Control of REM sleep by ventral medulla GABAergic neurons

Abstract

Rapid eye movement (REM) sleep is a distinct brain state characterized by activated electroencephalogram and complete skeletal muscle paralysis, and is associated with vivid dreams. Transection studies by Jouvet first demonstrated that the brainstem is both necessary and sufficient for REM sleep generation, and the neural circuits in the pons have since been studied extensively. The medulla also contains neurons that are active during REM sleep, but whether they play a causal role in REM sleep generation remains unclear. Here we show that a GABAergic (γ-aminobutyric-acid-releasing) pathway originating from the ventral medulla powerfully promotes REM sleep in mice. Optogenetic activation of ventral medulla GABAergic neurons rapidly and reliably initiated REM sleep episodes and prolonged their durations, whereas inactivating these neurons had the opposite effects. Optrode recordings from channelrhodopsin-2-tagged ventral medulla GABAergic neurons showed that they were most active during REM sleep (REMmax), and during wakefulness they were preferentially active during eating and grooming. Furthermore, dual retrograde tracing showed that the rostral projections to the pons and midbrain and caudal projections to the spinal cord originate from separate ventral medulla neuron populations. Activating the rostral GABAergic projections was sufficient for both the induction and maintenance of REM sleep, which are probably mediated in part by inhibition of REM-suppressing GABAergic neurons in the ventrolateral periaqueductal grey. These results identify a key component of the pontomedullary network controlling REM sleep. The capability to induce REM sleep on command may offer a powerful tool for investigating its functions.

Extended Data Figure 1. Comparison of spontaneous and laser-induced REM sleep

a, Mean EEG spectrogram and EMG amplitude before and after laser onset (averaged across all trials with laser onset falling on NREM sleep). b, Mean EEG spectrogram and EMG amplitude before and after spontaneous REM onset outside of laser stimulation periods (only REM episodes with duration >70 s were included). c, Comparison of EEG power spectra during spontaneous (gray) and laser-induced (blue) REM sleep. Blue shading, ±s.e.m. for laser-induced REM sleep.

Extended Data Figure 2. Effect of laser stimulation on brain states in eYFP control mice

a, Brain states in all trials from 5 mice aligned by time of laser stimulation (top) and probability of wake, NREM or REM states before, during, and after laser stimulation (bottom). Shading, 95% confidence intervals (CI). Blue bar, period of laser stimulation. Laser stimulation caused no significant change in the probability of any brain state (P > 0.34, bootstrap). b, Similar to (a), for trials in which laser onset fell on NREM sleep. c, Trials in which laser onset fell on wakefulness.

Extended Data Figure 3. Optogenetic activation of vM glutamatergic neurons induces wakefulness

a, Probability of wake, NREM or REM states before, during, and after laser stimulation (20 Hz, 120 s) in VGLUT2-Cre mice injected with AAV expressing ChR2-eYFP into the vM (n=3 mice). Shading, 95% CI. Blue bar, period of laser stimulation. Laser stimulation caused a significant increase in wakefulness (P<0.001, bootstrap) and decrease in NREM sleep (P<0.001). b, Mean durations of REM sleep episodes with and without laser stimulation. Each pair of dots represent data from one mouse. Laser stimulation shortened the duration of REM sleep episodes (n=3 mice, P=0.008, paired t-test). Error bar, ±s.d. c, Probability of wake (left), NREM (middle), and REM (right) states before, during, and after laser stimulation (20 Hz, 120 s) at different laser powers (color coded).

Extended Data Figure 4. Effect of laser stimulation of vM GABAergic neurons on transition probability between each pair of brain states in ChR2 (a-f) and eYFP control (g-l) mice

a, Probability of NREM (N)→REM (R) state transition within each 20 s period in ChR2 mice. Blue shading, period of laser stimulation (20 Hz, 120 s). Error bar, ± s.d. (bootstrap, n=6 mice). Baseline transition probability (red dashed line) was computed after excluding the laser stimulation period. The probability during laser stimulation was significantly higher than the baseline (P=0.02, bootstrap). b, Similar to (a), for NREM→wake (W) transition. The probability during laser stimulation was not significantly different from baseline (P=0.24). c, REM → wake, probability during laser stimulation was significantly lower than baseline (P<0.001), consistent with the effect of vM GABAergic neurons on prolonging REM duration (Fig. 2). d, REM→NREM, which rarely occurs in rodents. Laser stimulation caused no significant effect (P>0.99). e, Wake→REM, which rarely occurs in normal mice. Laser stimulation had no significant effect (P>0.99). f, Wake→NREM. Laser stimulation caused a significant reduction of the transition probability (P<0.001), indicating that during wakefulness vM GABAergic neuron activity has a wake-maintenance effect. g-l, Similar to (a-f), for eYFP control mice. Laser stimulation had no significant effect on any transition probability (P>0.05).

Extended Data Figure 5. Pharmacogenetic inactivation of vM GABAergic neurons reduces REM sleep

a, Brain states in a control (vehicle injection) and a CNO session from an example mouse. The recording session started 20 min after vehicle or CNO injection. b, Probability of each brain state during the first 2.5 hrs of the recording session, following injection of vehicle (gray) or two dosages of CNO (different shades of blue). Error bar, ±s.e.m. (n=6 mice). *, P<0.05; **, P<0.01; one-way ANOVA with post hoc Dunnett's test. c, Similar to (b), but during the second half of the recording session (2.5 to 5 hrs). There was no significant difference between control and CNO at any dosage (P>0.12). d, Latency of first REM sleep episode (from the beginning of each recording session). e, Frequency of REM episodes during the first 2.5 hrs of the recording session. f, Duration of REM episodes during the first 2.5 hrs of the session. The reduction of REM sleep caused by pharmacogenetic inactivation of vM GABAergic neurons appears to be due to the reduction of frequency rather than duration of REM episodes.

Extended Data Figure 6. Optogenetic identification of vM GABAergic neurons

a, Left, reliability and temporal jitter of laser-evoked spikes in identified (blue) and unidentified (gray) units recorded in the vM. Note that the identified and unidentified units form two distinct clusters, with high reliability and low jitter for identified units. Dark/light symbols, data during 30/15 Hz laser stimulation. Middle, distribution of delays of laser-evoked spiking for 21 identified GABAergic neurons. Delay is defined as timing of the first spike following each laser pulse. Right, distribution of correlation coefficient between laser-evoked and spontaneous spike waveforms for all 21 identified vM GABAergic neurons. b, Fluorescence image of a coronal section showing the position of an electrolytic lesion at the end of the optrode tract (arrow). Blue, DAPI staining. c, Positions of the 21 identified vM GABAergic neurons from 5 mice. Each dot indicates one neuron. All 20 neurons with recording periods encompassing all three brain states showed maximal firing rates during REM sleep. The wakeful behavior during which the neuron showed the maximal firing rate is color-coded. Black, neurons for which the recording period did not include all wakeful behaviors. Schemes of brain sections adapted from Allen Mouse Brain Atlas (Website: © 2015 Allen Institute for Brain Science. Allen Mouse Brain Atlas [Internet]. Available from: http://mouse.brain-map.org).

Extended Data Figure 7. Activity of vM GABAergic neurons at REM sleep onset and offset

a, Mean firing rates of vM GABAergic neurons at REM onset. Shading, ±s.e.m. Dark blue, period in which firing rate was significantly higher than baseline (P<0.05, Wilcoxon signed-rank test); baseline was defined as the average firing rate during the 10 s intervals 60 s before and after REM sleep. b, Mean firing rates at REM offset.

Extended Data Figure 8. Effect of laser stimulation of vM neurons on several wakeful behaviors

a, Probability of moving (MV), running (RU), eating (ET), and grooming (GR) before, during, and after laser stimulation (20 Hz, 120 s) in GAD2-Cre mice injected with AAV expressing ChR2-eYFP into the vM (n=8 mice). Unclassified behaviors are not shown. Blue bar, period of laser stimulation. Laser stimulation caused a significant increase in eating (P=0.008, Wilcoxon signed-rank test). b, Similar to a, but in control mice expressing eYFP (n=4 mice).

Extended Data Figure 9. Firing rates of unidentified vM neurons

a, Firing rates of unidentified units in the three brain states. Each line represents data from one neuron. Gray bar represents average over units (n=24). b, Left, relative firing rates of the units across brain states. The firing rates of each unit were normalized by its maximum. Right, normalized firing rates across different wakeful behaviors. c, Mean firing rates of unidentified vM neurons at REM onset. Shading, ±s.e.m. Dark blue, period in which firing rate was significantly higher than baseline (P<0.05, Wilcoxon signed-rank test). d, Mean firing rates of unidentified vM neurons at REM offset.

Extended Data Figure 10. Co-expression of GABA and glycine in vM neurons

a, Fluorescence image of ChR2-eYFP expressing neurons in the vM of a GAD2-Cre mouse injected with Cre-inducible AAV. b, Immunohistochemical staining for glycine. c, Superposition of eYFP-expression and glycine staining. In total, 94% (273/289) of eYFP-expressing cells were glycine positive (n=3 mice).

Figure 1. Optogenetic activation of vM GABAergic neurons induces REM sleep

a, Top, schematic of optogenetic experiment (adapted from Paxinos and Franklin mouse brain atlas). Bottom, fluorescence image of vM (dashed box in schematic) in a GAD2-Cre mouse injected with AAV expressing ChR2-eYFP (green). Blue, DAPI. GI, gigantocellular reticular nucleus; LPGI, lateral paragigantocellular nucleus; IO, inferior olive; Amb, ambiguus nucleus. ChR2-eYFP was expressed within ~400 μm from injection site and mainly localized within the LPGI. b, Example experiment. Shown are EEG power spectrogram, EMG amplitude, and brain states (color coded). Blue shading, laser stimulation period (20 Hz, 120 s). c, Brain states in all trials from 6 mice (top) and probability of wake, NREM or REM states (bottom) before, during, and after laser stimulation. Shading, 95% confidence intervals (CI). Blue bar, laser stimulation period. d, Similar to (c), for trials with laser onset falling on NREM sleep (probability of NREM is 1 immediately before laser onset). e, Trials with laser onset falling on wakefulness.

Figure 2. Activation of vM GABAergic neurons prolongs REM sleep duration

a, Schematic of closed-loop stimulation. Laser was turned on after spontaneous REM onset and turned off at REM termination. b, Example recording containing REM episodes with and without laser stimulation (blue shading). c, Mean REM sleep duration with and without laser stimulation, in ChR2 (left, n=6 mice) and eYFP control mice (right, n=4). Each pair of dots, data from one mouse. Error bar, ±s.d. d, Similar to (c), but with green laser stimulation in Arch/Halo (left, n=4 mice) and eYFP control (right, n=4) mice.

Figure 3. Firing rates of identified vM GABAergic neurons across brain states

a, Example recording of spontaneous and laser-evoked spikes from a vM neuron. Blue ticks, laser pulses (30 Hz). b, Comparison between laser-evoked (blue) and averaged spontaneous (red) spike waveforms from this unit. c, Spike raster showing multiple trials of laser stimulation at 15 and 30 Hz. d, Firing rates of an example vM GABAergic neuron. e, Left, mean firing rates of the neuron in (d) during different brain states. Error bar, ±s.e.m. Right, firing rates of the neuron during different wakeful behaviors. f, Firing rates of 20 identified vM GABAergic neurons during different brain states. Each line shows firing rates of one unit; gray bar, average across units. g, Relative firing rates across brain states. The rates of each neuron were normalized by its maximum. All 20 neurons showed maximum firing rates during REM sleep (the difference was significant at P<0.005 for all 20 neurons compared to NREM and wakefulness; Wilcoxon rank sum test, post-hoc Bonferroni correction). h, Relative firing rates of 17 vM GABAergic neurons during different wakeful behaviors.

Figure 4. Inhibition of vlPAG GABAergic neurons by vM projections promotes REM sleep

a, Top, schematic showing injection of AAV expressing ChR2-eYFP into the vM of a GAD2-Cre mouse. Bottom, fluorescence image of vM axons in the pons and midbrain (position of coronal section indicated by dashed line in schematic). Scale bar, 500 μm; Green, eYFP; Blue, DAPI. b, Top, schematic showing simultaneous injections of RetroBeads (RB) in the pons and FluoroGold (FG) in the spinal cord. Bottom, fluorescence image of vM, showing neurons labeled by RB and FG. Among the 1235 FG- and 881 RB-labeled neurons, only 27 were double labeled. Scale bar, 500 μm. c, Left, probability of wake, NREM or REM states before, during, and after laser stimulation of vlPAG GABAergic neurons (20 Hz, 300 s; n=6 mice). Shading, 95% CI; Blue bar, laser stimulation period (20 Hz, 300s). Right, mean REM durations with and without vlPAG stimulation (n=5 mice). Each pair of dots, data from one mouse. Error bar, ±s.d. d, Left, schematic showing rabies-mediated transsynaptic tracing. TVA, EnvA receptor; G, rabies glycoprotein; RVdG, G-deleted rabies virus. Middle, fluorescence image of vlPAG in a GAD2-Cre mouse. Scale bar, 500 μm. Bottom middle, enlarged view of region in white box showing starter cells (yellow, expressing both GFP and mCherry, arrowheads; Scale bar, 20 μm). Right, rabies-labeled presynaptic neurons in vM (same brain as in middle panel). Scale bar, 500 μm. e, Rabies-labeled presynaptic neurons in vM are GABAergic and glycinergic. Lower panel, enlarged view of region in white box, containing GFP labeled neurons expressing glycine (Gly, arrowheads; Scale bar, 50 μm). GABA and glycine co-exist in a high percentage of vM neurons (Extended Data Fig. 10), suggesting that the glycinergic rabies-labeled neurons in the vM are also GABAergic. In total, 82% (185/226) rabies-labeled cells were glycine positive (n=3 mice). f, Left, probability of wake, NREM or REM states before, during, and after laser stimulation of vM axons in vlPAG (20 Hz, 120 s; n=5 mice). Shading, 95% CI. Right, mean REM durations with and without vM axon stimulation (n=5 mice). Error bar, ± s.d.