Parallel cholinergic circuit in oculomotor nucleus to control eye movements and REM sleep

Abstract

Sleep suppresses most motor output, yet rapid eye movement sleep (REMs) is marked by brief muscle twitches, including eye movements (EMs), indicating selective motor preservation. How these opposing features are coordinated remains unclear. Here we show that cholinergic neurons in the oculomotor nucleus (nIII^ChAT) possess dual functional organization governing REMs termination and oculomotor execution in mice. We identify a ventrolateral periaqueductal gray-projecting nIII^ChAT subset whose activity dissociates from EMs but undergoes a progressive increase prior to REMs termination. Optogenetic activation of this subset suppresses REMs without affecting EMs, whereas extraocular muscles-projecting nIII^ChAT neurons reliably induced EMs without altering sleep-wake states. These motor and sleep-modulating functions of nIII^ChAT neurons are coordinated by genetically distinct Vglut2- and Vgat-expressing neurons in the upstream nucleus papilio. Collectively, our findings revealed a microcircuitry underlying REMs exit, where motor related function of nIII is preparing the mammal to enter the subsequent wakefulness with elevated motor capabilities.

While eye movements (EMs) serve as a critical physiological marker for distinguishing NREM from REM sleep, they have traditionally been characterized as a secondary byproduct of the sleep state. Here authors show that a dual role for cholinergic neurons of the oculomotor nucleus (nIII) in coordinating REM sleep and its associated EMs. This work provides insights into oculomotor activity during the late phase of REM sleep, suggesting a mechanism that prepares mammals to transition into wakefulness with enhanced motor readiness.

Introduction

REMs is an “active” phase of sleep, hallmarked by prominent EMs¹. During REMs, postural motor neurons are inhibited, and muscles are relaxed, with the exception of EOMs². As a prominent pool of cholinergic neurons within the midbrain^3,4, the oculomotor nucleus (nIII) has been suggested to modulate oculomotility, not only during wakefulness⁵ but also during REMs, although there are some discrepancies^6,7. In alert animals, spontaneous EMs consist of voluntary and involuntary saccades that move and keep the eyes to a visual target, enabling visual acquisition^8,9. These modalities of EMs are coordinated by nIII and other nuclei, the trochlear nucleus, and the abducent nucleus. Specifically, abducting and adducting EMs are mediated by motoneurons in the abducens nucleus and nIII, respectively, whereas the torsional rotation of the eyeball is controlled by the trochlear nucleus via its innervation of the superior oblique muscle^9,10. With regard to EMs in sleep, Herrera et al. provided evidence that the activation of the nIII is responsible for the occurrence of EMs during REMs⁷, spawning new data on the behavior of this visuomotor system during sleep.

Ludwig Mauthner postulated the existence of a putative “sleep center” situated near the nIII within the midbrain¹¹. Since then, multiple nuclei in close proximity to the nIII have been identified as potential regulators of sleep, such as GABAergic neurons in the dorsal norepinephrine bundle, cholinergic neurons in the laterodorsal tegmentum (LDT) in the genesis of REMs^12–14, as well as the glutamatergic neurons in the dorsal deep mesencephalic nucleus (dDpMe) in the promotion of NREM sleep (NREMs)^15,16. A recent study identified sleep-active neurons within the perioculomotor (pIII) region and Edinger-Westphal nucleus (EWcp), both of which were adjacent to the nIII^17,18. It was found that activating Calca neurons in the pIII region or activating GHSR signaling in EWcp facilitates NREMs^17,18. Meanwhile, the nIII has abundant afferent connections with the midbrain reticular formation, a key regulator of REMs^19,20. Cholinergic neurons within the nIII displayed progressively augmented activity with increasing REM sleep deprivation duration, as evidenced by a marked increase in c-Fos expression in nIII^ChAT neurons²¹. These findings suggest that nIII^ChAT neurons might play an important role in regulating both vigilance states and EMs. However, the definite functional role of nIII^ChAT neurons in sleep regulation is yet to be studied.

Here, we first observed the elevated c-Fos expression of nIII^ChAT neurons during recovery sleep after deprivation in mice. Using fiber photometry and an electrophysiological approach, we identified a population of cholinergic neurons in the nIII that are active in response to repeated REM-to-wake (R-W) transitions. Calcium imaging revealed that those neurons were functionally distinct from the oculomotoneurons and formed a densely interconnected network. Viral tracing experiments revealed that, distinct from extraocular muscles (EOMs)-projecting nIII^ChAT neurons, this newly identified nIII^ChAT projects to the ventrolateral periaqueductal gray (vlPAG). We then selectively inhibited and activated these nIII^ChAT neurons to investigate their involvement in sleep regulation. Finally, using optogenetics-assisted electrophysiological recording combined with a retrograde labeling approach, we unveiled that the upstream nucleus papilio (NP) functions as an initiator. The NP^Vglut2-nIII-EOMs pathway is responsible for oculomotility in REMs, whereas the NP^Vgat-nIII-vlPAG circuit gates the switch of REMs. Taken together, our results implicate the functional diversity of nIII^ChAT neurons and underscore the dual role of the nIII in motor function and REMs regulation mediated by distinct subsets of cholinergic neurons.

Results

nIII^ChAT population activity correlates with REMs and EMs

Through immunostaining co-localization with choline acetyltransferase (ChAT), we first identified that the majority of neurons (82.12% ± 3.38) in the nIII are cholinergic (Fig. 1a, b). To date, although the detailed topological projections of these neurons to the EOMs have been previously described²², we have a limited understanding of the coding of oculomotor behavior because the activity profile of nIII^ChAT neurons during oculomotor behavior has not been specifically characterized^19,22. To first investigate the involvement of nIII^ChAT neurons in EMs, we stereotactically injected AAV2/9-hSyn-FLEX-GCaMP6s-WPRE into the nIII of ChAT-Cre mice, along with the implantation of an optic fiber targeting the nIII (Fig. 1c). Immunostaining results showed the expression of GCaMP6s in the nIIIs (Fig. 1d). Subsequently, we recorded calcium signals of nIII^ChAT neurons during EMs induced by the optokinetic reflex (OKR) test (Supplementary Fig. 1a). EMs in head-fixed mice were captured by an infrared camera and pupil position was converted into a temporal-nasal (T-N) and dorsal-ventral (D-V) coordinate system (Supplementary Fig. 1b). We found that nIII^ChAT neurons exhibited enhanced activity during EMs occurring in T-N or D-V directions, while showed the reduced responses during EMs in N-T direction (Supplementary Fig. 1c–e). The peak response of nIII^ChAT neurons was observed when the eyeball moved from the T-N direction, ranging between 120°−160° (Supplementary Fig. 1f).

Fig. 1. Population activities of nIII^ChAT neurons across sleep-wake states.

a Confocal image showing co-localization of NeuN⁺ (green) and ChAT⁺ (red) within the nIII. Scale bar, 50 μm. b Fraction of ChAT⁺ neurons in the nIII (n = 3 mice). c Schematic of fiber photometry and in vivo recording setup. d Confocal image of GCaMP6s expression and optical fibers placement in ChAT-cre mice. Scale bar, 100 μm. e Experimental timeline. f Representative EEG power, EMG, EMs, and fluorescence traces across spontaneous sleep-wake cycles. Red dots in the magnified EM traces indicate EM events. g Average activity of nIII^ChAT neurons across vigilance states with (EMs-enriched episodes) or without (EMs-free episodes) EMs. Each dot represents one mouse (n = 4 mice). Data are presented as mean ± SEM. Two-way repeated-measures ANOVA followed by Tukey’s multiple comparisons test. *p < 0.05, **p < 0.01, ***p < 0.001. h Distribution of EMs and calcium signals during REM-to-wake transitions (88 episodes). Upper: raster plot and histogram of EMs aligned to transition. Middle: fluorescence heatmap of individual transitions sorted by mean activity. Lower: average ΔF/F (z-scored) during the transition. Lines represent the mean across episodes, and shaded areas indicate 95% confidence intervals (CIs). Each REM and wake episode was temporally normalized. Dashed line indicates the transition time point. i Same as (h) for NREM-to-REM transitions (73 episodes). j REM-to-wake transitions for EMs-free episodes. Upper: raster plot and histogram of EMs aligned to transition. Middle: fluorescence heatmap of individual transitions sorted by mean activity. Lower: average ΔF/F (z-scored) during the transition. The dashed line indicates the time point of transition. k Same as (j) for NREM-to-REM transitions for EMs-free episodes. Data represent at least three biological replicates. Each dot indicates one mouse. Error bars represent SEM across mice. All data and significance values for graph (g) are shown in Supplementary Table 1.

To investigate the activity of nIII^ChAT neurons during sleep, we conducted a c-Fos staining experiment after 8 h sleep deprivation (SD, using the small-platform-over-water method) or during the subsequent sleep recovered (SR) in binocularly enucleated mice (Supplementary Fig. 2a, c). We observed that the proportion of c-Fos⁺/ChAT⁺ neurons among nIII^ChAT neurons was higher in the SR group (37.74%) compared with the SD group (14.01%) (Supplementary Fig. 2g, h). We also deprived the REMs of mice specifically by employing a closed-loop sleep interference system (Supplementary Fig. 2b), which achieves a state-of-the-art performance in REMs classification (Supplementary Fig. 2d, e)²³. During the 8-h deprivation period, the proportion of REMs was reduced from 8.1% in the control group to 1.2% in the deprived group, accompanied by a significant increase in wakefulness (Supplementary Fig. 2f). Subsequently, we observed a significantly increased c-Fos expression in the SR^REM group compared to the SD^REM group (Supplementary Fig. 2i, j). These evidences indicate that the activation of nIII^ChAT neurons was associated with sleep, especially REM episodes.

To monitor the activity of nIII^ChAT neurons throughout sleep-wake cycles, mice were implanted with a custom-made polysomnography recording device²⁴ to simultaneously record electroencephalogram (EEG), electromyogram (EMG) and EMs signals(Fig. 1e). Calcium imaging, along with EEG/EMG/EMs recordings, was conducted for 1 h during each session in the light phase (Fig. 1e). Sleep-wake states were scored based on the polysomnography recording according to previous protocol²⁴. As shown in Fig. 1f, calcium signals of nIII^ChAT neurons exhibited increased activity during REMs (highlighted in green). Taking the average calcium signal of the entire recording as the baseline, the overall activities of nIII^ChAT neurons was highest during REMs compared to wake and NREMs (Fig. 1g). Moreover, we categorized REMs episodes into EMs-enriched and EMs-free episodes and found that EMs-enriched REM episodes exhibited slightly lower activity than EMs-free REM episodes (Fig. 1g, Supplementary Fig. 3a). The calcium signals of nIII^ChAT neurons during wake were smaller than those during REMs episodes and showed more negative values in EMs-free wake episodes compared to EMs-enriched wake episodes (Fig. 1g). These observations suggest the existence of a subpopulation of nIII^ChAT neurons that are hyperactive during REMs episodes without EMs.

Given that REMs are typically followed by wakefulness in a physiological context^12,25, here we further examined the neuronal activity during REM-to-wake (R-W), NREM-to-REM (NR-R), NREM-to-wake (NR-W), and wake-to-NREM (W-NR) transition. To account for variability in episode duration, the time axis was normalized to enable direct comparisons of neuronal activity across transitions (Supplementary Fig. 3d). We found that nIII^ChAT neurons displayed increased activity preceding R-W transitions, which declined upon entry into wakefulness (Fig. 1h). The activity of nIII^ChAT neurons gradually increased after the time point of NR-R transitions (Fig. 1i). These trends in neuronal activity remained consistent during EMs-free R-W and EMs-free NR-R transitions (Fig. 1j, k), excluding the possibility that increase in calcium activity during REMs was simply due to the EMs. Furthermore, the heightened activity of nIII^ChAT neurons was accompanied by the rising cumulative probability of R-W transitions, while concurrently decreasing the probability of NR-R transitions (Supplementary Fig. 3b, c). In contrast, the calcium signals showed no significant changes during NR-W and W-NR transitions (Supplementary Fig. 3e, f), especially during EMs-free episodes (Supplementary Fig. 3g, h). These findings exhibited that the nIII^ChAT neurons exhibit selective hyperactivity during REMs, independent of EMs occurrence.

Optrode recording from non-EOMs-projecting nIII^ChAT neurons

To further corroborate these findings, we performed in vivo electrophysiological recordings of non-EOMs-projecting nIII^ChAT neurons (Supplementary Fig. 4a). We first injected AAV2/9-hEF1α-DIO-hChR2(H134R)-EYFP into the nIII of ChAT-Cre mice and implanted an optrode at nIII. Each optrode comprises 20 channels in total, including 16 nickel-titanium alloy wire electrodes for spike recording arranged around an optical fiber, along with 4 additional silver wire electrodes for EEG and EMG recordings. Using an optrode, we were able to identify ChR2-expressing nIII^ChAT neurons based on laser-evoked firing and EMs while simultaneously recording EEG/EMG signals (Supplementary Fig. 4b). Subsequently, we injected retrograde AAV2/2Retro Plus-CAG-DIO-taCaspase3-TEVp-WPRE-pA into the EOMs to selectively ablate nIII^ChAT neurons that project to EOMs. This ablation almost completely abolished laser-evoked EMs (Supplementary Fig. 5a, b), confirming that the recorded neurons barely innervated EOMs.

After 3 weeks’ recovery, single units exhibited reliable laser-evoked spiking from ChR2-tagged, putative non-EOMs-projecting nIII^ChAT neurons (Supplementary Fig. 5c, d). We then performed long-term spike recordings combined with EEG and EMG monitoring (Supplementary Fig. 4b). Supplementary Fig. 4c, d shows the activity of non-EOMs-projecting nIII^ChAT neurons across the sleep-wake cycle. In line with the fiber photometry findings (Fig. 1g), nIII^ChAT neurons exhibited the highest firing rate during REMs (Supplementary Fig. 5e). Their firing rate increased gradually over the course of each REM episode before the transition to wake or after the transition from NREMs (Supplementary Fig. 5f, g). The consistency of this activity pattern across recorded units supports the notion that nIII^ChAT neurons are indeed active during REMs, even in the absence of EMs.

Silencing nIII^ChAT neurons prolongs REM sleep episodes

We first confirmed the role of nIII^ChAT neurons in regulating EMs by conducting bilateral injections of AAV-CAG-DIO-taCaspase-3 into the nIII of ChAT-Cre mice (Fig. 2a). Three weeks after the injection, the density of nIII^ChAT neurons in the ablation group reduced to 13.1% of the control group (control: 529.0 neurons/mm², ablation: 58.0 neurons/mm²) (Fig. 2b, c). We conducted OKR experiments to examine the effects of bilaterally ablating nIII^ChAT neurons on EMs (Fig. 2d). We found significantly smaller amplitudes of OKR-evoked EMs in both N-T and T-N directions for the ablation group compared to the control group (Fig. 2e, f).

Fig. 2. Loss of nIII^ChAT function attenuates REM episodes and elongates REM duration.

a Schematic of viral injection (upper) and sleep recording (lower) in ChAT-Cre mice. b Confocal images showing selective ablation of nIII^ChAT neurons. ChAT, red. NeuN, green. Scale bar, 50 μm. c Quantification of remaining nIII^ChAT neurons (control n = 3, ablation n = 4). Two-sided unpaired t-test. d EMs induced by grating motion on the OKR platform before nIII ablation. e Representative EMs traces along nasotemporal (N-T) and temporonasal (T-N) directions in the control (gray trace) and ablation group (blue trace). f Quantification of OKR-induced EMs amplitude (control: n = 7, ablation: n = 5). Two-way ANOVA followed by Tukey’s multiple comparisons test. g Time course of wake (left), NREMs (middle), and REMs (right) across 24 h (n = 5). Two-way RM ANOVA followed by Bonferroni post hoc analysis. h Total time spent in each vigilance stage (n = 5). Two-sided unpaired t-test for the 24 h time phase. Two-way ANOVA followed by Bonferroni’s test for the comparison of day and light phases. i Duration of REM episodes in the ablation group compared to the control group (n = 5). Two-way ANOVA followed by Bonferroni post hoc analysis for the histogram. Two-sided Kolmogorov–Smirnov test for cumulative probabilities. j Number of REM episodes in the ablation group compared to the control group (n = 5). Two-way ANOVA followed by Bonferroni post hoc analysis. k Number of REM-to-wake transitions during 24 h sleep recording (Control: n = 5, Ablation: n = 5). Two-sided unpaired t-test. l Schematic depicting stGTACR-mediated inhibition of nIII^ChAT neurons. m Confocal image of viral expression and optical fiber placement within nIII. The dashed gray box indicates the trace of the optical fiber. n Optogenetic inhibition paradigm during the light phase. o Percentage of wake, NREMs, or REMs before, during, and after optogenetic inhibition in ChAT-Cre mice (n = 6). Lines represent the mean across mice, and shaded areas indicate 95% CIs. The gray shielding indicates the time window of stimulation. p Quantification of state percentages before (pre-stim.) and during (stim.) optogenetic inhibition (n = 6). Two-sided paired t-test. All data are presented as mean ± SEM. Each dos indicates individual mice. Data represent at least three biological replicates. All data and significance values for graphs (c, f–k, and p) are shown in Supplementary Table 1. *p < 0.05, **p < 0.01, ***p < 0.001, n.s., no significant.

The hourly amounts of REMs in ChAT-Cre mice were markedly increased at the expense of wake during both the light and dark phases following the ablation of nIII^ChAT neurons (Fig. 2g, h). The EEG power spectra of each brain state (wake, NREMs, and REMs) following ablation-induced increase of REMs were similar to those displayed during spontaneous sleep-wake (Supplementary Fig. 6a). Detailed analysis revealed that the substantial increase in total REMs time observed in nIII^ChAT-ablated mice was due to an augmentation not in the number but in the duration of REMs episodes (Fig. 2i, j). As a consequence of less fragmentation of REMs in nIII^ChAT-ablated mice, the number of R-W transitions was decreased (Fig. 2k). The ablation of nIII^ChAT neurons had no significant effect on the duration and number of wake or NREMs episodes (Supplementary Fig. 6b, c). We also did not observe a significant change in the overall amplitude and frequency of EMs during wake, NREMs, and REMs (Supplementary Fig. 6d, e). These findings indicate that ablation of nIII^ChAT neurons likely facilitates the maintenance of REMs and consequently impairs the physiological transition from REMs to wake.

To clarify the necessity of nIII^ChAT neurons in regulating REMs, beyond the cell ablation approach, we have performed acute and cell-type-specific optogenetic inhibition of nIII^ChAT neurons during polysomnographic recordings. We expressed AAV2/9-hSyn-DIO-stGtACR2-mCherry-WPRE-pA in the nIII^ChAT neurons in ChAT-Cre mice (Fig. 2l, m). Three weeks post-injection, optogenetic inhibition of nIII^ChAT neurons was performed during the light phase (12:00–18:00) in 30 sessions, each comprising a 2 min, 20 Hz stimulation block (indicated by blue arrows) delivered at 10 min intervals (Fig. 2n). We found that optogenetic inhibition of nIII^ChAT neurons induced a reliable EMs in the nasal-to-dorsotemporal direction (Supplementary Fig. 6f, g). We further examined the dynamic changes in EEG, EMG, and EMs signals across different vigilance states. We found that the occurrence rate of EMs rapidly increased during the stimulation period across multiple trials (Supplementary Fig. 6h). The EEG power spectra and EMG of each brain state during inhibition were similar to those displayed during spontaneous sleep-wake cycle, indicating that optogenetically evoked EMs did not interfere with EEG or EMG signals and therefore did not compromise the accurate classification of sleep stages (Supplementary Fig. 7a–f). We then evaluated the sleep architecture and observed that optogenetic inhibition of nIII^ChAT neurons significantly increased the proportion of REMs (Fig. 2o, p). These results are consistent with the nIII^ChAT neuronal ablation experiment, indicating that silencing nIII^ChAT neurons could lead to the generation and maintenance of REMs.

Activation of the nIII^ChAT neurons facilitates the termination of REMs and the occurrence of EMs

Given that the fundamental role of the nIII is oculomotor control, we injected AAV2/9-hEF1α-DIO-hChR2(H134R)-EYFP into the nIII (Supplementary Fig. 8a, b). Three weeks after viral expression, optogenetic activation was performed using 480 nm laser illumination, and the eyeball position of head-fixed mice was recorded with an infrared camera (Supplementary Fig. 8c). We observed a reliable ipsilateral EMs in T-N direction but had no effect on the pupil constriction (Supplementary Fig. 8d–f). These results confirm that we selectively activated the nIII without affecting the adjacent Edinger-Westphal (EW) nucleus. The amplitude of EMs increased as the pulse durations increased from 5 ms to 100 ms, and reached a plateau as the durations increased further (Supplementary Fig. 8e, f). When the duration of the optogenetic pulse exceeded 500 ms, it might induce a prolonged contraction of the oculomotor muscles, which was subsequently followed by sustained ocular immobility (Supplementary Fig. 8e). Moreover, time-locked EMs induced by low-frequency laser pulses (3.3 Hz, 100 ms duration) were robust, while 20 Hz optical stimulation induced very smaller amplitude of EMs (Supplementary Fig. 8g, h, Supplementary Movie 1 and 2).

We next examined the role of nIII^ChAT neurons in REMs modulation using acute optogenetic activation. Optical stimulation was delivered for 2 min per trial during polysomnographic recordings, with stimulation initiated at least 20 s after the onset of stable NREM or REM sleep. A total of 10–15 blocks were conducted, with a minimum interval of 10 min between blocks (Supplementary Fig. 9a). Optogenetic stimulations were conducted between 12:00 and 17:00, a period when mice exhibit high sleep stress and increased propensity to fall asleep²⁶. We also examined the dynamic changes in EEG, EMG, and EMs across different vigilance states (Supplementary Fig. 10). We found that optogenetic activation reliably induced EMs regardless of the current vigilance state, with rapidly increased EMs rate during the stimulation window (Supplementary Fig. 9b, c). The EEG power spectra and EMG activity across brain states during optogenetic activation were comparable to those observed during spontaneous sleep-wake cycles, ensuring the accurate identification of sleep stages. (Supplementary Fig. 10a–f). We found that activating nIII^ChAT neurons induced awakening of mice, producing EEG and EMG patterns characteristic of wakefulness (Supplementary Fig. 9d, e), accompanied by a significant reduction in theta EEG oscillations during stimulation, which are typically dominant during REMs (Supplementary Fig. 9f). Optical activation at 3.3 Hz and 20 Hz resulted in a 57.7% and 74.2% reduction, respectively, in the fraction of REMs compared to the baseline period (Supplementary Fig. 9g, i). In comparison, no significant change was observed across all vigilance states during 590 nm (with no activation of ChR2-expressing neurons) laser illumination (Supplementary Fig. 9h, i).

We further dissected the role of nIII^ChAT on REMs by applying optogenetic stimulation during REMs. When optical stimulation was delivered during REMs episodes, the probability of REMs dramatically decreased from 57.1% in the control group to 18.8% with 3.3 Hz stimulus or to 14.3% with 20 Hz stimulus, accompanied by a complementary increase in wake and NREMs (Supplementary Fig. 9j, k). The increase in wake was more notable than the increase in NREMs. Furthermore, activating nIII^ChAT neurons results in a notable reduction in the latency to wake when juxtaposed with the control group (Supplementary Fig. 9l). This reduction in latency from R-W aligns with a comparable latency index (defined as the ratio of REMs latency to the total duration of the REMs episode), indicating the dissipation of intrinsic REMs pressure²⁷.

To evaluate whether the REMs-suppressing effect of optogenetic activation of nIII^ChAT neurons could be influenced by homeostatic sleep pressure, we also performed a long-term (1 h) optogenetic experiment while simultaneously recording the electrophysiological activity of ChR2-tagged nIII^ChAT neurons using an optrode (Supplementary Fig. 11a, b). Prolonged activation also led to a significant decrease in the probability of REMs and an increase in the probability of wake compared to the control condition, although this effect progressively diminished toward the end of the long-term stimulation (Supplementary Fig. 9m). Similarly, the firing rate of single-unit also shows a gradual decline with stimulation time, while the spontaneous firing activity in the control group (no-stim.) remained stable (Supplementary Fig. 11c). That means that the progressively reduced effect by the end of the long-term stimulation may owing to the fatigue and desensitization of ChR2-tagged neurons²⁸. Additionally, the amount of time mice spent in REMs decreased, accompanied by an increase in the duration of wake episodes, compared to the control group (Supplementary Fig. 9n). Taken together, these results provide evidence that activating nIII^ChAT neurons can result in the REMs exit and promote the R-W transition.

Effect of optogenetic activation of the non-EOMs-projecting nIII^ChAT neurons on sleep

To exclude potential interference from EMs on sleep architecture changes during optogenetic stimulation, it is essential to distinguish nIII^ChAT neurons projecting to EOMs from those involved in REMs. To this end, we first investigated the projections of nIII^ChAT neurons to the EOMs and found that EOMs-projecting nIII^ChAT neurons constituted 33.60% of ipsilateral and 11.50% of contralateral nIII^ChAT neurons, respectively, based on data from 6 mice, with 3 representative sections analyzed per animal (Supplementary Fig. 12a-c). We eliminated these EOMs-projecting nIII^ChAT neurons by injecting AAV2/2-Retro-DIO-taCaspase3-WPRE into EOMs in ChAT-Cre mice. Subsequently, AAV-hEF1α-DIO-hChR2-GFP was injected into the nIII followed by optic fiber implantation (Fig. 3a). Despite anatomical labeling and caspase expression in EOMs-projecting nIII^ChAT neurons may not be complete, the genetic ablation resulted in a 39% reduction in the number of nIII^ChAT neurons (Fig. 3b, c), a proportion that closely matches the percentage of EOMs-projecting nIII^ChAT neurons (Supplementary Fig. 12c). The amplitude of EMs elicited by optogenetic activation of nIII^ChAT neurons was dramatically decreased after neuronal ablation (Fig. 3d and Supplementary Fig. 13a, Supplementary Movie 3 and 4). Optogenetic activation (at 3.3 Hz or 20 Hz for 2 min) of the rest nIII^ChAT neurons (putative non-EOMs-projecting nIII^ChAT neurons) did not change the occurrence rate of EMs during stimulus time window across trials (Supplementary Fig. 13b, c), confirming the functional loss of EOMs-projecting nIII^ChAT neurons.

Fig. 3. Non-EOMs-projecting nIII^ChAT neurons modulate R-W transition.

a Schematic of ChR2-mediated activation after ablating EOMs-projecting neurons. b Confocal image of residual nIII^ChAT neurons after retrograde ablation. Scale bar, 50 μm. c Fraction of residual nIII^ChAT neurons after retrograde ablation (n = 4). Two-side unpaired t-test. d Peak amplitude of laser-evoked EMs after ablation (n = 7). Two-sided unpaired t-test. e Stimulation paradigms (3.3 Hz, 20 Hz; 2 mW) applied during sleep or REMs. f Confocal image of viral expression and optical fiber placement. The dashed gray box indicates the trace of the optical fiber. g Representative EEG, EMG, and EMs traces during optical stimulation. Blue shielding area indicates the illumination coverage time window. h Percentage of wake, NREM, and REM before, during, and after activation (3.3 Hz and 20 Hz, n = 8). Lines represent the mean across mice, and shaded areas indicate 95% CIs. The gray shielding indicates the time window of stimulation. i Same as (h) for control stimulation (590 nm, n = 8). Lines represent the mean across mice, and shaded areas indicate 95% CIs. The gray shielding indicates the laser stimulation window. j State probabilities during REM-specific activation. Lines represent the mean across mice, and shaded areas indicate 95% CIs. The gray shielding indicates the time window of stimulation. k Pie charts of sleep-wake state when optogeneticly activating the nIII^ChAT neurons during REMs. l Changes in state percentages during optogenetic stimulation relative to baseline (n = 8). One-way RM ANOVA followed by Tukey post hoc analysis for each state. m Definition of latency index for R-W transition. n Correlation between latency index and preceding REM duration. Shaded areas represent 95% CIs of the linear regression. Pearson’s correlation analysis was used to assess the relationship between variables. o Latency from stimulation onset to wake when laser stimulation during REMs (n = 8). One-way ANOVA followed by Tukey post hoc analysis. All data are presented as mean ± SEM. Each dos indicates individual mice. Data represent at least three biological replicates. All data and significance values for graphs (c, d, l, and n, o) are shown in Supplementary Table 1. *p < 0.05, **p < 0.01, ***p < 0.001.

As previously described, we performed short-term optogenetic stimulation during sleep to evaluate the function of the non-EOMs-projecting nIII^ChAT neurons after the ablation of EOMs-projecting one (Fig. 3e, f). REMs were immediately interrupted upon activating non-EOMs-projecting nIII^ChAT neurons (Fig. 3g), and the decreased probability of REMs was accompanied by an elevation in wake compared to the control group (Fig. 3h, i, l). Additionally, the power of REMs-dominated theta oscillation (6–9 Hz) also decreased during stimulation (Supplementary Fig. 13d). When the optical stimulation was specifically delivered during the REM period, the probability of REMs decreased drastically from 55.6% to 19.7% upon 3.3 Hz optogenetic stimulus or to 16.3% by 20 Hz optogenetic stimulus (Fig. 3j-k). However, no discernible alteration was noted when the optogenetic stimuli were presented during wake or NREMs (Supplementary Fig. 13e–h). Consistently, the latency from the onset of light stimulus to the R-W transition was significantly shortened upon optogenetic stimulations (Fig. 3m–o).

To further dissociate EMs control from REMs modulation, we injected retroAAV-hSyn-Cre-WPRE-pA into the EOMs of C57 mice and subsequently injected AAV2/9-hEF1α-DIO-ChR2(H134R)-GFP-WPRE into the nIII to selectively express ChR2 in nIII neurons projecting to the EOMs (Supplementary Fig. 14a, b). Optogenetic activation of these neurons induced robust EMs (Supplementary Fig. 14c, d), yet did not significantly alter the proportion of time spent in different sleep states (Supplementary Fig. 14e, f). Collectively, these results support the conclusion that activation of nIII^ChAT neurons directly drives R-W transitions, rather than exerting an indirect influence via evoked EMs.

Functionally distinct nIII^ChAT subnetworks

To explore how nIII^ChAT neurons contribute to R-W transition and EMs, we monitored the activity of nIII^ChAT neurons at single-cell resolution by endoscopic calcium imaging. AAV2/9-hSyn-FLEX-GCaMP6s-WPRE was injected into the nIII of ChAT-Cre mice, and a gradient refractive index (GRIN) lens was implanted over the injection site, along with a custom-made EEG/EMG/EMs polysomnography recording device²⁴ (Fig. 4a, b). The expression of GCaMP6s in nIII^ChAT neurons was shown by immunostaining 4 weeks post-injection (Fig. 4c). We recorded the activity profiles of nIII^ChAT neurons for 2 h per session from 5 mice, at least two sessions per mouse (Supplementary Movie 5). Each session encompassed multiple episodes of wake, NREMs, and REMs (Fig.4d). We observed prominent calcium signals in spatially separated individual nIII^ChAT neurons during EMs in REMs or wake (cells 15, 18, 31 in Fig.4e, f), as well as during R-W transition (cells 3, 4 in Fig.4e, f). There were also some neurons that exhibited non-specific responses across these episodes (Fig. 4f).

Fig. 4. Subgroups of nIII^ChAT neurons show distinct response to R-W transition and EMs.

a Schematic of calcium imaging and polysomnography recording in ChAT-Cre mice. b Schematic of viral injection and GRIN lens implantation into the nIII. c The confocal image of GCaMP6s expression and GRIN lens placement. d Experimental timeline. e Representative fields of view (FOVs) of nIII^ChAT neurons imaged with a miniature microscope. f Representative calcium activity across vigilance states and EMs. The green shielding indicates the REM episodes, and the gray shielding indicates the occurrence of EM events during REMs. The red dots label the events of EMs. g Correlation matrix (left) and clustering of neuronal activity (right) during R-W transitions, showing functional ensembles. h Same as (g) for EMs-related activity. i Functional neuronal network of R-W specific neurons (green) and EMs specific neurons (purple), where every node represents a neuron and each link represents a functional connection between neurons. The overlapping nodes indicate shared participation. Pie chart summarizes neuron proportions. j Average calcium activity of R-W-specific neurons and inactive neurons during R-W transition (left) and EMs (right). Lines represent the mean across neurons, and shaded areas indicate 95% CIs. The dashed line indicates the time point of transition or onset of EMs. k Average calcium activity of EMs-specific neurons and inactive neurons in response to EMs (left) and R-W transition (right). Lines represent the mean across neurons, and shaded areas indicate 95% CIs. The dashed line indicates the onset of EMs or the time point of transition.

At the population level, we calculated trial-averaged traces aligned to the onset of EMs. nIII^ChAT neurons exhibited an immediate response after the onset of EMs across sleep-wake states (Supplementary Fig. 15a, b). We also investigated the activity of overall nIII^ChAT neurons throughout the different vigilance states, and found that the nIII^ChAT neurons displayed high activity during REMs, compared to that in wake and NREMs (Supplementary Fig. 15c). Further analysis demonstrated that averaged calcium signals progressively elevated preceding the EMs-free R-W transition and declined after entering wakefulness (Supplementary Fig. 15d, e). This observation is consistent with our fiber photometry results (Fig. 1j).

To identify the correlation of nIII^ChAT neuron activities to either R-W transition or EMs, we computed two correlation coefficient matrices: one is between the calcium signals of all nIII^ChAT neurons and the EMs events (Fig. 4g, left); the other is between the calcium signals of all nIII^ChAT neurons and R-W transitions (Fig. 4h, left). Based on these two matrices, the dendrogram was constructed using unsupervised hierarchical clustering, employing Ward’s linkage to calculate the distance between each pair of clusters. Subsequently, the order corresponding to the peak value of the contrast index among the clusters is defined as the optimal cluster order (Fig. 4g, h, upper right, see Methods for details). Neurons within the clusters that met the threshold (correlations exceeding the 95% confidence level) were considered a neuronal ensemble and visualized in the neuronal network graph. The neuronal ensembles were identified as “R-W active” (Fig.4g, lower right) or “EMs active” (Fig.4h, lower right) subpopulations. Neurons not involved in these processes were categorized as “ R-W inactive” or “EMs inactive” (Supplementary Fig. 15f).

To evaluate network connectivity strength, we calculated the closeness centrality, measured by the shortest path distance of each neuron to its connected neurons. Our analysis revealed that R-W active and EMs active neurons exhibited higher connectivity compared to their inactive counterparts (Supplementary Fig. 15g). Furthermore, R-W active neurons demonstrated significantly stronger connections than the EMs active neurons (Supplementary Fig. 15g). Among the 42 functionally identified neurons, R-W specific and EMs specific neurons accounted for 40.1% (17/42) and 33.3% (14/42) of the population, respectively (Fig. 4i). For the R-W specific neurons, we observed elevated neuronal activity at the EMs-free R-W transition (Fig. 4j left and Supplementary Fig. 15i, j). However, no responses were detected when aligned with the occurrence of EMs (Fig. 4j right and Supplementary Fig. 15h). In contrast, EMs-specific neurons exhibited notably elevated GCaMP6s signals during EMs, but not during R-W transition (Fig. 4k and Supplementary Fig. 15k, l). We also examined the activity of R-W active neurons across three other types of vigilance state transitions and found these neurons did not alter their activity during transitions from W-NR, NR-R, and NR-W (Supplementary Fig. 16a–c). Additionally, approximately 26.2% (11/42) of nIII^ChAT neurons were overlapping neurons, existing in both R-W active and EMs active ensembles (Fig. 4i). Those overlapping nIII^ChAT neurons show increased activity in coherence with the elevated EMs during R-W transition (Supplementary Fig. 16d, e).

In summary, we identified a distinct subpopulation of R-W active cholinergic neurons within the nIII. These neurons exhibited activation preceding the R-W transition. Their dense interconnectivity may contribute to intrinsically driven reverberatory activity, thereby facilitating the initiation of the R-W transition.

nIII^ChAT neurons control the R-W transition through the vlPAG pathway

To further investigate the neural circuit mechanisms by which the nIII regulates REMs, we conducted anterograde tracing to identify its downstream targets. AAV2/8-hSyn-Flex-mWGA-mCherry-WPRE-pA (mWmC) virus was a genetically encoded monosynaptic anterograde tracer developed by Tsai et al.²⁹. The anterograde and trans-synaptic labeling capabilities of mWmC have been validated through striatal injections, which resulted in robust mCherry-labeled somas in the substantia nigra pars reticulata (SNr)²⁹. We injected the mWmC virus into the nIII in ChAT-Cre mice to identify direct postsynaptic targets of nIII^ChAT neurons (Fig. 5a). We observed distinct somatic mCherry signals in several downstream regions, including the ventrolateral periaqueductal gray (vlPAG), ventral tegmental area (VTA), median raphe nucleus (MnR), locus coeruleus (LC), and medial parabrachial nucleus (MPB) (Fig. 5b). By injecting AAV2/9-hEF1α-DIO-hChR2(H134R)-GFP-WPRE-pA into the nIII of ChAT-Cre mice, we observed clear axonal projections within vlPAG, LC, MPB and LC along the rostral-caudal axis in coronal sections (Supplementary Fig. 17).

Fig. 5. vlPAG-projecting nIII^ChAT population suppresses REMs.

a Schematic of anterograde tracing of nIII^ChAT neurons to label the downstream nuclei. b Representative images showing mCherry-labeled somatic in downstream regions (vlPAG, VTA, MnR, LC, MPB). Scale bar, 50 µm. c Schematic of retrograde labeling from vlPAG and EOMs. d Left: images of vlPAG-projecting (GFP) and EOMs-projecting (CTB555) nIII^ChAT neurons. Right: quantification of projection proportions (n = 4) (n = 4 mice, CTB555 = 33.33 ± 6.583%, GFP = 56.88 ± 11.98%, overlap = 10.40 ± 3.497%). e Schematic of viral injection (left) in nIII and optrode implantation (right) in vlPAG. f Example raster plot and spike waveform of laser-evoked unit. Each thin trace represents an individual spike, and the bold trace indicates the averaged waveform of spontaneous (gray) and laser-evoked (purple) spikes of an example unit. Dashed line, laser onset time. g Pulse-dependent firing of a ChR2-tagged unit during stimulation. h Proportions of activated, suppressed, and non-responsive units in vlPAG. i Schematic of viral injection in the nIII and fiber implantation in the vlPAG. j Representative EEG, EMG, and EMs traces during optical activation of the nIII^ChAT terminal within the vlPAG. k Left: percentage of wake, NREM, and REM before (pre-stim.), during (stim.), and after stimulation (n = 5). Right: quantification of state changes. Lines represent the mean across mice, and shaded areas indicate 95% CIs. Two-way ANOVA followed by Bonferroni post hoc analysis. **p < 0.01. Error bars represent SEM across mice. Data represent at least three biological replicates. All data and significance values for graph (k) are shown in Supplementary Table 1.

Previous studies suggest that vlPAG played a crucial role in gating REMs, while activating vlPAG GABAergic neurons in mice suppresses the initiation and maintenance of REMs^12,20. We therefore explored the spatial distribution of EOMs- and vlPAG-projecting cholinergic neurons within nIII. By injecting retrograde tracer (CTB555) and retroAAV2/8-hSyn-Flex-GFP-WPRE into the EOMs and vlPAG, respectively (Fig. 5c), we found that nIII^ChAT neurons were labeled either by CTB or GFP, but rarely co-labeled by both fluorophores in the same neuron (Fig. 5d). 33.3% of nIII^ChAT neurons project to the EOMs, whereas 56.9% send innervation to the vlPAG. To further validate the functional connectivity of nIII-vlPAG projection, we performed in vivo electrophysiological recordings combined with optogenetic stimulation (Fig. 5e). Upon photostimulation of nIII^ChAT terminals within the vlPAG, we observed reliable optogenetically evoked spiking in a single unit (Fig. 5f, g). Among the 61 well-isolated units recorded from 3 mice, 32.8% exhibited optogenetically activated responses, and 9.8% showed optogenetically suppressed responses to nIII^ChAT axon activation (Fig. 5h and Supplementary Fig. 18).

To examine whether nIII^ChAT neurons regulates REMs through nIII-vlPAG circuit, we stereotactically injected AAV2/9-hEF1α-DIO-hChR2(H134R)-EGFP-WPRE-pA into the nIII of ChAT-Cre mice and implanted an optic fiber into the vlPAG, along with an EEG/EMG/EMs polysomnography recording device (Fig. 5i). Using the same optical stimulation protocol as above (20 Hz, 2 mW), we found a reduction in REMs coupled with an elevation in wake state when activating the axon terminals of vlPAG-projecting nIII^ChAT neurons (Fig. 5j, k and Supplementary Movie 6).

We next sought to examine the effect of activation of vlPAG-projecting nIII^ChAT neurons on EMs. RetroAAV2/2-Plus-hEF1a-DIO-hChR2-mCherry-WPRE were injected into the vlPAG of ChAT-Cre mice, followed by implanting an optical fiber above the nIII (Supplementary Fig. 19a, b). Optogenetic stimulation at 480 nm was then applied in head-fixed mice. Laser-induced EMs were recorded by a camera, as previously described. As shown in Supplementary Fig. 19c, optogenetic activation of vlPAG-projecting nIII^ChAT neurons did not elicit significant EMs compared to the control group. These anatomical and functional experiments provide evidence that vlPAG-projecting nIII^ChAT neurons specifically regulate REMs rather than EMs.

Additionally, the LC located in the pons is also highly related to arousal, and activating the LC noradrenergic neurons effectively induces the transition of mice from sleep to wake^30,31. To determine whether the nIII neurons projecting to the vlPAG and LC constitute the same or distinct populations, we injected Cre-dependent retrograde viruses (retroAAV2/8-hSyn-Flex-GFP-WPRE-pA and retroAAV2/8-hSyn-Flex-mCherry-WPRE-pA) into the vlPAG and LC, respectively, in ChAT-Cre mice (Supplementary Fig. 20a). Co-localization analysis of GFP and mCherry signals in the nIII region revealed that a small subset of nIII^ChAT neurons (24.6% of GFP⁺ cells and 18.2% of mCherry⁺ cells) were double-labeled (Supplementary Fig. 20b, c), indicating partial overlap between the two projection populations.

To investigate the functional role of the nIII-LC pathway in sleep regulation, we implanted an optical fiber above the LC in ChR2-tagged ChAT-Cre mice and optogenetically stimulated LC-projecting nIII^ChAT axons (Supplementary Fig. 20d, e). In contrast to the REM-suppressing effect observed with nIII-vlPAG pathway activation, activation of nIII-LC projections resulted in a slight increase in REM probability (Supplementary Fig. 20f, g). Using the same approach, we also performed optogenetic activation of nIII^ChAT projections to the VTA and MPB (Supplementary Fig. 20h, j). However, activation of these pathways did not produce significant changes to sleep architecture, suggesting that the nIII^ChAT-VTA and nIII^ChAT-MPB circuits are not likely to be directly involved in sleep regulation (Supplementary Fig. 20i, k).

NP acts as a shared upstream regulator of cholinergic subpopulations in the nIII nucleus

To label long-range inputs to nIII, we injected AAV2/2-Retro-hSyn-hChR2(H134R)-EYFP into the nIII (Supplementary Fig. 21a). GFP-labeled neurons located in several brain regions, including the nucleus papilio (NP), the lateral habenular nucleus (LHb), lateral (dentate) cerebellar nucleus (Lat), interstitial nucleus of medial longitudinal fasciculus (Ri), lateral hypothalamic area (LH), root of abducens nerve (6 N), medial vestibular nucleus, parvicellular part (MVePC), medial (fastigial) cerebellar nucleus (Med) and interposed cerebellar nucleus, posterior part (IntP) (Supplementary Fig. 21b–k). Among these brain regions, NP is known for its role in EMs regulation during REMs⁷. However, there is currently no direct evidence showing that NP innervates cholinergic neurons within the nIII, nor have any studies reported involvement of the NP-nIII pathway in REMs regulation.

We first validated the NP-nIII pathway by injecting an anterograde trans-synaptic AAV1/2-hSyn-Cre-WPRE into the NP and CTB488 into EOMs of Ai14 mice, and found cholinergic neuronal soma intermingled with tdTomato-labeled soma in the ipsilateral nIII area (Fig. 6a, b and Supplementary Fig. 22a, b). There is also a subset of CTB488⁺/tdTomato⁺ co-labeled nIII^ChAT neurons (Supplementary Fig. 22b, c), indicating that the NP neurons innervate both EOMs-projecting and non-EOMs-projecting nIII^ChAT neurons. Additionally, we employed a rabies virus (RV)-mediated trans-synaptic retrograde tracing approach, which has been shown to efficiently label monosynaptic inputs to selectively infected starter cells³². To specifically label nIII^ChAT neurons, we first injected two Cre-dependent helper viruses (rAAV-EF1α-DIO-EGFP-T2A-TVA and rAAV-EF1α-DIO-oRVG(19G)) into the nIII in ChAT-Cre mice (Fig. 6c). After 2 weeks, glycoprotein (G) gene-deleted RVs expressing mCherry (RV-EnVA-G-mCherry) were injected into the same site. The starter neurons, which are shown in yellow, were characterized by the expression of both the helper virus and RV in cholinergic neurons (Fig. 6c). Immunofluorescence staining revealed robust mCherry-positive somata in the NP (Fig. 6d). These results provided direct anatomical evidence of a synaptic connection from the NP to nIII^ChAT neurons.

Fig. 6. NP mediates nIII cholinergic activity to modulate EMs and REMs.

a Schematic of anterograde tracing downstream region. Right: Confocal image of viral expression in the NP two weeks post-injection. b Anterograde trans-synaptic labeling of NP-projecting nIII^ChAT neurons. c Schematic of monosynaptic rabies tracing (left) and confocal image (right) of starter nIII^ChAT neurons. d Confocal image of NP neurons labeled by RV-mediated monosynaptic retrograde tracing from the nIII. e Schematic of viral injection in NP and optrode implantation in nIII. f Example raster plot (left) and waveform (right) of laser-evoked spikes. Each thin trace represents an individual spike, and the bold trace indicates the averaged waveform of spontaneous (gray) and laser-evoked (purple) spikes of an example unit. g Example raster plot of laser-inhibited spikes. h Proportions of activated, suppressed, and non-responsive units in nIII. i Schematic of viral injection and fiber implantation. j Representative image showing optic implantation trace within NP. k Eye position (left) and averaged EMs traces (right) during optogenetic activation. Lines represent the mean across mice, and shaded areas indicate 95% CIs. l Quantification of EM amplitude evoked by optogenetic activation of nIII-projecting neurons in NP (n = 3). Two-sided paired t-test. *p < 0.05. m Stimulation paradigm for 20 Hz. n Representative EEG, EMG, and EMs during stimulation of NP terminals in nIII. o Percentage of wake, NREM, and REM before, during, and after stimulation. Lines represent the mean across mice, and shaded areas indicate 95% CIs. The gray shielding indicates the time window of optical stimulation. p Quantification of state probabilities before (pre-stim.) and during (pre-stim.) laser stimulation (n = 5). Two-way ANOVA followed by Bonferroni post hoc analysis. ***p < 0.001. Error bars represent SEM across mice. Data represent at least three biological replicates. All data and significance values for graphs (l and p) are shown in Supplementary Table 1.

In addition to RV-mediated anatomical tracing, we also measured functional connections between NP and nIII^ChAT neurons using in vivo optogenetics-assisted electrophysiological recordings. We injected AAV2/9-hSyn-ChrimsonR-tdTomato-WPRE-pA into the NP and implanted an optrode in the nIII (Fig. 6e and Supplementary Fig. 23a, b). Optogenetic activation of NP terminals within the nIII induced a pulse-dependent increase or decrease in firing rate of single-unit (Fig. 6f, g and Supplementary Fig. 23c). Among 40 well-identified nIII units recorded from 3 mice, 17.9% exhibited optogenetically activated responses (Supplementary Fig. 23d), and 15.4% showed optogenetically suppressed responses (Fig. 6h and Supplementary Fig. 23e), demonstrating that nIII received both excitatory and inhibitory synaptic inputs from the NP.

As part of the dorsal paragigantocellular nucleus (DPGi), we wonder whether NP participates in nIII-mediated sleep regulation. To investigate this possibility, we injected AAV2/2Retro-hSyn-Cre-WPRE into the nIII and AAV2/9-Flex-ChrimsonR-tdTomato-WPRE into the NP of C57 mice (Fig. 6i, j). Upon optogenetic activation of nIII-projecting NP neurons, we observed robust T-N directional EMs (Fig. 6k, l). Furthermore, when 20 Hz stimulation was applied for two min per train across multiple trains during polysomnographic recordings, there was a significant increase in the fraction of REMs accompanied by a reduction in NREMs (Fig. 6m–p).

Distinct neuronal populations of the nucleus papilio innervating nIII^ChAT neurons

Lodged at the dorsomedial boundary of the dorsal paragigantocellular nucleus (DPGi), the NP comprises a heterogeneous population of neurons, including inhibitory GABAergic neurons, excitatory glutamatergic neurons, and a substantial number of neuropeptidergic neurons³³. To further identify the neuron types upstream of nIII^ChAT neurons contributing to motor behavior and sleep regulation, we employed a rabies virus (RV)-mediated trans-synaptic retrograde tracing approach. To specifically label NP neurons projecting to EOMs-innervating nIII neurons and to vlPAG-projecting nIII neurons, we first injected AAV2/2-Retro-hSyn-Cre-WPRE-pA into either the extraocular muscles or the vlPAG of C57 mice (Fig. 7a, e). Two Cre-dependent helper viruses, rAAV-EF1α-DIO-EGFP-T2A-TVA and rAAV-EF1α-DIO-oRVG(19G), were unilaterally injected into the nIII (Fig. 7a, e). After 2 weeks, RV-EnVA-△G-mCherry was injected into the nIII, as described in Fig. 6c. Using this strategy, the RVs can only infect TVA-expressing cells and require RG expression to spread retrogradely into presynaptic cells. The starter neurons, which are shown in yellow, were limited to the nIII region and characterized by the expression of both the helper virus and RV in cholinergic neurons (Fig. 7b, f). We mapped the afferent neurons in the NP by imaging 3 brain slices from each of four mice. Immunofluorescence and RNAscope assays show that 93.3% of the monosynaptic inputs from NP to EOMs-projecting nIII^ChAT neurons expressed the glutamatergic marker Vglut2 mRNA, while scarcely expressing the GABAergic marker Vgat mRNA (Fig. 7c, d). On the other hand, 88.3% of NP neurons innervating vlPAG-projecting nIII^ChAT neurons expressed the GABAergic marker Vgat, but not the glutamatergic marker Vglut2 mRNA (Fig. 7g, h). No co-expression of Vgat and Vglut2 was observed in NP (Fig. 7d–h). We also examined neurons expressing ChAT and Cart; none of these neurons exhibited detectable co-expression with mCherry (Supplementary Fig. 24). This finding suggests that the functional heterogeneity of nIII^ChAT neurons most likely arises from differences in the gene profiles of upstream NP neurons.

Fig. 7. nIII^ChAT-projecting NP neurons express mutually exclusive marker genes.

a Schematic diagram of a strategy to label the EOM-projecting NP neuron. After 3 weeks of AAV2/2-retro-hSyn-Cre expression in the extraocular muscles (EOMs), two helper (rAAV-EF1α-DIO-EGFP-T2A-TVA and rAAV-EF1α-DIO-oRVG(19G)) were injected into the nIII in C57 mice, followed by injection of modified rabies virus (RV) expressing mCherry (RV-EnVA-G-mCherry). b Confocal image showing a coronal section of the nIII 10 days post-RV injection. The dashed square region in the nIII is magnified below. EOMs-projecting nIII^ChAT neurons appear yellow due to co-expression of GFP and mCherry from helper and RV vectors, respectively. c Co-expression of mCherry-labeled EOMs-projecting NP neurons with the glutamatergic marker Vglut2 mRNA. d Percentages of mCherry-labeled neurons in the NP region that express Vglut2 mRNA, Vgat mRNA and Vglut2/Vgat overlap. Each dot indicates data from one mouse (n = 3 mice). Error bars represent SEM across mice. e Schematic diagram of strategy to label the vlPAG-projecting NP neuron. f Confocal image showing a coronal section of the nIII 10 days post-RV injection. The dashed square region in the nIII is magnified below. vlPAG-projecting nIII^ChAT neurons appear yellow due to co-expression of GFP and mCherry from helper and RV vectors, respectively. g Co-expression of mCherry-labeled vlPAGs-projecting NP neurons with the GABAergic marker Vgat mRNA. h Percentages of mCherry-labeled neurons in the NP region that express Vglut2 mRNA, Vgat mRNA, and Vglut2/Vgat overlap. Each dot indicates data from one mouse (n = 4 mice). Error bars represent SEM across mice. Data represent at least three biological replicates.

Discussion

Considerable studies have demonstrated that cholinergic neurons play a modulatory role in REMs control. Cholinergic neurons in the laterodorsal/pedunculopontine tegmental nuclei (LDT/PPT) function as REM-on neurons and form excitatory connections with glutamatergic neurons in the pontine reticular formation (PRF) to promote REMs^34,35. Systemic administration of cholinergic antagonists, as well as brainstem microinfusions, suppresses REMs³⁶. Conversely, the activation of cholinergic neurons in the basal forebrain has been shown to suppress both REMs and NREMs, thereby promoting wakefulness^37–39. Dysfunction in the cholinergic system has been linked to various sleep disorders, including narcolepsy, cataplexy, sleep paralysis, and hallucinations^40–42. It is noteworthy that most motor neurons responsible for controlling skeletal muscles are cholinergic. While the nIII was well referred to as a cholinergic motoneuronal pool involved in controlling EMs⁴³, its critical role in non-motor functions has hitherto been underestimated. Our findings revealed cholinergic neurons in the nIII serving not only as motor neurons but also playing a critical role in sleep regulation. The functional diversity of the cholinergic system has attracted significant research attention, focusing on the detailed mechanisms underlying the interaction between the cholinergic system in sleep and the cholinergic motor system.

The nIII is located in close proximity to the EW nucleus, which also contains cholinergic neurons. The cholinergic population in the EW represents preganglionic parasympathetic neurons (EWpg), which are primarily confined to the rostral part of the EW and are cytoarchitecturally distant from the nIII⁴⁴. Previous studies by Yi et al. and Kozicz et al. demonstrated that cholinergic neurons in the EW and those in the nIII do not overlap in the coronal plane^18,44. This anatomical separation, combined with the limited lateral spread of signal collection in fiber photometry recordings (typically within a radius of ~200–400 µm below the fiber tip)^45,46, minimizes the possibility of signal contamination from the EW when targeting the nIII. Although the EW also contains non-cholinergic, centrally projecting neurons (EWcp) that have been implicated in the regulation of NREMs¹⁸, our use of a Cre-loxP strategy in ChAT-Cre mice ensured that these non-cholinergic EWcp neurons were excluded from our recordings. Furthermore, when we implanted the optical fiber above the nIII for optogenetic stimulation, we observed laser-evoked EMs responses but not pupillary constriction (Supplementary Fig. 8), further indicating that EW neurons were not recruited in these experiments.

For downstream targets of nIII, although anatomical tracing has delineated the topographic organization of projections controlling different EOMs²², we have a limited understanding of the codify of oculomotor because the activity profile of nIII^ChAT neurons during oculomotor behavior has not been specifically reported. We provided the calcium imaging evidence demonstrating that nIII^ChAT neuronal activity encodes EMs in different directions. Although several studies have emphasized the nIII’s role in mediating eye elevation and adduction¹⁰, others have proposed its involvement in vertical and torsional components of oculomotor function⁴⁷. Optogenetic activation and inhibition experiments in our study provide direct causal evidence that the nIII plays an antagonistic role in regulating EMs. Specifically, activation of nIII^ChAT neurons drives EMs from the temporal side toward the nasal side, whereas their inhibition induces EMs from the nasal side toward the dorsotemporal direction. These findings reinforce the notion that the EMs were precisely controlled by agonistic and antagonistic muscle pairs, which are innervated by SIF and MIF motoneurons within the nIII, facilitating coordinated EMs essential for visual tracking and stability⁴⁸.

Parallel to the pathway of nIII projecting to the EOMs, we found that 56.88 % of nIII^ChAT neurons project to the vlPAG nucleus, which has been known as a REM-off area that provides inhibitory gating of REMs^12,49,50. GABAergic neurons in the vlPAG were selectively suppressed during REMs, and activating those neurons induces a striking suppression of REMs with the increase of NREMs and wake^12,20,25. Complete lesions of the vlPAG significantly increase the amount of REMs⁵¹. The vlPAG innervates wake-promoting serotonergic neurons in the dorsal raphe and noradrenergic neurons in the locus coeruleus^12,52. Our findings laid out a circuit substrate in nIII, connecting to previously identified switching circuitry for REMs⁵¹. Given current methodological limitations, it remains challenging to determine whether these two pathways share the same neuronal substrate, warranting further exploration.

The arousal-action neuronal circuit²⁵ regulating sleep and those controlling autonomic/somatic motor functions, suggests that a primary function of sleep is to suppress motor activity^25,53. During REMs, Glutamatergic neurons in the sublaterodorsal tegmental nucleus (SLD)^53,54, GABAergic neurons in the lateral paragigantocellular nucleus (LPGi)^55,56, and substantia nigra pars reticulata (SNr) are key nuclei for suppressing locomotor activity and facilitating the genesis of sleep⁵⁷. Interestingly, the muscle twitches, which occur during REMs, represent retained forms of motor activity⁵⁸. Brooks et al. suggest that muscle twitches during REMs might trigger sensorimotor activity in preparation for wakefulness⁵⁸. Indeed, compared to awakening from NREMs, animals are better prepared to respond to environmental threats after awakening from REMs^59,60. Another form of retained motor activity is elevated EM frequency, observed in the late stage of REMs in humans⁶¹. In this study, we confirmed the increased EM frequency towards REMs termination (Fig. 1h) and identified two distinct subpopulations of nIII^ChAT neurons in rodents. EMs-active neurons encode the modality of oculomotor behavior during REMs, whereas R-W active neurons exhibit elevated activity aligned with the progression toward R-W transition. Our study suggests that nIII functions as a crucial node, subserving the autonomic coordination between brain states and the oculomotor system during REMs. We also observed the presence of densely and retrogradely-labeled neurons in the sensory cortex upstream of nIII, particularly in the visual and auditory cortices (Supplementary Fig. 21). It’s reasonable to infer that the role of nIII in preparing the animal for receiving sensory input for wakefulness⁶², which also supports the notion that the threshold for wake during REMs is lower compared to that during NREMs^63,64. The elevated EMs at the end of REMs likely serve to maintain perceptual sensitivity to external stimuli during REMs and enable a rapid fight-or-flight response to potential danger signals.

Situated at the dorsomedial boundary of DPGi, the NP contains a heterogeneous population of neurons, including inhibitory GABAergic neurons, excitatory glutamatergic neurons, and neuropeptidergic neurons³³. Our study demonstrated that EOMs-projecting nIII^ChAT neurons and vlPAG-projecting nIII^ChAT neurons are innervated by glutamatergic and GABAergic neurons from the NP, respectively (Fig. 7). In vivo electrophysiological recordings of the NP-to-nIII functional pathway revealed that optogenetic activation of NP terminals in the nIII elicited a comparable proportion of significantly excited and inhibited neurons within the nIII. The functional divergence of nIII^ChAT subpopulations is likely orchestrated by distinct NP neuronal subtypes. This is consistent with previous reports showing that excitatory glutamatergic afferents from the NP innervate the nIII⁷. Interestingly, DPGi, as the surrounding of NP, also harbors a distinct pool of inhibitory GABAergic neurons, along with glycinergic and cholinergic neuronal populations^49,65. Inhibition of the DPGi selectively abolishes REMs⁶⁶, and activation of GABAergic neurons within the DPGi suppresses wake-promoting neurons, thereby promoting the initiation of REMs⁶⁷. In our study, optogenetic activation of nIII-projecting NP neurons also promoted REMs, analogous to the effect observed with photoinhibition of nIII^ChAT neurons (Fig. 6). These findings raise the possibility that the GABAergic neurons in NP represent a subpopulation of the inhibitory neurons within the DPGi. The inhibitory interactions between NP (REM-on) and the nIII (REM-off) support the flip-flop switch model proposed by Jun Lu et al.⁵¹, in which the periodic occurrence of REMs was generated by the dynamic reciprocal interactions between GABAergic REM-promoting and REM-antagonizing neurons^51,68,69.

In summary, our study testified to an emerging role for nIII^ChAT neurons in regulating REMs transition beyond their classical function in oculomotor control. The functional heterogeneity of nIII is realized through two parallel neural circuits. These findings provided mechanistic insights into the role of the nIII in sensorimotor and sleep controls, and contribute to our understanding of how EMs occur during the late REM period to prepare for brain state switching. Meanwhile, the nIII can be a novel therapeutic target for REMs behavior disorder^70,71 by fragmentizing the ongoing REMs.

Methods

Animals

ChAT-Cre transgenic mice (B6;129S6-Chat^tm2(cre)Lowl/J; Jax Stock#: 006410; RRID: IMSR_JAX:006410), and Ai14 reporter mice (B6.Cg-Gt(ROSA)26Sor^{tm14(CAG-tdTomato)Hze}/J; Jax Stock#: 007914; RRID: IMSR_JAX:007914) were purchased from Jackson Laboratory. Both lines were subsequently backcrossed onto a C57BL/6 background for further breeding, using wild-type C57BL/6 mice obtained from SLAC Laboratory Animal Co., Ltd. (Shanghai, China). Both male and female mice were used throughout the whole study. All mice were 8–12 weeks old, weighing between 26 and 30 g, at the start of experimental procedures. All mice were group-housed with ad libitum access to food and water, maintained at a constant temperature of 23 ± 2 °C, humidity of 50 ± 10%, on a 12 h light/dark cycle (light on at 8:30 am). After surgery, mice were allowed to recover from anaesthesia under a warming lamp, and were subsequently relocated to a new residence room for a minimum of 2 weeks to ensure full recovery before commencing sleep recording, calcium imaging, or fiber photometry experiments. All animal experimental procedures were approved by the Animal Experiment and Use Committee of the Shanghai Medical School of Fudan University.

Virus preparation

The following recombinant adeno-associated viral vectors (AAVs) were packaged and purchased from Shanghai Taitool Bioscience Co., Ltd.: AAV2/9-hEF1α-DIO-hChR2(H134R)-EYFP, AAV2/2-Retro-hSyn-hChR2(H134R)-EYFP, AAV2/2Retro-Plus-hEF1a-DIO-hChR2(H134R)-mCherry-WPRE-pA, AAV2/9-hSyn-DIO-stGtACR2-mCherry-WPRE-pA, AAV2/9-hSyn-mCherry-WPRE, AAV2/2Retro Plus-CAG-DIO-taCaspase3-TEVp-WPRE-pA, AAV2/9-CAG-DIO-taCaspase3-WPRE, AAV2/9-hSyn-FLEX-GCaMP6s-WPRE, pAAV-EF1α-DIO-GTACR1-GFP, AAV1/2-hSyn-cre-tdTomato, AAV2/8-hSyn-Flex-mWGA-mCherry-WPRE-pA, AAV2/9-hSyn-Flex-ChrimsonR-tdTomato-WPRE-pA, AAV2/9-hSyn-ChrimsonR-tdTomato-WPRE-pA, and AAV2/2Retro-hSyn-Cre-WPRE-pA. The retrograde tracer rabies virus RV-EnVA-ΔG-mCherry, together with the helper viruses rAAV-EF1α-DIO-EGFP-T2A-TVA and rAAV-EF1α-DIO-oRVG(19G), were purchased from Brain Case (Shenzhen) Biotechnology Co., Ltd. The titer was estimated to be ~10¹² to ~10¹³ genome copies ml⁻¹ and diluted to ~10¹² with 5% glycerin and 95% PBS prior to administration.

Immunohistochemistry

Mice were euthanised under deep anesthesia induced by a 2.5% Avertin/saline solution and subsequently subjected to transcardial perfusion with 0.1 M PBS, followed by fixation with 4% ice-cold paraformaldehyde (w/v) in PBS. The dissected brain tissues were post-fixed in 4% paraformaldehyde at 4 °C overnight and then immersed into 30% sucrose in 1× PBS solution for approximately 2 days at 4 °C. Brains were then embedded and frozen before sectioning at 30 μm slices using a freezing microtome (Leica CM520). The coronal sections were subjected to immunofluorescent staining according to the following protocol. Briefly, the tissues were permeabilized with PBST (0.3% Triton X-100 in PBS) for 30 min, incubated with blocking solution followed by primary antibody incubation overnight at 4 °C, and secondary antibody incubation for 2 h. Primary antibodies utilized were goat anti-ChAT (Millipore, ab144P, 1:100), rabbit anti-NeuN (Millipore, Cat# ABN78, 1:1000), rabbit rabbit anti-cleaved caspase-3 (Cell Signaling Technology, Cat# 9661, 1:400), chicken anti-GFP (Abcam, Cat# ab13970, 1:800), rabbit anti-Dsred (Takara Bio, Cat# 632496, 1:1000), rabbit anti-cFos (Millipore, Cat# ABE457, 1:1000), rabbit anti-mCherry (Sicgen, Cat# AB0081-300, 1:400), and rabbit anti-CART (Phoenix Pharmaceuticals, Cat# H-003-62, 1:1000). Secondary antibodies were purchased from Jackson ImmunoResearch Labs and included Alexa Fluor 647 donkey anti-rabbit IgG (Cat# 711-605-152, 1:800), Alexa Fluor 594 donkey anti-rabbit IgG (Cat# 711-585-152, 1:800), Alexa Fluor 488 donkey anti-rabbit IgG (Cat# 711-545-152, 1:800), Alexa Fluor 488 donkey anti-chicken IgG (Cat# 703-545-155, 1:800), Alexa Fluor 647 donkey anti-goat IgG (Cat# 705-605-003, 1:800), Alexa Fluor 594 donkey anti-goat IgG (Cat# 705-585-003, 1:800), Alexa Fluor 488 donkey anti-goat IgG (Cat# 705-545-003, 1:800). DAPI staining was performed prior to mounting. Fluorescence images were ultimately captured using a 20× objective on a microscope (VS 120, Olympus, Tokyo, Japan) and a 20× confocal microscope (Nikon, NI-SJ; Olympus, FV-3000, Tokyo, Japan). For neuronal counting, somas of neurons were quantified semi-automatically using ImageJ (Fiji, National Institutes of Health), followed by manual screening.

Fluorescence in situ hybridization (FISH)

FISH was conducted using the RNAscope Multiplex Fluorescent Reagent Kit v2 (Advanced Cell Diagnostics, Newark, CA, USA) following the manufacturer’s protocol. Brain sections (15–20 μm) were first incubated with hydrogen peroxide, followed by target retrieval and protease treatment (Protease Plus). Sections were then hybridized for 2 h with probes targeting Vgat (Cat# 319191) and Vglut2 (Cat# 319171), both from ACD. Signal amplification was achieved through sequential application of AMP1, AMP2, and AMP3 reagents (15–30 min each). Fluorescent detection was performed using horseradish peroxidase (HRP)-conjugated reagents (HRP-2 or HRP-3, 15 min) and Opal fluorophores (Opal 520 and 690, 1:750 dilutions; PerkinElmer, Waltham, MA, USA) for 30 min. HRP blocker was applied for 15 min following each fluorophore reaction to prevent cross-channel interference. All hybridization and incubation steps were carried out at 40 °C in a humidified chamber within a hybridization oven. Imaging of labeled sections was performed using a confocal laser scanning microscope (Olympus FV3000, Tokyo, Japan) equipped with a 20× objective.

Sleep deprivation and sleep recovery experiment

The mice were categorized into two groups: the sleep deprivation (SD) and the sleep rebound (SR) group. The small-platform-over-water method and a closed-loop sleep interference system were used. For sleep deprivation using the small-platform-over-water method, mice were gently placed on a handcrafted small-round platform (diameter: 2.5 cm) situated within a water tank (1 cm under the platform) at the start of the light period (7 am). For sleep deprivation using a closed-loop sleep interference system, the mice were implanted with an EEG and EMG electrode. The mice were habituated to the polysomagrahpic recording chamber at least 2 days prior to sleep deprivation. On the third day, EEG, EMG, and EM signals were recorded in real time and segmented into 4 s epochs. The segmented signals were subsequently transmitted to MLS-Net, an autonomous online sleep classifier, for analysis and determination of vigilance states. Upon detection of a REMs epoch, a TTL signal was sent via Arduino to trigger a rotarod at the bottom of the chamber, thereby interrupting REMs. After 8 h of sleep deprivation, mice in the SR group were transferred to their home cages for sleep recovery with polysomnography recording concurrently. Once the mice had slept for at least 3 h, they were immediately anesthetized and sacrificed for further c-Fos immunostaining and histological procedures. In contrast, mice in the SD group were directly sacrificed for immunostaining without undergoing a recovery period.

Optokinetic reflex (OKR)

The optokinetic response (OKR) in mice was elicited by presenting horizontally-drifting gratings within a virtual reality system (PhenoSys, Germany). The system comprised a floating trackball, facilitating unrestricted rotation, and surrounded by five screens. The head-fixed mouse was guided to view the drifting grating on the floating trackball, and their resultant EMs were recorded by a high-speed infrared (IR) camera (JAI, Denmark, with 512 × 496 image resolution and 62 Hz sampling frequency), in conjunction with an IR emitter. Grating patterns (contrast level, 100%, 0.3 cycle per degree, 5°/s speed, clockwise or counterclockwise motion) were generated using the PsychoPy library in Python, and each session lasted for 3 min. To analyze the activity of nIII cholinergic neurons in response to EMs in different directions, mice were pre-implanted with optic fiber cannulas, and synchronous nIII^ChAT neuronal activities were recorded via a fiber photometry system. Laser-evoked EMs were recorded using the same methods as those described above, with the exception of the drifting gratings.

Surgeries procedure

General surgical procedures and viral delivery

Mice were anesthetized with 1.5% isoflurane and positioned in the stereotaxic apparatus (RWD, Shenzhen, China) for central injections. The skin was carefully incised to expose the skull, and the overlying connective tissue was removed. A burr hole was made near the posterior fontanelle on the skull with a cranial drill. A micropipette (15–20 µm tip) filled with 0.1–0.2 µl virus solution was injected unilaterally into the nIII (anteroposterior (AP) − 3.8 mm, mediolateral (ML) ± 0.2 mm, and dorsoventral (DV) − 3.1 mm) for calcium imaging, fiber photometry recording, and optogenetic manipulation. Virus was injected bilaterally into the nIII for upstream nucleus tracing and nIII^ChAT-vlPAG pathway activation. For functional identification of the NP-nIII pathway, viruses were injected into the NP at stereotaxic coordinates AP − 6.15 mm, ML 0.1 mm, DV − 4.2 mm. For retrograde tracing from vlPAG and LC to the nIII, viral injections were targeted to the LC (AP − 5.37 mm, ML ± 0.1 mm, DV − 3.49 mm) and the vlPAG (AP − 3.6 mm, ML ± 0.4 mm, DV − 2.3 mm). A total of 75 nL (for unilateral) or 150 nL (for bilateral) AAV virus was delivered to each site using a Nanoject II (Drummond Scientific). To selectively activate the non-EOMs-projecting nIII^ChAT neurons, we injected retroAAV-hSyn-Cre-WPRE into the extraocular muscles of C57BL/6 mice, followed by AAV2/9-hEF1α-DIO-ChR2(H134R)-GFP-WPRE injection into the nIII. To activate the EOMs-projecting neurons, we injected retroAAV-hSyn-Cre-WPRE-pA into the EOMs of C57 mice and subsequently injected AAV2/9-hEF1α-DIO-ChR2(H134R)-GFP-WPRE into the nIII to selectively express ChR2 in EOMs-projecting nIII^ChAT neurons. For monosynaptic rabies tracing, 100 nL (for bilateral) virus was delivered to each side. AAV-DIO-TVA and AAV-DIO-RG were mixed at a 1:2 ratio (TVA:RG) prior to injection. The needle was left in place for at least 15 min at each site to facilitate diffusion. The respiratory status of the mice during the procedure was monitored, and anesthesia level was adjusted to maintain stable breathing. After the injection, the snipped scalp was sutured with surgical sutures, followed by disinfecting with iodine, and the mice were recovered for at least 2 weeks for further experiments. For neuronal retrograde tracing from EOMs to the nIII, CTB488 or CTB555 was injected into four sites (dorsal, ventral, temporal, and nasal) around the EOMs using glass micropipettes inserted across the conjunctiva. The injection volume was limited to 0.75–0.9 μL per site so that the six EOMs could be effectively infected. To isolate individual EOMs (EMs) innervating neurons in the nIII, a more elaborate operation was conducted to ensure individual neuromuscular junction was infected, as referred to in related articles²². To avoid CTB leakage and ensure infectious effectiveness, the injection volume was limited to no more than 0.08 μL. After CTB injection, animals were kept warm and gently massaged until they regained their righting reflex. They were then returned to a home cage for long-term recovery and observation.

Enucleation

For sleep deprivation experiments, bilateral enucleation of the eyeballs was carried out on wild-type mice aged 2–3 months. Prior to the surgical procedure, mice were deeply anesthetized and maintained with 1.5% isoflurane throughout the operation. The eyelids were meticulously incised, and the connective tissue and muscle enveloping the eyeball were carefully excised. After ensuring that the optic nerve was severed, the eyeball was extracted. Subsequently, the eyelids were repositioned and sutured. Enucleated animals were euthanized and sacrificed for brain sectioning after sleep deprivation or sleep rebound.

EEG and EMG implantation

To record the electroencephalogram (EEG), electromyogram (EMG), and EMs (EMs), we have developed a device that has been validated for effectively recording both EMs and vigilance states during optogenetic manipulation experiments²⁴. The device contained two stainless steel screws for EEG recording, with the leads from the screw electrodes placed respectively above the frontal (AP = 1 mm; ML = +1.5 mm) and parietal (AP = −2 mm; ML = −1.5 mm) lobes on the surface of the skull. EMG signals were recorded from two electrodes inserted into the neck musculature to record postural tone. EM signals were acquired based on the magnetic displacement sensor integrated in the device. Finally, the entire assembly was then coated with dental cement.

Optic fiber implantation (for fiber photometry and optogenetic experiments)

Two weeks after virus injection, optic fiber cannulas (200 μm in diameter, 1.25 mm O.D., 0.48 numerical aperture, Newdoon, Shanghai) were inserted into the target region for optogenetics or fiber photometry experiments, respectively. The insertion depth ensured that the tip of the cannula was positioned between 50 and 200 μm above the site of virus injection, which was achieved through a craniotomy hole made in the mouse skull corresponding to the region of interest. Once the cannula was inserted, a metal frame was affixed to the skull to act as a head-fixation holder. Dental cement was then applied to completely cover the exposed skull surface and securely anchor the optic fiber in place.

Optrode implantation (for electrophysiological recording)

The micro-drive optrode consisted of four tetrodes surrounding a central optical fiber (100 μm core diameter, 1.25 mm outer diameter, 0.22 numerical aperture). Each tetrode was constructed by twisting together four strands of nickel-titanium alloy wire, with all four tips exposed to form individual recording sites. The spacing between adjacent tetrodes was approximately 200 μm, and the distance from the tips of the tetrodes to the end of the optical fiber ranged from 200 to 500 μm. A screw fixed to the base of the electrode assembly allowed for adjustment of the electrode depth by turning the screw, controlling the relative distance between the electrode interface and the base. For implantation, a small unilateral craniotomy was performed above the target brain region in adult mice (8 weeks old) following viral injection. The optrode was initially lowered to a position approximately 500 μm above the target site. Four additional holes were drilled into the skull to accommodate stainless steel screws for EEG and grounding purposes. Two silver-insulated EEG wires were soldered to screws positioned in the frontal cortex, while the ground and reference electrodes were attached to screws implanted in the cerebellum. Dental cement was then applied to secure the EEG/EMG screws and optrode, and to seal the exposed skull surface. Mice were allowed to recover for at least 2–3 weeks before subsequent behavioral or electrophysiological recordings.

GRIN lens implantation (for calcium imaging)

Two weeks after the injection of the GCaMP6s virus, the mice underwent implantation of a gradient refractive index (GRIN) lens (400 μm in diameter, 6 mm length, Inscopix), which was firmly mounted onto a stereotaxic holder also provided by Inscopix. The GRIN lens was connected to a micro-camera, which allowed real-time monitoring of the fluorescent signals as the lens was gradually descended into the brain parenchyma. Upon the apparent visualization of GCaMP6s fluorescence under illumination from the microscope’s LED, the lens was affixed to the skull using a bioadhesive silicone elastomer (Kwik-Sil™, World Precision Instruments Inc.) and dental cement. To prevent hemorrhage during lens implantation and avoid severing the sagittal sinus vessel located in the midline of the parietal lobe, an electrocoagulation pen was employed to cauterize the vessel. The microscope’s baseplate was securely fastened to the skull, guided by the microscope’s view field, using dental cement to facilitate subsequent attachment of the microscope to the head. In experiments necessitating the recording of EEG, EMG, and EMs, the GRIN lens needed to be implanted first, and then the baseplate and polysomnography recording device were affixed to the skull.

Polysomnography recordings and analysis (EEG/EMG/EM data acquisition, processing, and sleep-wake state classification)

After 2 days of habituation to the EEG/EMG/EMs headmount in sleep recording chambers, mice were subjected to EEG/EMG and EMs signal collection during sleep. Signals were digitized at 500 Hz and band-pass filtered (0.5–50 Hz filtration for EEG, 30–1 k Hz filtration for EMG and EMs signal). Sleep states were offline scored using SleepSign software (Version 2.0, Kissei Comtec, Nagano, Japan) followed by manual calibration if necessary. All sleep-wake scoring was semi-automatic based on spectral signatures of EEG and EMG waveforms in 4 s epochs. Sleep-wake states were classified according to previously reported criteria^23,24. Sleep-wake state classification is manually performed based on visual and spectral signatures of EEG and EMG signals. States were assigned across consecutive non-overlapping 4 s windows. We defined wakefulness as desynchronized, low-amplitude, and high-frequency EEG rhythms with elevated EMG activity. NREM sleep was defined as high-amplitude, low-frequency (0.5–4 Hz) EEG activity with lower tonic EMG tone, whereas REM sleep was defined as containing prominent theta (6–9 Hz) oscillations on EEG and minimal EMG activity.

To investigate the effect of optogenetic stimulation on sleep architecture, trials consisting of 6-min time windows, aligned with the onset of laser stimuli, were extracted and analyzed according to the previously established methodologies¹⁷. Percentage change was defined as the ratio of the percentage observed during stimulation to that during the baseline period. The power spectral density of EEG data before and after optical stimulation was estimated using Fast Fourier Transform (FFT) algorithms. Trials containing EEG data epochs with high-amplitude artifacts were discarded from the analysis. The total duration and episode numbers of wake, NREM, and REM were obtained from the SleepSign software after automatic state scoring. The probabilities of different brain states in nIII^ChAT mice after laser onset falling into REMs and NREMs were calculated across multiple trials. The REM latency was defined as the time from the onset of the optical stimulation to the occurrence of the initial wake episode during the stimulation period, and the latency index was defined as the ratio of REM latency to the total duration of the REM episode.

Fiber photometry recording and analysis

To monitor the real-time activity of nIII^ChAT neurons in response to EMs and sleep transition, we recorded the GCaMP6s fluorescence signals using a fiber photometry system (Thinker Tech Nanjing Biotech). During the recording session, the pre-implanted fiber was connected to the system via a 1.5-m-long fiber-optic patch cord. A 473 nm and a 405 nm laser beams were used for GCaMP6s excitation and isosbestic wavelength, respectively. The power of 473 nm imaging light was set at 30–40 μW, and the 405 nm light power was adjusted to approximately match the GCaMP6s fluorescence signals. The GCaMP6s fluorescence was converted to voltage signals within the system and digitalized at a sampling rate of 100 Hz. To synchronize the fiber photometry with EEG/EMG/EMs recordings, an external trigger carrying transistor-transistor logic (TTL) pulses generated by a custom Arduino board was delivered into the fiber photometry, the EEG/EMG/EMs recording system, and the 480 or 590 nm optical maser for optogenetic performance during sleep, respectively. Data acquisition and signal visualization were controlled by custom-written software. For in situ calcium signals recording and optogenetic stimulation of the terminal of NP in nIII, we directly combined a 590 nm laser beam with the 480 nm laser beam within the fiber photometry system to activate NP axon collaterals within the nIII and synchronously collect fluorescence signals in ChrimsonR-tagged ChAT-Cre mice.

The photometry data were processed using a custom-written Python script. Briefly, the filtered 488 nm signal was first detrended, followed by the calculation of the ΔF/F using the formula F − F0/F0 (where F0 represents the average of the entire trace). Subsequently, the fluorescence traces were then normalized (Z-scored ΔF/F). For the quantification of Ca²⁺ signals across different vigilance transitions, we calculated the average ΔF/F fluorescence of all sleep-wake active neurons and aligned it with the transition time point (from R-W, NREM-to-others, and wake-to-NREM). Other possible combinations were excluded, as rodents infrequently transition directly from wake to REM, which is characteristic of narcolepsy. Due to the variable duration of each state episode, we performed time normalization as previously reported^12,72 and then averaged the ΔF/F under the same type of state transition across trials (Figs. 1 and 4). Specifically, if the duration of a target episode exceeded 60 s, we extracted ΔF/F signals within a 60 s window before and after the transition point. For episodes shorter than 60 s, signals were extracted using the actual episode duration both preceding and following the transition. The extracted traces were then interpolated to generate 60 s time-normalized activity profiles. Finally, all trials were temporally aligned to the transition point and visualized as a heatmap (Supplementary Fig. 3). To analyze the relationship between the duration of REMs preceding its termination, the average calcium signal, and the probability of REMs transition episodes, the Z-scored ΔF/F values of the last epoch at the NREM-to-REM transition or the first epoch of the R-W transition were averaged across all extracted trials from all mice. The probability of nIII activity is defined as the ratio of the cumulative number of REM episodes longer than the current REM episode duration to the total number of REM episodes (Supplementary Fig. 3b, c).

Calcium imaging and analysis

Deep-brain calcium imaging was conducted using the Inscopix nVoke miniscope in GCaMP6s-tagged ChAT-Cre mice, conducted 2 weeks post-implantation of the GRIN lens. To ensure consistency of the field of view across different recording sessions, the microscopes were refocused as necessary. While minor changes in the field of view were inevitable due to long-term imaging sessions, all data were analyzed independently to explicitly align cell signals across imaging sessions. Images were collected at a rate of 16 Hz with an LED power of 1 mW/mm² and a gain of 6. A total of 5 mice were recorded, each with at least two imaging sessions. Each imaging session lasted for 1–2 h, and the calcium movies from all of the acquired sessions were motion corrected using default settings in the inscopix data processing software. Single-cell activity traces were obtained using the CNMF-E algorithm⁷³. Relative changes in fluorescence trace (ΔF/F) were subsequently used in further analyses. A custom Arduino board was used to synchronize the calcium imaging data with sleep recordings via external TTL signals. EEG/EMG/EMs recordings and calcium imaging were recorded synchronously, and the timestamps values were utilized to accurately align calcium tracing with the sleep-wake states of the mice. All mice underwent a 2-day miniscope adaptation prior to the start of the formal recording sessions.

To delineate the activity profiles of individual neurons in response to EMs across different sleep states, we computed trial-averaged activity using the z-scored ΔF/F fluorescence trace of each neuron across all considered trials, which were then aligned to the onset of EMs across different vigilance states, respectively. The baseline was defined as the time window 1 s preceding the onset of EMs, and the subsequent 4-s interval was designated as the response window. For the analysis of neuronal activity during sleep-wake transitions, we computed the average ΔF/F fluorescence across all neurons and aligned it to the transition time point, spanning 16 s before and after the transition. The use of time-normalized calcium signals facilitated alignment across the varying durations of vigilance states. To specifically assess the activity profile of REM active neurons during transitions, we extracted REM episodes characterized by the absence of EM events within 4-s epochs during REM stages. Additionally, the distribution of EMs during these transition episodes was visualized through raster plots and histograms with a bin size of 4 s.

To analyze the neuronal network, the Pearson correlation coefficient was calculated between each pair of identified neurons, yielding a correlation matrix for EMs and R-W transition, respectively. Based on the correlation coefficient matrix, we constructed a dendrogram to quantify the linkage between pairs of neurons and calculated a contrast index. The contrast index was defined as the ratio between the correlation coefficient values within (inner cluster) or without (outer cluster) the selected cluster order. The contrast index was implemented to determine the optimal number of subgroups for clustering⁷⁴. Specifically, each cluster order corresponds to a contrast index value, and the cluster order with the highest contrast index is identified as the optimal cluster order. Neurons within the cluster, as determined by the selected optimal cluster order, were analyzed using a threshold defined by the correlation coefficient at the 95% confidence level. Correlations between pairs of neurons exceeding this threshold were used to construct a functional network graph. In the functional network, each node represented a neuron, and each connection denoted significant coactivity between them (Fig. 4g, h). Neurons within the selected clusters in the network were fitted with an ellipse and named as “REM active” or “EMs active” neurons, respectively. Isolated neurons with no connections to any other neurons were excluded, and the remaining neurons were defined as “REM inactive” or “EMs inactive” neurons, respectively. Neurons that occupied both networks were defined as “overlapped” (Fig. 4i). The percentage of each subpopulation was presented using a Venn diagram (Fig. 4l). This allows a better visualization of the co-classification of nIII^ChAT neurons in encoding EMs and REM-wake transition information.

Optogenetic manipulation and analysis

To conduct optogenetic experiments, an optic fiber cable (1.5 m in length and 200 μm in diameter; ThorLabs) was affixed on one end to a pre-implanted optical ferrule and on the other end to either a 473 nm or 590 nm laser generator (BioLED Teolinc) via an FC/PC adapter. For optogenetic activation of nIII^ChAT neurons, a 480 nm blue laser at a frequency of 3.3 Hz or 20 Hz with a power of 2 mW was applied for 2 min per trial with intervals of 20–25 min. Each mouse underwent a total of 10–20 trials within a 24 h period. Stimulation trials were initiated remotely following the observation of stable sleep events, as determined through objective assessment of EEG and EMG waveforms. To examine the effect of activation of nIII^ChAT neurons on the REMs, optical stimulation was specifically delivered at least 16 s after the onset of REMs, with 5–10 trials conducted per mouse. The control group received 590 nm laser stimulation in ChR2-tagged ChAT-Cre mice. For optogenetic inhibition of nIII^ChAT neurons, a 590 nm laser at a frequency of 20 Hz with a power of 2 mW was applied for 2 min per trial with intervals of 15 min during 1:00 pm–6:00 pm at the light phase. Each mouse underwent a total of 30 trials within a 24 h period. To activate the nIII-vlPAG and nIII-LC pathways, the sleep states of mice were monitored in real-time, and laser stimulation was delivered at least 16 s after the mice entered NREMs or REMs. To activate nIII^ChAT neuron terminals within the VTA or MPB, 480 nm light pulses were delivered at 20 Hz. Each stimulation train lasted 2 min, and a total of 30 trains were applied with 15-min intervals. For optogenetic-assisted electrophysiological recordings, NP fibers within the nIII were stimulated using a 590-nm laser (2 mW, 40 pulses, 10 ms pulse width and 40 ms pulse interval) for 2 min. The laser frequency was controlled by a custom Arduino board with a preloaded script. Pulse signals were input to the acquisition system (Medusa, Bio-Signal Technologies, China) via a BNC cable for accurate timestamping of stimulation onset. For short-term optogenetic activation experiments, sleep recordings were performed at 2:00 pm–9:00 pm for each condition (3.3 Hz and 20 Hz) in every mouse. For long-term optogenetic activation experiments, continuous laser stimulation (20 Hz, 10 ms pulse duration and 40 ms pulse interval) was applied between 3:00 pm and 4:00 pm on the test day for each frequency, with sleep recordings starting at 2:00 pm and lasting for 3 h. Mice were connected to the recording device at least 2 days prior to the experiment to habituate the recording box and to prevent sleep interference from fiber cables during free moving.

To assess the effect of optic stimulation on sleep architecture in ChAT-Cre mice, the EEG/EMG/EMs traces per session were meticulously scored into specific states, and then each trial was extracted according to the time stamps. Note that trials with proportion of wake epochs exceeding 70% in the first 2 min after stimulation were discarded in the following analysis. The proportion of each state is defined as the ratio of trials containing the target state at the current time point to the total number of aligned trials. The parameters of vigilance states (number, duration, amount of time of wake, REM, and NREMs episodes) were referred to related works⁷⁵ and analyzed by custom-written Python scripts. For the probability analysis of wake, REM, and NREMs from trials when the laser stimulation fell exactly into REM or NREM epoch after offline sleep scoring.

Optogenetic-assisted electrophysiological recording and analysis

Micro-drive optrodes were purchased from Kudou (Suzhou) Brain-Computer Technology Co., Ltd. (China). Each optrode comprises 20 channels in total, which are soldered to a 32-pin connector (Omnetics, USA). Sixteen nickel-titanium alloy wire channels (20 μm internal diameter; impedance 250-900 kΩ) are arranged around an optical fiber for spike recording, while 4 additional silver wire electrodes are included for EEG and EMG recording. All wire joints were sealed with AB epoxy resin, and exposed nickel-titanium alloy wires were protected using polyethylene glycol 2000. Neuronal recordings were performed in a quiet room within a shielded enclosure starting from the 7th day after electrode implantation. Electrodes were connected to a multi-channel acquisition system (Apollo, Bio-Signal Technologies, China), and signals were digitized at a 3 kHz sampling rate. Spike signals and local field potentials (LFPs) were filtered online in the ranges of 300–7000 Hz and 0.7–200 Hz, respectively. The electrode was advanced daily by 60 μm increments until clear light-evoked neuronal spikes were recorded.

Data, including spike waveforms and timestamps, were saved as *.plx files using Plexon data acquisition software. Offline spike sorting was performed using Offline Sorter (Plexon, USA) based on the valley-seek method. Noisy or ambiguous waveforms were manually excluded in three-dimensional feature space. Sorted units and their timestamps were exported for further analysis using the PlexonIO Python package to generate raster plots and firing rate analyses. To identify ChR2-tagged neurons, both single-pulse (10 ms, 2 mW, 4 s interval) and high-frequency trains (20 Hz, 10 ms pulse width, 1 s duration, 2 mW, 40 repetitions) of optical stimulation were delivered. For recordings from ChR2-expressing nIII neurons, NP-nIII, and nIII-vlPAG neurons, units were classified as increased response if their average firing rate within 1 s post-stimulation exceeded the mean baseline activity by more than three standard deviations (mean + 3 SD). Units were classified as decreased response if the post-stimulation firing rate dropped below the threshold of mean − 3 SD of baseline activity.

To analyze nIII^ChAT neuron firing rates across sleep-wake transitions, we applied time normalization and neuronal activity visualization procedures analogous to those used in the fiber photometry analysis. To assess the temporal fidelity of ChR2-tagged nIII^ChAT neurons during prolonged optogenetic stimulation, we applied continuous optical stimulation for 1 h. The stimulation period was divided into four 15 min quartiles, and neuronal firing rates were quantified by averaging within each quartile.

Cell counting data analysis

All images for cell counting were acquired using a confocal microscope (Olympus FV3000, Tokyo, Japan) equipped with a 20× objective lens. Z-stack images were obtained with a step size of 1.5 μm across 7–8 layers, resulting in a total thickness of approximately 25 μm. The acquired Z-stacks were processed using maximum intensity projection and saved in TIFF format. Cell quantification was performed using the counting analysis tool in Adobe Photoshop (Adobe Systems Incorporated, San Jose, CA).

For immunostaining, three consecutive 30-μm-thick coronal sections were used per mouse; for FISH experiments, 20-μm-thick sections were used. For cellular ablation analysis, the proportion of ChAT-positive neurons co-labeled with NeuN was manually quantified in four mice. To assess projection patterns of nIII^ChAT neurons to the extraocular muscles (EOMs), we counted CTB555⁺ neurons, ChAT⁺ neurons, and their double-labeled counterparts on both the ipsilateral and contralateral sides relative to the injection site in six mice. For quantification of projection ratios from nIII^ChAT neurons to the vlPAG and LC, we calculated the percentage of double-labeled neurons (GFP and mCherry co-expressing) relative to either GFP⁺ or mCherry⁺ populations from three mice. To determine neurotransmitter identity, the proportion of Vglut2⁺ or Vgat⁺ cells among mCherry-labeled neurons was quantified. This was performed in three mice for NP-EOMs projections and in four mice for NP-vlPAG projections. Values were averaged across animals to obtain group-level estimates. All quantification was performed by experimenters who were blinded to experimental conditions and treatment groups.

Statistical analysis

No statistical methods were used to predetermine sample sizes. Instead, sample sizes were chosen based on the number of biological replicates required to achieve a reliable statistical analysis. To minimize experimenter bias and ensure technical reproducibility, immunostaining, optogenetic, and micro-endoscopic recording experiments were independently performed by at least two trained experimenters, with each condition replicated at least twice. All data were acquired and processed using instrument software without manual interference.

Data were presented as mean ± standard error of the mean. Statistical tests were analyzed with GraphPad Prism (GraphPad 8.0 Software). The two-tailed Student’s t test (paired or unpaired) was used for single value comparisons, while one-way ANOVA followed by post-hoc Tukey test analysis was employed to compare more than two groups. For group comparisons with multiple measurements, the two-way repeated-measures ANOVA was performed. In all cases, statistical significance was defined as p < 0.05. Post-hoc significance values were denoted as *p < 0.05, **p < 0.01, and ***p < 0.001. n.s. indicates not significant.

Schematic creation

Mouse brain coronal outlines, anatomical schematics, and graphical illustrations presented in Fig. 1a, c, e; 2a, l; 3a; 4a, b; 5a, c, e, i; 7a, e and Supplementary Figs 1a; 2a, b; 4a, b; 8a, c; 11a; 12a; 14a; 15f; 17a; 19a; 20a, d, e, h, j; 21a; 22a and 23a were originally created by the authors using Adobe Illustrator (Adobe Inc.). These illustrations were manually generated as vector graphics for this study and do not contain reproduced or adapted third-party copyrighted material.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

J.Z., B.Y., Z.H., and C.J. conceived the study. C.J., J.Z., and B.Y. wrote the paper. C.J. performed the optogenetic experiments, circuit mapping, and fiber photometry recording. X.T. and Q.M. performed the optogenetic experiments and chronic endoscopic recording. Y.L., E.C., and L.C. performed EMs recording experiments. X.T. participated in the ablation experiment. C.J., Y.L., H.W., and Y.X. analyzed the data. Y.C. helps us record and analyze sleep states. All authors contributed to the discussion and editing of the manuscript.

Peer review

Peer review information

Nature Communications thanks Peng Zhong and the other anonymous reviewer(s) for their contribution to the peer review of this work. A peer review file is available.

Funding

This study is funded by the NSF of China (Grant No. T2325008, 82588301 [to J.Z.]; Grant No. 32100803 [to B.Y.]), MOST of China (Grant No. 2022ZD0208604, 2022ZD0208605, 2024YFC2510800, 2024ZD0530302 [to J.Z.]; Grant No. 2022ZD0210000 [to B.Y.]), Key scientific technological innovation research project by Ministry of Education to J.Z., Shanghai Basic Research Special Zone Program (Grant No. 21TQ013 [to J.Z.]; Grant No. 25TQ009 [to B.Y.]), Lingang Laboratory, Grant No. LGL-8998-13 and LGL-5925-11. New Cornerstone Science Foundation XPLORER PRIZE, SANS Explorer Scholar, Shanghai Municipal Special Program for Basic Research on General AI Foundation Models (Grant No. 2025SHZDZX026D19).

Data availability

All data supporting the results of this study are available within the paper and its supplementary information. Source data underlying the statistical graphs are provided with this paper. Additional research materials generated in this study have been deposited in Figshare and are publicly available at 10.6084/m9.figshare.27684144. Source data are provided with this paper.

Code availability

Our scripts for analyzing sleep probability during optogenetic stimulation, fiber photometry, and micro-endoscopic calcium imaging, together with detailed user instructions, sample data, and example outputs, have been deposited in Code Ocean and are publicly available at https://codeocean.com/capsule/9306840/tree/v1 (10.24433/CO.7736716.v1)⁷⁶. In addition, we have created a dedicated GitHub repository for this project, which is accessible at https://github.com/18810386233/Parallel-circuit-in-oculomotor-nucleus-to-control-eye-movements-and-REM-sleep.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41467-026-74768-5.