A REM-active basal ganglia circuit that regulates anxiety

REM sleep has been hypothesized to promote emotional resilience, but any neuronal circuits mediating this have not been identified. We find that in mice, somatostatin (Som) neurons in the entopeduncular nucleus (EPSom)/internal globus pallidus are predominantly active selectively during REM sleep. This unique REM activity is necessary and sufficient for maintaining normal REM sleep. Inhibiting or exciting EPSom neurons reduced or increased REM sleep duration, respectively. Activation of the sole downstream target of EPSom neurons, Vglut2 cells in the lateral habenula (LHb), increased sleep via the ventral tegmental area (VTA). A simple chemogenetic scheme to periodically inhibit the LHb over 4 days selectively removed a significant amount of cumulative REM sleep. Chronic REM reduction correlated with mice becoming anxious and more sensitive to aversive stimuli. Therefore, we suggest that REM sleep, in part generated by the EP→LHb→VTA circuit identified here, could contribute to stabilizing reactions to habitual aversive stimuli.


43
The hypothesis that sleep serves a restorative function has focused largely on NREM 44 sleep, the stage of sleep that occurs first [1]. Recently, however, REM sleep in humans 45 has been found to be the deepest and most subjectively satisfying stage of sleep [2,3], 46 and is also the sleep state where capillary blood flow in the mouse brain is selectively 47 boosted [4]. REM sleep has been hypothesized to promote emotional health or resilience 48 The LHb receives diverse inputs [30][31][32][33][34][35], but including a unique projection from 73 somatostatin (Som)-glutamate-GABA neurons in the entopeduncular nucleus [26,[36][37][38]. 74 We performed anterograde tracing by injecting AAV-DIO-mCherry in the EP of Som-Cre 75 mice and confirmed that EP Som neurons project exclusively to the LHb as previously 76 reported [26,36,[39][40][41] (Figure 1A and 1B). We then used calcium photometry to study 77 the natural dynamics of the EP Som neurons across the sleep-wake cycle. AAV-  GCaMP6s was injected into the EP of Som-Cre mice with an optical fiber implanted over 79 the EP to acquire population activity of EP Som neurons ( Figure 1C and Figure S1A). We  Recordings from EP Som -GFP control mice displayed no such vigilance state-dependent 86 variations ( Figure S1B and S1C). 88 To assess whether EP Som activity is necessary for natural REM sleep, we inhibited these 89 neurons chemogenetically. AAV-DIO-hM4Di-mCherry (referred to as "hM4Di" hereafter) 90 was bilaterally injected into the EP of Som-Cre mice (Figure 2A Figure 2B and 2C). Reduced REM sleep amounts were due to decreased 97 numbers of long (>1-min) REM-sleep episodes (from 54% to 29% P = 0.03; n = 7). In 98 We reasoned that EP Som neurons regulate REM sleep via their sole downstream target, 123 the LHb. Three experiments were carried out to confirm this. First, we examined whether 124 the output from EP to LHb correlated with REM sleep by monitoring the Ca 2+ transients at 125 EP→LHb terminals. We injected AAV-DIO-GCaMP6s in EP Som neurons and positioned 126 the optical fiber over the LHb. Our results showed that as for the EP Som somata, the Ca 2+ 127 transients at EP→LHb terminals peaked during REM sleep ( Figure S4). NREM-REM 128 sleep transitions were associated with prominent increases of activity at these terminals 129 (P = 0.0002). Next, we examined whether the LHb is similarly modulated during REM   contributing to sleep-wake regulation [42,43]. In-vivo fiber-photometry recordings at 158 LHb→DRN and LHb→VTA terminals across sleep-wake cycle were achieved similarly as 159 described above, by expressing AAV-DIO-GCaMP6s in LHb Vglut2 neurons, and with optical 160 fibers placed over terminals in either the DRN or VTA ( Figure 5A). We observed a wake-161 specific calcium signal at the LHb→DRN terminals, while the activity was absent during   171 We assessed the functional significance of REM sleep generated by the circuit. We 172 discovered that repetitive dosing of LHb Vglut2-hM4Di mice with CNO led to a chronic reduction 173 of REM sleep. LHb Vglut2-hM4Di mice were given saline or CNO once a day at the beginning of the inactive lights-ON phase for consecutive four days. This reduced REM sleep by 175 24%, 15%, 11% and 7% over the four days but did not change the measurable amount of 176 NREM sleep or WAKE ( Figure S10). Then, 16 hours after the last saline or CNO 177 administration on day 4, mice were evaluated in several behavioral tests ( Figure 6A). In

185
To further assess whether REM sleep contributes to the performance of innate 186 behaviors, such as defensive behaviors, we performed a looming test in which a moving 187 disk is presented to mimic distal threats. In this test, mice generally respond with freezing  197 The pathways that contribute to REM sleep generation are still being elucidated [21][22][23][24]198 46-49]. The basal ganglia, a collection of subcortical nuclei that include the nucleus 199 accumbens, the caudate-putamen, the globus pallidus externa, the EP (globus pallidus downstream target, the LHb cells, fires with aversive events or disappointment [26,28,207 29], in other words, when an outcome is worse than expected [30,54]. This 208 EP→LHb→VTA circuit helps animals learn about negative experiences and/or adopt 209 passive coping strategies [54][55][56]. That this circuitry seems to have an exact mapping with 210 the REM sleep circuitry we report here seems more than a coincidence. Empirically, we 211 discovered that a simple chemogenetic scheme to periodically chemogentically inhibit the 212 LHb over 4 days selectively removed a significant amount of cumulative REM sleep, 213 without affecting measurable amounts of NREM or wakefulness. Testing 16 hours later 214 after the last inhibitory dose, chronic REM reduction correlated with mice becoming more 215 sensitive to aversive stimuli. Therefore, we suggest that baseline REM sleep could 216 contribute to stabilizing reactions to habitual aversive stimuli.

217
The LHb is a relay nucleus with many channels (reflecting diverse inputs), and receives 218 innervations from about forty brain regions, including from the EP, but also from the 219 hypothalamus, midbrain and brainstem areas [30][31][32][33]35]. Our data showed that the input 220 from the preoptic area of the hypothalamus is selectively active during wakefulness, likely 221 contributing to the wake-active calcium signals in the LHb we observed.  is more active during depression [60], and we predict that it is this habenula overactivity 242 that causes the enhanced REM in this condition. The enhanced REM in people living with 243 depression is often assumed to be an adverse outcome. However, this enhanced REM 244 could, in fact, be beneficial and aid emotional processing, rather than being harmful. (0.5 Hz, -3dB, an FFT size of 512 was the designated time window) using a digital filter.

321
The EMG signals were band-pass filtered between 5-45 Hz (-3dB). Power in the delta 322 (0.5-4 Hz), theta (6-10 Hz) bands and theta to delta band ratio were calculated, along with 323 the root mean square (RMS) value of the EMG signal (averaged over a bin size of 5 s).

324
All of these data were used to define the vigilance states of WAKE, NREM and REM by 325 an automatic script. Each vigilance state was screened and confirmed manually afterward.

326
The peak frequency during NREM epochs were analyzed using Fourier transform power 327 spectra to average power spectra over blocks of time. Electronic Design).

375
The photometry signal was aligned with the EEG and EMG recordings. For each 376 experiment, the photometry signal F was normalized to the baseline signal using