Striatal adenosine A2A receptor neurons control active-period sleep via parvalbumin neurons in external globus pallidus

Dysfunction of the striatum is frequently associated with sleep disturbances. However, its role in sleep-wake regulation has been paid little attention even though the striatum densely expresses adenosine A2A receptors (A2ARs), which are essential for adenosine-induced sleep. Here we showed that chemogenetic activation of A2AR neurons in specific subregions of the striatum induced a remarkable increase in non-rapid eye movement (NREM) sleep. Anatomical mapping and immunoelectron microscopy revealed that striatal A2AR neurons innervated the external globus pallidus (GPe) in a topographically organized manner and preferentially formed inhibitory synapses with GPe parvalbumin (PV) neurons. Moreover, lesions of GPe PV neurons abolished the sleep-promoting effect of striatal A2AR neurons. In addition, chemogenetic inhibition of striatal A2AR neurons led to a significant decrease of NREM sleep at active period, but not inactive period of mice. These findings reveal a prominent contribution of striatal A2AR neuron/GPe PV neuron circuit in sleep control.


Introduction
It is widely known that the striatum (caudate putamen), which resides in the forebrain and serves as the primary input nucleus of the basal ganglia, is involved in an array of physiological processes including motor control, habit formation, and goal-directed behaviors (Durieux et al., 2011;Graybiel, 2008;Li et al., 2016). Up to 90% of patients with Parkinson's disease (PD) exhibit severe sleep disturbances, which is one of the most frequent non-motor symptoms (Arnulf et al., 2008). Since the substantia nigra pars compacta (SNc), which projects primarily to the striatum, is the major area of degeneration in PD, dysfunction of the striatum may contribute to sleep disturbances in PD patients. However, to date, few studies have examined the role of the striatum in sleep-wake regulation.
The limited reports on the role of the striatum in sleep-wake regulation are controversial. In animal studies, surgical removal of the striatum in cats (Villablanca, 1972) and striatal excitotoxic lesions in rats (Mena-Segovia et al., 2002) decrease time spent in sleep. However, electrical lesion of the rat striatum selectively increases rapid eye movement (REM) sleep (Corsi-Cabrera et al., 1975), and lesion by ibotenic acid induces an increase in the non-rapid eye movement (NREM) and a decrease in the REM sleep (Qiu et al., 2010). In humans, a H 2 15 O PET study shows that cerebral blood flow (represents activity of the brain) in caudate nucleus, which is equivalent to the rostral striatum of rodents (Kreitzer, 2009), increases during REM sleep and decreases during slow-wave sleep (Braun et al., 1997). Since lesion and imaging methods exhibit limitations, specific manipulation of neuronal activity with simultaneous electroencephalogram (EEG) recording provides a powerful tool to understand the role of the striatum in sleep-wake cycles. The striatum contains GABAergic medium spiny neurons (MSNs, 95%) and interneurons (5%) (Kreitzer, 2009). MSNs are divided into two projection neuron classes. One class consists of striatopallidal neurons that express the adenosine A 2A receptors (A 2A Rs) and dopamine D 2 receptors (D 2 Rs), and project to the external globus pallidus (GPe). The other class consists of striatonigral neurons that express adenosine A 1 receptors (A 1 Rs) and dopamine D 1 receptors (D 1 Rs), and project primarily to the substantia nigra pars reticulata (SNr) and the internal globus pallidus (GPi) (Kreitzer, 2009). Both A 1 Rs and A 2A Rs have been reported to regulate sleep (Basheer et al., 2004;Thakkar et al., 2003;Urade et al., 2003). Among them, A 1 Rs contribute to sleep induction in a region-dependent manner, whereas A 2A Rs play a predominant role in sleep induction (Huang et al., 2005;Lazarus et al., 2012;Wang et al., 2017). Moreover, it has been reported that the expression level of A 2A Rs is altered in the striatum of PD patients (Mishina et al., 2011;Ramlackhansingh et al., 2011), which may change the activity of striatal A 2A R neurons (Gerfen et al., 1990;Mitchell et al., 1989), thus contributing to sleep disturbances in PD patients. However, it is unknown about the role and circuits of striatal A 2A R neurons in regulation of sleepwake behavior.
To address these questions, we employed a chemogenetic technique known as designer receptor exclusively activated by designer drugs (DREADD) (Alexander et al., 2009), which specifically and non-invasively manipulates neuronal activity based on the principle of Cre/LoxP recombination (Farrell and Roth, 2013), and neural tracing, immunoelectron microscopy, as well as optogenetic and electrophysiological methods. We selectively manipulated activity of striatal A 2A R neurons in Adora2a-Cre mice to topographically investigate their contributions to sleep and characterize the functional connectivity between striatal A 2A R neurons and neurons in the GPe. Then, in Pvalb-Cre mice, we studied the role of GPe parvalbumin (PV) neurons, which are downstream targets of A 2A R neurons, in sleep-wake behavior. In addition, using Adora2a/Pvalb-Cre mice expressing DREADD in striatal A 2A R neurons and selective lesion GPe PV neurons, we examined the neuronal circuit for sleep induced by activation of A 2A R neurons in the striatum. The results revealed for the first time that A 2A R neurons in the rostral and central striatum contribute to sleep-wake behavior through the striatal A 2A R neuron/GPe PV neuron pathway.
Based on the medial-lateral and rostral-caudal axis of the striatum, we injected DREADD-AAV into four subregions (rostral, centromedial, centrolateral and caudal) to investigate their contributions to sleep regulation mediated by striatal A 2A R neurons. Following vehicle injection (i.p.) at 7 p.m. (light off), the beginning of the active period when mice usually show high levels of arousal, mice expressing hM3Dq receptors in the rostral striatum ( Figure 1A) displayed long bouts of wakefulness marked by low EEG slow-wave activity and high electromyogram (EMG) activity ( Figure 1B). However, CNO (1 mg/kg) injection induced an increase in NREM sleep and remarkably shortened NREM sleep latency ( Figure 1K), with a decrease in wakefulness for 3 hr ( Figure 1B and C). The amount of CNO-induced NREM sleep was significantly increased by 2.2-fold during the 3 hr postinjection period as compared with vehicle ( Figure 1I). Consistent with these results, the amount of wakefulness was significantly decreased by 32% following administration of CNO ( Figure 1B, C and J), but REM sleep was not significantly changed (Figure 1-figure supplement 3A and E). CNO injection induced an increase in the mean duration of NREM sleep with a decrease in wakefulness, and had no effect on the episode numbers of 3 vigilance stages (Figure 1-figure supplement 4). Notably, chemogenetic activation of A 2A R neurons in the rostral striatum did not change NREM sleep during the subsequent 9 hr of the active period ( Figure 1I). Along with the increase of NREM sleep amount after CNO injection, however, no change was detected in EEG power density of NREM sleep during the 3 hr post-CNO injection as compared with vehicle ( Figure 1D), which suggests that the increased sleep was similar to a physiological sleep pattern. Similar to the rostral striatum, chemogenetic activation of A 2A R neurons in the centromedial or centrolateral striatum increased NREM sleep by 2.2-fold and 1.8-fold, respectively ( Figure 1I and J and Figure 1-figure supplements 3B, C, F, G, 4 and 5).
Unexpectedly, administration of CNO did not change the total amount, the episode number, the mean duration and the latency of NREM sleep, or EEG power density in mice expressing hM3Dq receptors in the caudal striatum ( Figure 1I) when compared with vehicle ( Figure   To determine whether output pattern varies in different subregions of the striatum, we examined pallidal projections by injecting CMV-lox-stop-hrGFP-AAV ( Figure 2A) into different subregions of the striatum in Adora2a-Cre mice. This virus caused robust expression of humanized Renilla green fluorescent protein (hrGFP) in the cytosol of A 2A R neurons (Zhang et al., 2013). Using immunofluorescence, we found that hrGFP-expressing neurons displayed typical morphology of MSNs ( Figure 2B) in the striatum, and positive signals for A 2A R immunoreactivity ( Figure 2C-F). Furthermore, we found 3 types of axonal arrangement based on different virus injection sites in the striatum. In sagittal sections, axons of A 2A R neurons in the rostral striatum were found in the rostral GPe with a discoidal field paralleling the strio-pallidal border ( Figure 2G). Axons from the central striatum were distributed not only in the rostral but also the caudal GPe, forming similar discoidal areas paralleling the strio-pallidal border ( Figure 2G). In contrast, axons from the caudal striatum were distributed only in the caudal GPe ( Figure 2G). In addition, it is notable that axons of the lateral striatum projected preferentially to the lateral GPe. These findings indicate that the projections of A 2A R neurons in different subregions of the striatum are organized topographically in the GPe and suggest that the topographical projections of A 2A R neurons may contribute to the discrepancies in A 2A R neuron-mediated sleep.
Striatopallidal terminals formed more symmetric synapses with PVpositive neurons than PV-negative neurons in the GPe Neurons within the GPe can be divided into PV-positive and PV-negative neurons . To examine synapses between axon terminals of A 2A R neurons and PV-positive neurons expressing PV or PV-negative neurons not expressing PV in the GPe, we processed mouse GPe samples expressing hrGFP originating from A 2A R neurons in the striatum for immunoelectron microscopy. In the GPe, hrGFP-IR elements were filled by floccular diaminobenzidine (DAB) reaction products and represented the terminal of striatal A 2A R neurons. PV-IR ones were filled with punctate reaction products of Vector very intense purple (V-VIP) and represented the GPe PV neurons (Li et al., 2002). However, PV-unlabeled dendrites and PV-unlabeled perikarya in the GPe, were not filled with DAB or VIP reaction products and represented the PV-negative neurons in the GPe. We observed that hrGFP-IR terminals established symmetric synapses with dendrites that were labeled or unlabeled for PV ( Figure 2H and J). Moreover, a small number of perikarya, which were labeled or unlabeled for PV, received symmetric synapses from hrGFP-IR terminals ( Figure 2I and K). However, hrGFP-labeled terminals formed significantly more synapses with PV-IR profiles than PV-unlabeled profiles (81.2% vs. 18.8%, n = 377 synapses from five mice; Figure 2L) in the rostral GPe. In contrast, in the caudal GPe, hrGFP-IR terminals preferentially formed synapses with PV-unlabeled profiles than PV-labeled profiles (61.3% vs. 38.7%, n = 275 synapses from five mice; Figure 2L).
These anatomical findings indicate that A 2A R neurons in the rostral striatum preferentially form symmetric synapses with PV-positive neurons in the rostral GPe, while A 2A R neurons in the caudal striatum preferentially form synapses with PV-negative neurons in the caudal GPe, indicating a rostral-caudal variation in striatopallidal connections.

Optogenetic stimulation of the striatopallidal terminals inhibited GPe neurons
To explore the functional nature of striatopallidal connections, we employed an optogenetic-assisted circuit mapping approach (O'Connor et al., 2015). Channelrhodopsin-2 (ChR2), a blue light-gated cation channel, was expressed in A 2A R neurons by injecting hSyn-DIO-ChR2-mCherry-AAV into the striatum of Adora2a-Cre mice ( Figure 3A and Figure 3-figure supplement 1A). After 3 weeks, acute coronal brain slices containing the striatum or GPe were prepared for in vitro patch-clamp recording. We first tested responses from somata of ChR2-expressing neurons, which were presum- ably    Next, cells in the GPe were randomly patch-clamped while blue light flashes (1 ms) at 10 Hz were used to stimulate axon terminals of A 2A R neurons. Previous studies have demonstrated two nonoverlapping cell classes in the GPe, one expressing PV, and the other expressing forkhead box protein P2 (FoxP2), a transcription factor (Abdi et al., 2015;Dodson et al., 2015;Hernández et al., 2015;Mallet et al., 2012). Thus, to identify the cell type of recorded GPe neurons, we added biocytin to the pipette solution, and performed immunostaining using PV as marker for PV-positive neurons, and FoxP2 as marker for PV-negative neurons, after recording. We found that light-evoked inhibition could be recorded in PV-positive and PV-negative neurons in the GPe. In the cell-attached patch mode, photostimulation decreased the firing rate of most PV-positive neurons, which showed immunoreactive signals for PV but not FoxP2 ( Figure 3B and C), to 58% of the spontaneous firing rate (from 30.3 ± 2.9 to 17.6 ± 2.6 Hz, n = 19 from 10 mice; Figure 3D-F), and the firing rate recovered immediately when photostimulation was terminated. In PV-negative neurons which were FoxP2-IR ( Figure 3G and H), photostimulation decreased the firing rate to 25% of the spontaneous firing rate (from 9.2 ± 0.9 to 2.3 ± 0.8 Hz, n = 12 from 10 mice; Figure 3I-K). Notably, a rebound in the firing rate after photostimulation was observed in 6 of 12 PV-negative neurons (10 mice).
In the whole-cell voltage-clamp mode, flashes of blue light evoked fast inhibitory postsynaptic currents (IPSCs) in both PV-positive and PV-negative neurons ( Figure 3F and K) with a latency of less than 5 ms ( Figure 3N), indicating a direct connection between terminals of A 2A R neurons and PV-positive or PV-negative neurons. In addition, the light-evoked IPSCs were completely abolished by picrotoxin (PTX, 100 mM; Figure 3F and K), indicating that these responses were mediated by GABA released from axon terminals of A 2A R neurons and postsynaptic GABA A receptors on GPe neurons. Furthermore, light-evoked IPSCs were recorded in 76% of neurons (45 of 59 neurons from 10 mice) in the rostral GPe, and 53% of neurons (10 of 19 neurons from 5 mice) in the caudal GPe ( Figure 3L). In addition, the amplitude of the first IPSC evoked by blue light flashes at 10 Hz was significantly larger in PV-positive neurons than in PV-negative neurons (860.4 ± 157.3 pA, n = 23 vs. 234.3 ± 69.7 pA, n = 18, from 10 mice) in the rostral GPe ( Figure 3M). Notably, we did not detect connections between terminals of A 2A R neurons and PV-positive neuron in the caudal GPe, possibly due to their low numbers in this region. Finally, the spontaneous firing rates of PV-positive and PVnegative neurons recorded in current condition are consistent with previous studies ( Figure 3O) . Taken together, these data support anatomical studies indicating that striatal A 2A R neurons preferentially innervate and inhibit PV neurons in the rostral GPe.
In addition, we examined the effects of photostimulation of ChR2-expressing A 2A R neuron terminals in the GPe on sleep-wake behavior in freely moving mice. Light stimulation for 1 hr with optical fiber implanted into the GPe, containing ChR2-expressing A 2A R neuron terminals, remarkably increased NREM sleep by 1.8-fold (Figure 3-figure supplement 2), strongly suggesting that striatal A 2A R neurons promoted sleep by inhibiting neurons, more likely PV neurons, in the GPe.

Chemogenetic inhibition of PV neurons in the GPe promoted NREM sleep
PV neurons in the GPe serve as an important downstream target for striatal A 2A R neurons. To test whether PV neurons in the GPe are involved in sleep, we transduced a Cre-recombinase-enabled chemogenetic inhibitory system, a Gi-coupled DREADD, hM4Di receptor, using AAV microinjection into the GPe of Pvalb-Cre mice ( Figure 4A). The hSyn-DIO-hM4Di-mCherry-AAV caused robust and confined expression of hM4Di receptors in the GPe ( Figure 4B). Immunofluorescence staining of brain slices with anti-PV antibody revealed that mCherry was exclusively expressed in PV neurons ( Figure 4C), confirming the requirement for Cre activity to enable expression of hM4Di receptors. Bath application of CNO (5 mM) reduced the spontaneous firing rate of PV neurons expressing hM4Di receptors in the GPe ( Figure 4D). Systemically, CNO (1 mg/kg, i.p.) injections caused mice to fall asleep with an increased NREM sleep and decreased wakefulness, and this effect was sustained for 3 hr ( Figure 4E-G). The amount of CNO-induced NREM sleep was significantly increased by 1.6- and CNO (closed blue circle) injections during active period in mice expressing hM4Di receptors in PV neurons of the GPe. n = 6, two-way repeated measures ANOVA, paired t test, NREM: F 1,10 = 18.698, p=0.002 (ANOVA), *p=0.026, *p=0.033, **p=0.001 (t-test); REM: F 1,10 = 1.209, p=0.297 (ANOVA); Wake: F 1,10 = 14.614, p=0.003 (ANOVA), *p=0.036, *p=0.043, **p=0.002 (t-test). (G) Total amounts of NREM sleep, REM sleep, and wakefulness during the 3 hr post-injection period (7 p.m.-10 p.m.) and the following 9 hr of the active period (10 p.m.-7 a. m.) following vehicle or CNO injections in mice expressing hM4Di receptors in GPe PV neurons. n = 6, paired t test, NREM: **p=1.972E-4, *p=0.027; REM: p=0.088, p=0.106; Wake: **p=2.508E-4, *p=0.028. (H) Relative average EEG power spectrum of NREM sleep and quantitative changes in power for delta (0.5-4.0 Hz) frequency bands (insert) during the 3 hr period after CNO and vehicle injections. n = 6, paired t test, p=0.258. See Figure 4source data 1. DOI: https://doi.org/10.7554/eLife.29055.020 The following source data is available for figure 4: Source data 1. Sample size (n), mean and SEM are presented for the data in fold with a decrease in wake by 26% during the 3 hr post-injection period as compared with vehicle ( Figure 4F and G). However, REM sleep ( Figure 4F and G), the latency to the first NREM sleep and EEG power density of NREM sleep ( Figure 4H) during the 3 hr post-CNO injection was not altered. These findings demonstrate that inhibition of PV neurons in the GPe mimics the effect of activation of striatal A 2A R neurons and confirm PV neurons of the GPe as a critical downstream target for striatal A 2A R neuron-mediated sleep.

Lesion of PV neurons in the GPe abolished the increase in NREM sleep caused by activation of striatal A 2A R neurons
To test whether A 2A R neurons in the striatum promote NREM sleep by innervating PV neurons in the GPe, we crossed Adora2a-Cre mice with Pvalb-Cre mice to generate Adora2a/Pvalb-Cre mice expressing Cre recombinase in A 2A R neurons and PV-positive neurons. Using the Adora2a/Pvalb-Cre mice, hSyn-DIO-hM3Dq-mCherry-AAV was injected into the rostral and central striatum to express hM3Dq receptors (mCherry+) in A 2A R neurons ( Figure 5A-D). Then, mice, injected with hSyn-DIO-hM3Dq-mCherry-AAV, were microinjected with Flex-taCasp3-TEVp-AAV into the GPe to kill PV-positive neurons ( Figure 5C) . Non-lesion mice, injected with hSyn-DIO-hM3Dq-mCherry-AAV, were microinjected with DIO-eGFP-AAV in the GPe ( Figure 5A). In the GPe, we quantified the number of PV-IR neurons, human neuronal protein HuC/HuD (HuCD)-IR neurons  which represented total neurons, and FoxP2-IR neurons which represented the PV-negative neurons after three weeks of microinjection of taCasp3 or eGFP viral vector ( Figure 5E-G). We found that microinjection of taCasp3 vector significantly decreased the number of PV-IR and HuCD-IR neurons compared with the control group (p<0.001; Figure 5E-G). In contrast, microinjection of taCasp3 vector did not change the number of FoxP2-IR neurons (p=0.738; Figure 5E-G). These data suggested that PV neurons were eliminated in the GPe of the lesion group, but PV-negative neurons were not affected. In freely moving mice, CNO (1 mg/kg) injections caused non-lesion mice that expressed hM3Dq receptors on the striatal A 2A R neurons to fall asleep with an increase in NREM sleep lasting for 4 hr ( Figure 5H and J). The amount of CNO-induced NREM sleep was significantly increased by 2.6-fold, with a decrease in wakefulness by 46% during the 4 hr post-injection period as compared with vehicle ( Figure 5H and J).
It is very important to note that the PV-lesion mice also expressing hM3Dq receptors in the striatal A 2A R neurons did not fall asleep after CNO injection. There was no significant change in NREM sleep and wakefulness following CNO injections in the PV-lesion mice ( Figure 5I and J). In addition, REM sleep was not changed in both groups following CNO injection ( Figure 5HJ). Thus, lesion of PV neurons in the GPe abolished the increase in NREM sleep caused by activation of A 2A R neurons in the striatum, indicating that the striatal A 2A R neurons control sleep behavior via striatal A 2A R neurons/GPe PV neurons.

Chemogenetic inhibition of striatal A 2A R neurons induced wakefulness during active period
We demonstrated that chemogenetic activation of A 2A R neurons in the rostral, centromedial and centrolateral but not caudal striatum promoted NREM sleep. To examine whether striatal A 2A R neurons are necessary for sleep under baseline conditions, we bilaterally injected hSyn-DIO-hM4Di-mCherry-AAV in the rostral and central striatum of Adora2a-Cre mice ( Figure 6A, C and D) to chemogenetically inhibit A 2A R neurons. Whole-cell current-clamp recordings of A 2A R neurons expressing hM4Di receptors showed decreased firing in response to 350 pA current injection during CNO (5 mM) application, and reversal after washout ( Figure 6B). Administration of CNO (1 mg/kg) at 7 p.m. (active period) significantly decreased NREM sleep by 70% and REM sleep by 74% in freely moving mice ( Figure 6E-G), with an increase in wakefulness during the 4 hr post-injection period as compared with vehicle ( Figure 6E-G). Notably, no sleep rebound was observed in these mice during the subsequent active period ( Figure 6G). Along with decreases in amount of NREM sleep and REM sleep after CNO injection, no change was detected in the delta power density of NREM sleep and the theta power density of REM sleep during the 4 hr post-CNO injection as compared with vehicle ( Figure 6H and I). Unexpectedly, administration of CNO at 9 a.m. (inactive period), which is a time of high sleep pressure in mice, did not change time spent in each stage or the power density of NREM sleep, REM sleep, and wakefulness, compared with vehicle ( Figure 6-figure supplement 1).

Discussion
This work constitutes the first investigation of striatal A 2A R neuron contributions to regulation of sleep-wake behavior. We showed that activation of A 2A R neurons in the rostral and central, but not caudal striatum, promotes NREM sleep during active period. The topographical study revealed that striatal A 2A R neurons in the rostral, central, and caudal striatum, send axons to the rostral, rostral plus caudal, and caudal areas of the GPe, respectively. The pathway from striatal A 2A R neurons to GPe PV neurons was found to be responsible for sleep control by striatal A 2A R neurons. It is worth to note that inhibition of striatal A 2A R neurons induces a decrease in sleep during active period, indicating that striatal A 2A R neurons are necessary for sustaining sleep during active period.

Striatal A 2A R neuron/GPe PV neuron pathway in sleep regulation
The present study showed that activation of striatal A 2A R neurons promotes NREM sleep via the GPe. Mallet et al. (2012) identified two major populations of GPe neurons, 'prototypic' and 'arkypallidal' neurons. Most (93%) prototypic neurons expressing PV are fast firing (~50 Hz) compared to arkypallidal neurons expressing FoxP2 (~10 Hz) in the GPe (Abdi et al., 2015;Dodson et al., 2015). They both fire at a higher rate during active period when compared to slow wave sleep state, thus are wake active (Abdi et al., 2015). Using immunoelectron microscopy as well as a combination of optogenetic stimulation and patch-clamp recording, we found that striatal A 2A R neurons preferentially form inhibitory synapses with PV-positive neurons (prototypic neurons) in the GPe, suggesting that activation of A 2A R neurons in the striatum inhibits PV neurons in the GPe to induce NREM sleep.
In the present study, inhibition of GPe PV neurons mimicked the effect of activation of A 2A R neurons in the striatum with an increase in NREM sleep. Most importantly, specific lesions of GPe PV neurons abolished the increase in NREM sleep caused by activation of striatal A 2A R neurons. Therefore, A 2A R neurons control sleep behavior by innervating PV neurons of the GPe, which indicates the importance of the striatal A 2A R neuron/GPe PV neuron pathway in sleep induction. However, whether inhibition of striatal A 2A R neurons induced wakefulness also through PV neurons, remains to be investigated in the future.
The mechanism through which PV neurons in the GPe regulate sleep-wake behavior remains unknown. Although the majority of synaptic outputs of GPe PV neurons target the subthalamic nucleus (STN), lesions of the STN have a minimal effect on sleep-wake patterns (Qiu et al., 2010), suggesting that PV neurons in the GPe may communicate with other structures in the brain to influence sleep-wake behavior. Recent studies have shown direct GABAergic projections from the GPe to GABAergic interneurons and, to a lesser extent, to pyramidal cells in the cerebral cortex Saunders et al., 2015). Therefore, inhibition of GPe neurons may disinhibit GABAergic interneurons in the cortex to suppress firing of pyramidal cells and promote sleep. Furthermore, it has been reported that GABAergic neurons in the GPe send axons to the thalamic reticular nucleus (TRN) (Gandia et al., 1993;Mastro et al., 2014) and regulate spiking rates of neurons in the TRN (Villalobos et al., 2016). Given that direct activation of GABAergic neurons in the TRN increases the amount of NREM sleep (Herrera et al., 2016;Ni et al., 2016), we propose that neurons in the GPe, probably PV-positive neurons, may regulate sleep-wake behavior by influencing activity of TRN neurons. Taken together, these results suggest that neural pathways from striatal A 2A R neurons to GPe PV neurons and then to cortical interneurons or TRN neurons may be responsible for sleep-wake regulation.

Figure 5 continued
The following source data is available for figure 5: Source data 1. Sample size (n), mean and SEM are presented for the data in Figure 5. DOI: https://doi.org/10.7554/eLife.29055.023

Subregion-specific diversity of striatal A 2A R neurons in sleep regulation
Subregion-specific diversity of the striatum is not completely understood. Increasing evidence has shown functional heterogeneity along a medial-lateral axis in the striatum in goal-directed action, habit formation, and motor learning (Durieux et al., 2012;Li et al., 2016;Rothwell et al., 2015;Vicente et al., 2016); however, little is known about variation along the rostral-caudal axis. A recent study reported that methyl-CpG-binding protein two in the rostral but not caudal striatum, is critical for maintaining local dopamine content and psychomotor control (Su et al., 2015). In the present study, activation of A 2A R neurons in the centromedial or centrolateral striatum induced similar increases in the amount of NREM sleep, suggesting an equal contribution to sleep regulation along the medial-lateral axis in the striatum. However, manipulating A 2A R neurons in the rostral but not caudal striatum changed the sleep-wake state, strongly suggesting a functional heterogeneity in sleep regulation along the rostral-caudal axis of the striatum.
The mechanisms underlying the functional discrepancy in sleep regulation along the rostral-caudal axis of the striatum remain to be determined. Only one previous study using nonspecific tract tracing in rats showed that the projection of all neurons in the rostral and central striatum is arranged into separate zones of globus pallidus (Wilson and Phelan, 1982). On account of limitations of the nonspecific method as well as the complexity of striatal subregions and striatal neurons, specific tracing using Cre provides a powerful tool to understand topographical projections of A 2A R neurons in different subregions, including the rostral, centromedial, centrolateral, and caudal striatum. We revealed the topographical projection of striatal A 2A R neurons, in which A 2A R neurons in the rostral, central, and caudal striatum send axons to the rostral, rostral plus caudal, and caudal areas of the GPe, respectively. Combined with the above results from synaptology, we conclude that only A 2A R neurons in the rostral and central striatum project to the rostral GPe and connect primarily with PV neurons through inhibitory synapses to control sleep behavior. Recently, reports form clinical studies have indicated variation in subregions of the striatum in neurodegenerative diseases, such as PD, and psychiatric disorders, such as bipolar disease (Altinay et al., 2016;Chou et al., 2015;Jung et al., 2014). Our results provide experimental evidence suggesting subregion-specific targets such as the caudate nucleus of human, which is considered equivalent to the rostral striatum of rodents (Stoffers et al., 2014;Su et al., 2015), for therapeutic intervention of sleep disturbances in clinical cases.

Necessity of striatal A 2A R neurons for normal sleep at active period
Homeostatic drive is a major sleep regulating factor. Adenosine, which is released as a neuromodulator in the brain, has been proposed to act as one of the most potent endogenous somnogens to accumulate in the brain during wakefulness and promote physiological sleep through activation of adenosine A 1 Rs or A 2A Rs (Basheer et al., 2004;Huang et al., 2005;Huang et al., 2014). Among adenosine receptors, A 2A Rs play a predominant role in sleep induction, whereas A 1 Rs contribute to sleep induction in a region-dependent manner (Huang et al., 2014). Moreover, A 2A Rs are present at high concentration in the striatopallidal neurons of the striatum. Evidence has shown that the extracellular adenosine accumulates in the striatum during the active period in rats (Huston et al., 1996). Therefore, we predicted that striatopallidal neurons expressing A 2A Rs may be important in sleep induction of rodents. In the present study, activation of A 2A R neurons in the striatum induced NREM sleep without any significant change in EEG delta power, suggesting that the induced sleep was similar to physiological sleep.
Interestingly, and somewhat more surprisingly, inhibition of A 2A R neurons only decreased NREM sleep in active period, but did not alter sleep-wake profiles during the inactive period. During the active period, levels of extracellular adenosine increase in the striatum. Adenosine then acts on excitatory A 2A Rs expressed on A 2A R neurons and increases activation of striatopallidal neurons to produce sleep. However, during the inactive period, A 2A R neurons show low activity because of low levels of extracellular adenosine which results from reduced metabolism of brain tissue and enhanced removal of metabolic products compared with the active period (Huston et al., 1996;Xie et al., 2013). Thus, chemogenetic inhibition could not further suppress activity of striatal A 2A R neurons that had been in a state of very low activity during the inactive period. This may explain why the sleep-wake profile was not changed following chemogenetic inhibition of A 2A R neurons during the inactive period. In addition, we found that inhibition of A 2A R neurons increased wake without eliciting a homeostatic sleep response during the active period in mice. Although a homeostatic rebound of sleep following sleep deprivation is a widely accepted phenomenon, our data suggested that increased wake during normal active period is not enough to induce sleep rebound. Together, these findings clearly reveal for the first time that activated A 2A R neurons in the striatum are essential for maintaining a certain amount of sleep during the active period and emphasize the importance of striatal A 2A R neurons in physiological mechanism of sleep regulation.

New strategies for treatment of sleep disorders
The present results get access to insight into sleep regulation by striatal A 2A R neurons as well as the therapeutic value of striatal A 2A R neurons for sleep disturbances related to striatal dysfunction, such as excessive daytime sleepiness (EDS) in PD. The main pathological characteristic of PD is loss of dopaminergic neurons in the SNc, which sends dense dopaminergic innervation to the striatum. In PD patients, EDS has been reported to be a frequent sleep disturbance (Tholfsen et al., 2015), but the neuronal mechanisms remain to be elucidated. In humans, Schulz and Falkenburger (2004) reported that the degree of dopaminergic terminal loss in the striatum appears to be more pronounced than the magnitude of SNc dopaminergic neuron loss. In a PD animal model, Gerfen et al. (1990) demonstrated that mRNA encoding D 2 Rs increased in striatopallidal neurons. D 2 Rs are localized primarily in dendritic spines of striatopallidal neurons and form heteromers with A 2A Rs (Ferré et al., 2007). Consistent with D 2 Rs, levels of A 2A Rs also increased in the caudate and putamen of postmortem PD patients (Villar-Menéndez et al., 2014). Other studies have reported that A 2A Rs in the striatum are over-activated in the PD brain, which may lead to increased activation of striatopallidal neurons (A 2A R neurons) in the striatum (Gerfen et al., 1990;Mitchell et al., 1989), and contribute to EDS. Our findings also suggest that EDS in PD patients may be attributed to overactivation of A 2A Rs in the striatum. Therefore, pharmacological antagonism of A 2A Rs may be an effective way to treat EDS using a specific A 2A R antagonist or a non-specific antagonist such as caffeine Huang et al., 2005;Rodrigues et al., 2016).
In summary, our findings demonstrate the importance of striatal A 2A R neurons in controlling sleep behavior and reveal subregion diversity underlying this process. Understanding the circuitry through which striatal A 2A R neurons contribute to sleep regulation provides insight into the striatal A 2A R neuron/GPe PV neuron pathway for sleep regulation and suggests a potential treatment strategy to ameliorate sleep disturbances.
Three weeks after injections, mice used for in vivo studies were chronically implanted with EEG and EMG electrodes for polysomnographic recordings under chloral hydrate (5% in saline, 720 mg/ kg) anesthesia. The EEG electrode implant consisted of 2 stainless steel screws (1 mm diameter) inserted through the skull into the cortex (anteroposterior, À1.0 mm and left/right, À1.5 mm from bregma). The EMG electrodes consisted of two insulated stainless steel, Teflon-coated wires that were bilaterally placed into both trapezius muscles. All electrodes were attached to a microconnector and fixed to the skull with dental cement (Qu et al., 2010). For optogenetic stimulation, the fiber optic cannula (200 mm diameter; Newton Inc., Hangzhou, China) was placed in the GPe: 0.4 mm posterior and 2.2 mm lateral to bregma, 3.3 mm deep from the skull and fixed on the skull using dental cement. The scalp wound was closed with surgical sutures, and each mouse was kept in a warm environment until it resumed normal activity as previously described (Qu et al., 2010).

Sleep-wake monitoring
After the surgical procedures, animals were allowed to recover in individual chambers for at least 7d. Then, each animal was transferred to an insulated sound-proofed recording chamber and connected to an EEG/EMG headstage. The recording cable was connected to a slip-ring device (CFS-22) so the mice could move freely in their cages without tangling the cable. Mice were habituated to the recording cable for 3-4 d before polygraphic recordings. For chemogenetics, EEG/EMG signals mice were recorded over a 24 hr baseline period. This was followed by injection of vehicle (saline, i. p.) at 7:00 p.m. (active period) or 9:00 a.m. (inactive period). Administration of CNO (1 mg/kg in saline, LKT lab, Saint Paul, USA) was performed 24 hr after vehicle injection, and EEG/EMG signals were recorded over the next 24 hr.

Sleep scoring and analysis
EEG and EMG signals were amplified and filtered (EEG,EMG,, digitized at a sampling rate of 128 Hz, and recorded using SleepSign for Animal (Kissei Comtec) (Huang et al., 2005). When completed, SleepSign was used to automatically scored polygraphic recordings off-line (10 s epochs for chemogenetics and 4 s epochs for optogenetics) into the 3 stages of wakefulness, REM sleep, and NREM sleep with the assistance of spectral analysis using fast Fourier transform (Huang et al., 2005;Qu et al., 2010). Briefly, NREM sleep was identified by a preponderance of high-amplitude, low frequency (<4 Hz) EEG activity and relatively low and unchanging EMG activity, whereas wakefulness was characterized by a preponderance of low-amplitude, fast EEG activity and highly variable muscle tone on EMG. REM sleep was identified by very low EMG activity and a low amplitude monotonous EEG containing a predominance of theta range (6-9 Hz) EEG activity. As a final step, defined sleep-wake stages were examined visually, and corrected, if necessary. The latency to NREM sleep are defined as the time between the end of the injection and the onset of the first NREM sleep episode lasting >20 s (Ulbrich et al., 2013). Scoring was done before histological examination and so the scorers were unaware of the extent of the receptor expression. The amount of time spent in wake, NREM, and REM sleep were determined from the scored data.
EEG power spectra were computed for consecutive 10 s or 4 s epochs within the frequency range of 0.5-25 Hz (0.5 Hz bins for 10 s and 0.25 Hz bins for 4 s), using a fast Fourier transform routine. To analyze EEG frequency bands, relative power bins were summed as follows: delta = 0.5-4 Hz, theta = 6-10 Hz, alpha = 12-14 Hz and beta = 15-25 Hz . The EEG power spectrum data are expressed as relative values to total power of NREM or REM sleep, and wakefulness. Two mice power spectrum data were removed, as there was no NREM and REM sleep following vehicle or CNO injection over 3 hr or 4 hr ( Figure 1H; Figure 6H and I).

Optogenetic stimulation in vivo
The optical fiber cannula was attached to a rotating joint (FRJ_FC-FC, Doric Lenses, Canada) to relieve torque. The joint was connected via a fiber to a 473 nm blue laser diode (Newton Inc., Hangzhou, China). Light pulses were generated through a stimulator (SEM-7103 Nihon Kohden, Japan) and output via an isolator (ss-102J, Nihon Kohden). For 1 hr photostimulation, we used programmed light pulse trains (5 ms pulses at 20 Hz for 50 s with 40 s intervals for 1 hr). Light stimulation was conducted from 9 p.m. to 10 p.m. EEG/EMG recorded during the same period on the previous day served as baseline. Light intensity was tested by a power meter (PM10, Coherent) before each experiment and calibrated to emit 20-30 mW/mm 2 from the tip of the optical fiber cannula.
Spectral analysis-compressed spectral array EEG power spectra of wake as well as NREM and REM epochs were analyzed offline using fast Fourier transformation (256 points, Hanning window, 0-24.5 Hz with 0.5 Hz resolution using SleepSign). Then the EEG power spectra data were converted into a dataset in 10 s epochs disregarded stages. Spectral analysis-compressed spectral array were created for ranges of 0-25 Hz and were 4 hr or 5 hr in length. An amplitude bar graph was simultaneously created as the average of 10 s epochs. Bit maps were exported in TIFF format for generation of figures by MATLAB (The MathWorks Inc., New York, USA) (Litvak et al., 2011).
Next, injection sites were viewed under high and low magnification to discern the regions of mCherry immunoreactivity. Under low magnification, histological sections were sampled and rotated to match the mouse atlas, and outlines containing the injection regions were then drawn and merged using Matlab software. A color map showed regions of overlap, and the number of overlapping regions was displayed in decreasing order with red as the maximum followed by yellow, gray, and blue.
Brain samples containing the striatum and GPe were cut into coronal sections (40 mm thick) with a vibratome (VT1000S, Leica) and collected in 0.05 M PB for hrGFP and PV double staining. The basic immunohistochemical protocols were the same as those described above. Briefly, the sections were placed in 0.05 M PB (pH 7.4) containing 25% (w/v) sucrose and 10% (v/v) glycerol for 1 hr and then freeze-thawed with liquid nitrogen to enhance antibody penetration (Li et al., 2002;Li et al., 2001). The sections were then incubated in rabbit anti-hrGFP antibody (1:3000, Agilent) in 0.05 M PB containing 5% normal donkey serum for 30-36 hr at 4˚C. Next, the sections were incubated with donkey anti-rabbit biotinylated IgG (1:1000, Jackson ImmunoResearch) for 3 hr at RT, followed by incubation in ABC complex for 3 hr at RT. After hrGFP immunoractivity was visualized with DAB (Vector Laboratories) following the same procedure used for light microscopy, sections were incubated in goat anti-PV antibody (1:2000, Swant) containing 5% normal donkey serum for 30-36 hr at 4˚C. The sections were then incubated in donkey anti-goat biotinylated IgG (1:1000, Jackson ImmunoResearch), followed by ABC complex for 3 hr each at RT. PV immunoreactivity was visualized using a chromogen of the Vector VIP substrate kit (Vector Laboratories) (Li et al., 2002;Smiley and Mesulam, 1999).
The double-labeled sections were osmicated with 2% OsO4, dehydrated in a graded series of ethanol, and embedded flat in Epon 12 (Ted Pella, Redding, USA) using embedding capsules (TAAB, Berks, UK). The sections embedded in Epon were observed under a light microscope and areas of the GPe were sampled. Ultrathin sections were cut at a thickness of 70 nm using an ultramicrotome, stained with uranyl acetate and lead citrate, and then examined in a CM-120 transmission electron microscope (Philips, Netherlands).
During recording, slices were submerged in a low-volume (~170 mL) recording chamber and continuously superfused at 2 mL/min with warm (30-32˚C) ACSF containing 25 mM d-(-)À2-amino-5phosphonopentanoic acid (d-APV) and 5 mM NBQX to block NMDA and AMPA receptors, respectively. Recordings were guided using a combination of fluorescence and infrared differential interference contrast (IR-DIC) video microscopy using a fixed stage upright microscope (BX51WI, Olympus) equipped with a water immersion lens (40Â/0.8 W) and an IR-sensitive CCD camera (IR1000, DAGE MTI). Recording electrodes (3-5 MW) were filled with an internal solution consisting of (in mM): 105 potassium gluconate, 30 KCl, 4 ATP-Mg, 10 phosphocreatine, 0.3 EGTA, 0.3 GTP-Na, 10 HEPES (pH 7.3, 295-310 mOsm). The internal solution also contained 0.2% biocytin. Random recordings were obtained from neurons in the striatum expressing mCherry and from neurons in the region of the GPe innervated by A 2A R neurons expressing mCherry. Recordings were conducted in the cellattached or whole-cell configuration using a Multiclamp 700B amplifier (Molecular Devices), a Digidata 1440A interface and Clampex 10.3 software (Molecular Devices). Optical stimulation was delivered to slices via an optical fiber (200 mm core, Thorlabs, Newton, USA) coupled to 470 nm diodepumped solid-state continuous-wave laser system (OEM Laser Systems Salt Lake City, USA). Stimulation consisted of either a single 1 ms pulse or trains of 1 ms pulses delivered at 10 Hz. Output of the laser was <2 mW. Light-evoked responses and the effects of a GABA A receptor antagonist (picrotoxin, PTX, 100 mM) on these responses were recorded at a holding potential of À70 mV. All drugs were dissolved in ACSF. Series resistance (Rs) compensation was not used. Therefore, cells with Rs changes over 25% were discarded.
Thirty-two individual GPe neurons were intracellularly labeled with biocytin. Following whole-cell recordings, slices containing cells injected with biocytin were stored in a 4% paraformaldehyde solution overnight at 4˚C and then rinsed with PBS. To test for expression of PV or FoxP2, brain slices were incubated in goat anti-PV antibody (1:2000, Swant) or goat anti-FoxP2 (1:500, Santa Cruz) containing 3% normal donkey serum (v/v), 0.5% Triton X-100 (v/v) for 48 hr at 4˚C. This was followed by incubation with Alexa Fluor 647-conjugated donkey anti-goat (1:1000; Invitrogen) and Alexa Fluor 488 streptavidin (1:1000, Invitrogen) for 12 hr at RT. The slices were then incubated in PBS containing DAPI (1:3000) for 20 min. Finally, sections were washed in PBS and coverslipped with Fluoromout-G TM .

Statistics
All data subjected to statistical analysis in SPSS 19.0. The hourly amounts of the amounts of time spent in each stage of the sleep-wake profiles after vehicle or CNO injection were compared using two-way repeated-measures ANOVA and two-tailed paired Student's t tests. Two-tailed paired Student's t tests were used to compare (1) histograms of amounts of sleep and wakefulness and the latency to the first NREM sleep after vehicle or CNO injection, (2) the hourly amount of sleep and wakefulness after photostimualtion, (3) the percentage of hrGFP-immunoreactive terminals that formed synapses with PV-positive or PV-negative structures in the rostral or caudal GPe, (4) firing rates of PV-positive and PV-negative neurons in the GPe after baseline, light on, or light off, (5) the number of PV, HuCD, or FoxP2 neurons in the GPe of the control group or lesion group, (6) the effect of vehicle or CNO injection on EEG power density bands (delta, 0.5-4 Hz; theta, 6-10 Hz; alpha, 12-14 Hz; beta, 15-25 Hz). eIPSC amplitude, action potential latency, or firing rate of PV-positive and PV-negative neurons in the GPe after light on, and histograms of amounts of sleep and wakefulness in the control and lesion group were compared using Levene's test followed by independent-samples Student's t tests. Data are expressed as mean ± SEM. Statistical significance was considered with p<0.05.