReaChR and measured calcium activity in PVC (Figure 2A). Because the calcium 226 transients observable in PVC are typically small we used immobilized worms to reduce 227 measurement noise [47]. Upon RIS stimulation, PVC immediately displayed unexpectedly 228 strong calcium transients, which were slightly stronger during lethargus (Figure 2B,Figure 229 S6A inhibition. We hyperpolarized RIS optogenetically for one minute using ArchT and 253 measured the activity of PVC. Interestingly, PVC showed a small but significant activity 254 increase during RIS inhibition, an effect which was increased during lethargus (Figure 2C induce quiescence behavior because its inhibitory neuropeptides are not expressed [27,48]. 267 In aptf-1(-) mutant animals, calcium transients' maxima were reduced by about 35% 268 . These results are consistent with the idea that sleep induction is a self-269 enforcing process in which RIS-mediated inhibition of brain activity promotes further RIS 270 activation ( Figure 2F). 271

RIS depolarization is under homeostatic control 273 274
The design of the sleep circuit suggests an intimate mutual control mechanism of RIS and 275 command interneurons that could allow homeostatic control of sleep. Arousing stimulation 276 is known to inhibit sleep-active neurons and to increase subsequent sleep [24,25,27,29]. 277 We thus hypothesized that inhibition of RIS leads to its subsequent depolarization, forming 278 a homeostat that allows maintaining or reinstating sleep bouts. We tested this hypothesis 279 by optogenetically hyperpolarizing RIS and following its activity using calcium imaging. 280 We inhibited RIS directly for 60 seconds by expressing the light-driven proton pump 281 ArchT specifically in this neuron and used green light illumination to activate ArchT. We 282 followed RIS calcium activity using GCaMP during the experiment and quantified 283 behavior. Optogenetic hyperpolarization of RIS led to a decrease in intracellular calcium 284 and increased behavioral activity. Approximately one minute after the end of the inhibition, 285 RIS showed a rebound activation transient during which calcium activity levels increased 286 strongly and rose well above baseline levels, concomitant with a decrease in behavioral 287 activity. Overall brain activity measurements showed that behavioral activity and brain 288 activity correlated throughout the experiment ( Figure 3A). Strikingly, while the rebound 289 transient was also measurable outside of lethargus, the strength of the RIS rebound 290 depolarization was three-fold stronger during lethargus than before lethargus, indicating 291 that the propensity for RIS rebound activation is strongly increased during lethargus. 292

293
To test whether rebound activation of RIS mediates acute or chronic homeostasis, we tested 294 whether the strength of the rebound activation is a function of length of prior inhibition. 295 For this experiment we increased the length of the RIS inhibition and quantified the time 296 it took after the end of the stimulation until the rebound transient started as well as the peak 297 maximum of the rebound. After inhibiting RIS for five minutes, the rebound initiated 298 immediately after the end of the stimulation and the maximum that was reached exceeded 299 that observed after about one minute of RIS stimulation. Inhibiting RIS for ten minutes did 300 not further increase the occurrence or strength of the rebound transient. These results show 301 that RIS activation rebound transients rapidly saturate with increasing length of inhibition 302 ( Figure 3B, Figure S6B-D) Thus, RIS shows a rebound activation upon inhibition that leads 303 to its own subsequent activation. A rebound activation was not only seen when RIS was 304 inhibited directly, but we also saw a reactivation of RIS after it was inhibited indirectly 305 using a blue light stimulus when the worms returned to sleep ( Figure S7). The rebound 306 activation presents the translation of RIS inhibition into subsequently increased RIS 307 activity and thus sleep induction. Rebound activation of RIS does not seem to constitute a 308 chronic integrator of wake time but presents an acute homeostatic regulatory phenomenon 309 to induce, maintain, or reinstate sleep bouts. 310 311 If RIS is part of a homeostatic system and RIS inhibition triggers its subsequent activation, 312 does activation of RIS also trigger its own inhibition? To test this idea, we optogenetically 313 activated RIS and measured its activity before, during, and after the manipulation. We 314 optogenetically depolarized RIS for one minute using green light and the light-activated 315 cation channel ReaChR expressed specifically in RIS. Activity of RIS was followed using 316 calcium imaging and increased strongly during optogenetic depolarization but was also 317 reduced significantly after stimulation compared with baseline activity, indicating that RIS 318 activity also shows a negative rebound. Again, this effect was highly increased, about five-319 fold, in worms that were in lethargus compared with worms outside of lethargus (Figure 320 3C We reasoned that it should be possible to measure these changes that occur in command 352 interneuron activity upstream of RIS by characterizing neural activity and behavior in aptf-353 1(-) mutant worms. We quantified behavior and command interneuron calcium levels 354 across lethargus in aptf-1(-) mutant worms. Wild type animals showed successive sleep 355 bouts and a 70% reduction in locomotion speed by during lethargus. By contrast, aptf-1(-) 356 mutant animals almost never showed quiescence bouts ( Figure 2E), but nevertheless 357 locomotion speed was decreased by 25% during the lethargus phase (Figure 4). Consistent 358 with the behavioral activity reduction, there was a significant reduction of command 359 interneuron activity during lethargus also in aptf-1(-) mutant animals (Figure 4, Figure S8). 360 To further characterize the neuronal changes upstream of RIS-mediated sleep induction, 361 we imaged the activity of RIM as exemplary neurons during lethargus in aptf-1(-) mutants. 362 In wild type animals, RIM regularly showed activation transients before lethargus but did 363 not show many transients during lethargus. RIM showed not only a change in transient 364 frequency across the lethargus cycle but also a reduction in baseline calcium activity. In 365 aptf-1(-) mutant worms, RIM continued showing RIS transients during lethargus, 366 indicating that RIS inhibits RIM transients during sleep bouts. However, reduction of 367 baseline calcium activity was preserved in aptf-1(-), indicating that RIM activity is 368 dampened during lethargus independently of RIS at the level of baseline calcium activity. 369 Together these experiments indicate that a dampening of behavioral and neural baseline 370 activity occurs during lethargus that is independent of RIS. This neuronal baseline and 371 behavioral dampening itself appears not to be sufficient to constitute normal sleep bouts, 372 but could hypothetically lead to a weaker activity of reverse arousal neurons thus allowing 373 stronger or longer activation of forward command interneurons including PVC and thus 374 RIS activation [43,54]. 375 376 An arousing stimulus inhibits RIS through RIM 377 378 Arousal plays a major role in inhibiting sleep but the circuits that mediate the effect of 379 arousing stimuli on RIS inhibition are not well understood. We hence studied the circuit 380 by which stimulation of a nociceptor, the ASH neurons, leads to a reverse escape response 381 and inhibition of RIS [55]. We optogenetically stimulated ASH using ReaChR and green 382 light and followed RIS and RIM activities. ASH activation led to the activation of the RIM 383 neuron and triggered a backwards response as previously described [55,56]. 384 Simultaneously, RIS was inhibited ( Figure 5A). RIM can inhibit PVC through reverse 385 interneurons which it synchronizes, making it ideally suited to mediate the effects of ASH 386 stimulation [39,46]. To test whether RIM is required for RIS inhibition, we ablated RIM 387 genetically by expression of egl-1 from the tdc-1 promoter, which is specific to RIM and 388 RIC, and repeated the optogenetic stimulation of ASH. We used L4 stage animals as the 389 genetic ablation was not yet effective in L1 animals. In RIM-ablated animals, activation of 390 ASH caused the opposite effect on RIS activity. Instead of inhibiting RIS, ASH activated 391 RIS, while still increasing behavioral activity (Figure 5B). Consistent with our calcium 392 imaging data, ASH stimulation after RIM ablation predominantly caused a forward 393 locomotion response ( Figure 5C). Further supporting the idea that reverse interneurons 394 inhibit and forward PVC neurons activate RIS during stimulation, gentle tail touch 395 increased RIS activity more strongly when RIM was ablated (Figure S9 non-linearly to a strong activation of RIS, which in turn induces the shutdown of behavior 442 that is required for the sleep state [24,31,54,60,61]. The activation of RIS leads to the 443 secretion of inhibitory neuropeptides that inhibit key arousal neurons [27,48]. While PVC 444 is presynaptic to RIS and likely activates it directly, RIS is not presynaptic to PVC, 445 suggesting an indirect mechanism that perhaps could involve inhibition of reverse Dampening of neural activity independently of sleep-active neurons could be interpreted 458 as a neural equivalent of tiredness that leads to an increased propensity to activate sleep-459 active neurons and to induce sleep bouts. Thus, the sleep switch could present an amplifier 460 that translates reduced brain activity into sleep bouts (Figure 6    ]. 644 645 646 HBR1807 goeIs232 ]; 647 goeIs304 goeIs268 ]; 651 goeIs340 . 652

ZC1148
yxIs1 . DNA was prepared as follows: construct 30-100 ng/µl, co-injection marker 5-50 ng/µl, 726 pCG150 up to a concentration of 100 ng/µl if required. Positive transformants were 727 selected according to the presence of coinjection markers. The following plasmids were 728 generated for this study: 729 730 K31 nmr-1p::   All imaging experiments were conducted using either an iXon EMCCD (512x512 pixels), 754 an iXon Ultra EMCCD (1024x1024 pixels), a Photometrics Prime 95B back-illuminated 755 sCMOS camera (1200x1200 pixels) or a Nikon DS Qi2 (4908x3264 pixels). For the iXon 756 cameras the EM Gain was set between 100-200. The exposure times used were between 5-757 30ms. Andor IQ 2 and 3 and NIS Elements 5 were used for image acquisition. were pre-picked onto NGM plates with all-trans-retinal (ATR, Sigma Aldrich) and grown 815 at 25°C. During the two days after exposure to ATR, pretzel stage eggs or L1 worms were 816 taken from this plate for optogenetic experiments. For optogenetic experiments with L4 817 larvae an agar chunk containing a mixed population of growing worms was added to NGM 818 plates containing ATR. Worms for optogenetic experiments were taken from this plate 819 within the next two days. 820 821 Calcium imaging was conducted with an interval of 3s and with an exposure time of 5-822 30ms. A standard optogenetic protocol included calcium imaging during a baseline. This 823 was followed by a stimulation time, in which the worms were optogenetically stimulated. 824 The 585nm light exposure was continuous except for brief interruptions during the time 825 calcium imaging was conducted. After the optogenetic stimulation, calcium images were 826 acquired during a recovery period. 827

828
In mobile worms this standard protocol was preceded by 20 DIC frames that were taken 829 every 500ms to determine if the worm was pumping. The overall protocol was repeated 830 every 15-30min. L1 mobile worms were imaged with a 20x objective and a 0.7 lens. Mobile 831 L4 worms were imaged with either a 10x objective ( Figure 5A-C) or a 20x objective 832 (Figure1B/CEP, 1C/URY, S1A/CEP, S2A/URY). Fixed worms were usually imaged 833 between 1-4 trials. A delay preceded the standard protocol to allow the worm to recover 834 between trials. A 100x oil objective was used for the experiments. To specifically 835 manipulate PVC and SDQL in Figure 1B-C, S1A, S2A, the stimulating illumination was 836 restricted to the neuronal areas. The details for optogenetic experiments can be found in 837 Supplementary Table S1. The protocol was repeated every 15min. First, 20 DIC frames were taken every 500ms to 843 determine whether the worm was pumping or not. Next, baseline GCaMP was imaged for 844 3 min, the stimulation phase then lasted 18s and a recovery phase was imaged for 3 min. 845 The 490nm intensity for calcium imaging was 0.07mW/mm 2 . The 490nm intensity for 846 stimulation was set to 1.01 mW/mm 2 with a 20x objective. The same LED was used for 847 calcium imaging and stimulation. The intensity levels were controlled with Andor IQ2 848 software. 849

850
The RFP signal of the pan-neuronal strain was imaged in addition to the GCaMP signal 851 during the protocol every 3s with 585nm LED illumination, which was set to 852 0.17mW/mm 2 . 853 854 Spinning Disc Confocal Microscopy 855 856 L4 worms were fixed with Levamisole. Spinning disc imaging was done with an Andor 857 Revolution disc system using a 488nm (0.34 mW/mm 2 ) and a 565nm (0.34 mW/mm 2 ) laser 858 and a Yokogawa (Japan) CSU-X1 spinning disc head. Worms were imaged through a 100x 859 oil objective. In Figure S5A-B an additional 1.5 lens was used. Z-stacks with z-planes 860 0.5µm apart spanning a total distance of 10µm were taken and a maximum intensity 861 projection calculated in ImageJ. The exact speed and time thresholds were adjusted empirically to represent the worms' 937 behavior [84]. In Figure 2D-E worms had to have a speed below 5% of their maximum 938 smoothed speeds for at least 2min in order to be counted as sleeping. For all other 939 experiments the speed threshold was 10% and the time threshold was 2min. The sleep bout 940 analysis was carried out with a custom-written MATLAB script. 941

942
For stimulation experiments, the baseline and recovery time measurements were too short 943 to include a minimum time threshold in the sleep bout analysis. Hence, immobility was 944 used as a proxy for sleep. A mean of the wake speeds was calculated for each worm. In 945 most experiments a worm was counted as sleeping when its speed was below 10% of the 946 calculated mean of the wake speeds. In Figure 3A, C the threshold was adjusted to 30% 947 and in Figure S6 to 50% to account for a different locomotor behavior of the worms. RIS 948 signals and speeds of wild type and mutants were aligned to sleep bout onset for 949 comparison in Figure S2C, S3C, D. For GCaMP normalization ten data points before sleep 950 bout onset were taken as baseline in order to calculate DF/F. In Figure S7C  The data in Figure 3B was fitted to an asymptote with Origin software. The data in Figure  956 S5 was fitted to a logistic regression using Origin software. Exact functions and R 2 values 957 can be found in the respective figures. RIS shows rebound activation following hyperpolarization. Behavioral and brain 1331 activity measurements correlate throughout the whole experiment; *p < 0.05, **p 1332 < 0.01, ***p < 0.001, Wilcoxon Signed Rank Test for GCaMP and speed, 1333 Fisher's Exact Test for sleep fraction.    The dampening of neural and behavioral baseline activity levels during lethargus is 1499