To test the ability of A_2AR neurons in the different subregions of the striatum to regulate sleep in mice, we bilaterally injected a Cre-dependent synapsin-driven adeno-associated viral (AAV) vector containing excitatory modified muscarinic Gq-coupled hM3Dq receptors (hSyn-DIO-hM3Dq-mCherry-AAV, Figure 1—figure supplement 1A) into the striatum of Adora2a-Cre mice. Within the striatum of Adora2a-Cre mice, hSyn-DIO-hM3Dq-mCherry-AAV caused robust expression of hM3Dq receptors in A_2AR neurons (visualized with immunoreactivity for mCherry) (Figure 1A and E and Figure 1—figure supplement 1B and C). In addition, intraperitoneal (i.p.) injection of the specific hM3Dq ligand, clozapine-N-oxide (CNO, 1 mg/kg), but not vehicle, dramatically drove c-fos expression in hM3Dq-expressing A_2AR neurons (Figure 1—figure supplement 1B and C), which confirmed CNO-induced activation of hM3Dq-expressing A_2AR neurons in vivo. Moreover, whole-cell current-clamp recordings showed that a brief CNO (5 μM) perfusion induced depolarization and firing of hM3Dq-expressing A_2AR neurons in the striatum and the effect was reversible following CNO washout (Figure 1—figure supplement 1D). In addition, the hSyn-DIO-mCherry-AAV was injected into the striatum of Adora2a-Cre mice, to only express mCherry but not hM3Dq receptors in the A_2AR neurons. CNO application had no effect on membrane potential and firing of non-hM3Dq-expressing neurons in the striatum in vitro (data not shown) and sleep-wake profiles of non-hM3Dq-expressing mice in vivo (Figure 1—figure supplement 2). The abovementioned results confirmed CNO-mediated activation of striatal neurons only expressing hM3Dq receptors.

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_2AR 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 post-injection 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_2AR 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_2AR 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).

Figure 1 with 5 supplements see all

Download asset Open asset

Figure 1—source data 1

Sample size (n), mean and SEM are presented for the data in Figure 1.

https://doi.org/10.7554/eLife.29055.008

Download elife-29055-fig1-data1-v2.xlsx

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 1F-K and Figure 1—figure supplement 3D–H and 4D–H). These results indicate that A_2AR neurons in the rostral, centromedial, and centrolateral but not caudal striatum are involved in sleep-wake regulation.

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_2AR 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_2AR 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_2AR 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_2AR neurons in different subregions of the striatum are organized topographically in the GPe and suggest that the topographical projections of A_2AR neurons may contribute to the discrepancies in A_2AR neuron-mediated sleep.

Figure 2

Download asset Open asset

Figure 2—source data 1

Sample size (n), mean and SEM are presented for the data in Figure 2.

https://doi.org/10.7554/eLife.29055.014

Download elife-29055-fig2-data1-v2.xlsx

Neurons within the GPe can be divided into PV-positive and PV-negative neurons (Dodson et al., 2015). To examine synapses between axon terminals of A_2AR 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_2AR 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_2AR 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_2AR neurons in the rostral striatum preferentially form symmetric synapses with PV-positive neurons in the rostral GPe, while A_2AR neurons in the caudal striatum preferentially form synapses with PV-negative neurons in the caudal GPe, indicating a rostral-caudal variation in striatopallidal connections.

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_2AR 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 presumably A_2AR neurons exhibiting typical morphological and electrophysiological properties of striatal projecting neurons (Figure 3—figure supplement 1B–H). Photostimulation of cell bodies of A_2AR neurons elicited robust photocurrents (Figure 3—figure supplement 1I), and trains of brief blue light flashes (1–3 ms) evoked single action potentials at frequencies of 5–30 Hz with high fidelity (Figure 3—figure supplement 1J–M).

Figure 3 with 2 supplements see all

Download asset Open asset

Figure 3—source data 1

Sample size (n), mean and SEM are presented for the data in Figure 3.

https://doi.org/10.7554/eLife.29055.019

Download elife-29055-fig3-data1-v2.xlsx

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_2AR neurons. Previous studies have demonstrated two non-overlapping 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_2AR neurons and PV-positive or PV-negative neurons. In addition, the light-evoked IPSCs were completely abolished by picrotoxin (PTX, 100 μM; Figure 3F and K), indicating that these responses were mediated by GABA released from axon terminals of A_2AR 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_2AR 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 PV-negative neurons recorded in current condition are consistent with previous studies (Figure 3O) (Dodson et al., 2015). Taken together, these data support anatomical studies indicating that striatal A_2AR neurons preferentially innervate and inhibit PV neurons in the rostral GPe.

In addition, we examined the effects of photostimulation of ChR2-expressing A_2AR 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_2AR neuron terminals, remarkably increased NREM sleep by 1.8-fold (Figure 3—figure supplement 2), strongly suggesting that striatal A_2AR neurons promoted sleep by inhibiting neurons, more likely PV neurons, in the GPe.

PV neurons in the GPe serve as an important downstream target for striatal A_2AR 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 μM) 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-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_2AR neurons and confirm PV neurons of the GPe as a critical downstream target for striatal A_2AR neuron-mediated sleep.

Figure 4

Download asset Open asset

Figure 4—source data 1

Sample size (n), mean and SEM are presented for the data in Figure 4.

https://doi.org/10.7554/eLife.29055.021

Download elife-29055-fig4-data1-v2.xlsx

To test whether A_2AR 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_2AR 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_2AR 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) (Zhang et al., 2016). 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 (Dodson et al., 2015) 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_2AR 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).

Figure 5

Download asset Open asset

Figure 5—source data 1

Sample size (n), mean and SEM are presented for the data in Figure 5.

https://doi.org/10.7554/eLife.29055.023

Download elife-29055-fig5-data1-v2.xlsx

It is very important to note that the PV-lesion mice also expressing hM3Dq receptors in the striatal A_2AR 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_2AR neurons in the striatum, indicating that the striatal A_2AR neurons control sleep behavior via striatal A_2AR neurons/GPe PV neurons.

We demonstrated that chemogenetic activation of A_2AR neurons in the rostral, centromedial and centrolateral but not caudal striatum promoted NREM sleep. To examine whether striatal A_2AR 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_2AR neurons. Whole-cell current-clamp recordings of A_2AR neurons expressing hM4Di receptors showed decreased firing in response to 350 pA current injection during CNO (5 μM) 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). Therefore, the present results clearly indicate that A_2AR neurons in the striatum play a crucial role in maintenance of normal sleep during the active period.

Figure 6 with 1 supplement see all

Download asset Open asset

Figure 6—source data 1

Sample size (n), mean and SEM are presented for the data in Figure 6.

https://doi.org/10.7554/eLife.29055.027

Download elife-29055-fig6-data1-v2.xlsx

The present study showed that activation of striatal A_2AR 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_2AR neurons preferentially form inhibitory synapses with PV-positive neurons (prototypic neurons) in the GPe, suggesting that activation of A_2AR 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_2AR 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_2AR neurons. Therefore, A_2AR neurons control sleep behavior by innervating PV neurons of the GPe, which indicates the importance of the striatal A_2AR neuron/GPe PV neuron pathway in sleep induction. However, whether inhibition of striatal A_2AR 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 (Chen et al., 2015; 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_2AR neurons to GPe PV neurons and then to cortical interneurons or TRN neurons may be responsible for sleep-wake 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_2AR 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_2AR 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_2AR neurons in different subregions, including the rostral, centromedial, centrolateral, and caudal striatum. We revealed the topographical projection of striatal A_2AR neurons, in which A_2AR 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_2AR 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.

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 A₁Rs or A_2ARs (Basheer et al., 2004; Huang et al., 2005; Huang et al., 2014). Among adenosine receptors, A_2ARs play a predominant role in sleep induction, whereas A₁Rs contribute to sleep induction in a region-dependent manner (Huang et al., 2014). Moreover, A_2ARs 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_2ARs may be important in sleep induction of rodents. In the present study, activation of A_2AR 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_2AR 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_2ARs expressed on A_2AR neurons and increases activation of striatopallidal neurons to produce sleep. However, during the inactive period, A_2AR 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_2AR 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_2AR neurons during the inactive period. In addition, we found that inhibition of A_2AR 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_2AR neurons in the striatum are essential for maintaining a certain amount of sleep during the active period and emphasize the importance of striatal A_2AR neurons in physiological mechanism of sleep regulation.

The present results get access to insight into sleep regulation by striatal A_2AR neurons as well as the therapeutic value of striatal A_2AR 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 D₂Rs increased in striatopallidal neurons. D₂Rs are localized primarily in dendritic spines of striatopallidal neurons and form heteromers with A_2ARs (Ferré et al., 2007). Consistent with D₂Rs, levels of A_2ARs also increased in the caudate and putamen of postmortem PD patients (Villar-Menéndez et al., 2014). Other studies have reported that A_2ARs in the striatum are over-activated in the PD brain, which may lead to increased activation of striatopallidal neurons (A_2AR 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 over-activation of A_2ARs in the striatum. Therefore, pharmacological antagonism of A_2ARs may be an effective way to treat EDS using a specific A_2AR antagonist or a non-specific antagonist such as caffeine (Chen et al., 2013; Huang et al., 2005; Rodrigues et al., 2016).

In summary, our findings demonstrate the importance of striatal A_2AR neurons in controlling sleep behavior and reveal subregion diversity underlying this process. Understanding the circuitry through which striatal A_2AR neurons contribute to sleep regulation provides insight into the striatal A_2AR neuron/GPe PV neuron pathway for sleep regulation and suggests a potential treatment strategy to ameliorate sleep disturbances.

Pathogen-free adult male (8–10 weeks, 24–28 g) Adora2a-Cre mice (Durieux et al., 2009) (Tg(Adora2a-cre)2MDkde, RRID:MGI:3852493) on a C57BL/6J background and Pvalb-Cre mice (B6; 129P2-Pvalb^{tm1(cre)Arbr/J}, PV-Cre, RRID:IMSR_JAX:008069) on a 129 background were used (Do et al., 2016). Adora2a-Cre mice expressing Cre recombinase under the control of adenosine A_2A receptor gene (Adora2a) promoter were provided by Dr. Serge N. Schiffmann (Université Libre de Bruxelles, Brussels, Belgium) and Pvalb-Cre mice expressing Cre recombinase under the control of parvalbumin gene (Pvalb) promoter were provided by Dr. Miao He (Fudan University, Shanghai, China). We crossed Adora2a-Cre mice with Pvalb-Cre mice to generate Adora2a/Pvalb-Cre mice expressing Cre in A_2AR neurons and PV-positive neurons. The animals were housed in individual cages at constant temperature (22 ± 0.5°C) and relative humidity (60 ± 2%) on an automatically controlled 12:12 light/dark cycle (light on at 7 a.m.) with free access to food and water.

Naïve mice were anesthetized with chloral hydrate (5% in saline, 720 mg/kg) and placed in a stereotaxic apparatus (RWD Life Science, Shenzhen, China). To selectively express Cre-dependent hM3Dq receptors (for sleep-wake monitoring) or hrGFP (for anterograde tract tracing) in A_2AR neurons of different subregions in the striatum, hSyn-DIO-hM3Dq-mCherry-AAV, hSyn-DIO-mCherry-AAV or CMV-lox-stop-hrGFP-AAV was delivered bilaterally into the rostral, centromedial, centrolateral, or caudal area of the striatum (coordinates: AP = +1.2 mm, L = ±1.5 mm, DV = −2.4 mm; AP = +0.6 mm, L = ±1.5 mm, DV = −2.4 mm; AP = +0.6 mm, L = ±2.3 mm, DV = −2.4 mm; AP = −0.2 mm, L = ±2.5 mm, DV = −2.4 mm, respectively, as per the mouse atlas of Paxinos and Keith [George and Franklin, 2001]) of Adora2a-Cre mice. To selectively express Cre-dependent hM4Di receptors (for sleep-wake monitoring) in A_2AR neurons of the rostral and central striatum, hSyn-DIO-hM4Di-mCherry-AAV was delivered bilaterally into the rostral and central striatum (coordinates: AP = +1.2 mm, L = ±1.5 mm, DV = −2.4 mm; AP = +0.7 mm, L = ±1.6 mm, DV = −2.4 mm; AP = +0.3 mm, L = ±2.3 mm, DV = −2.4 mm) of Adora2a-Cre mice. Another group of Adora2a-Cre mice was microinjected into the striatum with hSyn-DIO-ChR2-mCherry-AAV to selectively express Cre-dependent ChR2. hSyn-DIO-hM4Di-mCherry-AAV was microinjected into the GPe (coordinates: AP = −0.3 mm, L = ±2.5 mm, DV = −3.0 mm) of Pvalb-Cre mice for sleep-wake monitoring. hSyn-DIO-hM3Dq-mCherry-AAV was injected into the striatum (coordinates: AP = +1.2 mm, L = ±1.5 mm, DV = −2.4 mm; AP = +0.8 mm, L = ±1.8 mm, DV = −2.4 mm), and Flex-taCasp3-TEVp-AAV or DIO-eGFP-AAV was injected into the GPe of Adora2a/Pvalb-Cre mice for sleep-wake monitoring (Zhang et al., 2016). Injections of the viral vector (100–200 nL) were performed using a compressed air delivery system as previously described (Chen et al., 2016).

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, −1.0 mm and left/right, −1.5 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 μm 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).

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.

EEG and EMG signals were amplified and filtered (EEG, 0.5–25 Hz; EMG, 20–200 Hz), 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 (Chen et al., 2016). 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).

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/mm² from the tip of the optical fiber cannula.

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).

Animals received CNO (1 mg/kg, 7 p.m.) and were killed 2 hr later by deep anesthesia with chloral hydrate (10%, 360 mg/kg). This was followed by transcardial perfusion with 10 mL saline, and then 100 mL of 4% paraformaldehyde in 0.1 M phosphate buffer (PB). The brains were removed, postfixed for 4–6 hr at 4°C, and then cryoprotected in 10%, 20%, and 30% sucrose in 0.1 M PB at 4°C until they sank. Tissues were embedded in OCT compound, and stored at −70°C before use (Li et al., 2001). The brains were coronally cut at a thickness of 30 μm on a cryostat (Leica 1950) in four series and were collected in 0.01 M phosphate-buffered saline (PBS, pH 7.4). The floating sections were double immunostained according to the following series of incubation steps (Anaclet et al., 2014): (1) 1:10000 diluted rabbit polyclonal anti-c-fos antibody (catalog number: ABE457, Millipore, RRID:AB_2631318) containing 3% normal donkey serum (v/v) and 0.25% Triton X-100 (v/v) for overnight at 4°C; (2) donkey anti-rabbit biotinylated IgG (1:1000, Jackson ImmunoResearch) for 2 hr at room temperature (RT); (3) ABC complex (Vectorlab, USA) for 2 hr at RT; (4) 3, 3’ diaminobenzidine (DAB, 0.2 mg/ml) and hydrogen peroxide (0.005%) in 0.05 M Tris buffer containing NiCl according to the manufacturer’s instructions (Vector Lab); (5) 1:8000 diluted rabbit polyclonal anti-mCherry antibody (catalog number: 632496, Clonetech, RRID:AB_10013483) containing 3% normal donkey serum (v/v) and 0.25% Triton X-100 (v/v) for overnight at 4°C; (6) donkey anti-rabbit biotinylated IgG (1:1000, Jackson ImmunoResearch) for 2 hr at RT; (7) ABC complex for 2 hr at RT; 8) DAB (0.2 mg/ml) and hydrogen peroxide (0.005%) in 0.05 M Tris buffer without NiCl. Sections were then mounted onto slides, dehydrated, and coverslipped.

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.

The floating sections from GPe lesion and control mice were incubated overnight at 4°C in PBS containing 3–5% normal donkey serum (v/v), 0.25% Triton X-100 (v/v) and primary antibodies that were as follows: rabbit anti-eGFP (1:2000, catalog number: A-11122, Life Technologies, RRID:AB_221569), goat anti-PV (1:3000, catalog number: PVG213, Swant, RRID:AB_2650496), mouse anti- human neuronal protein HuC/HuD (HuCD) (1:500, catalog number: A-21271, Life Technologies, RRID:AB_221448), and goat anti-forkhead box protein P2 (FoxP2) (1:500, catalog number: sc-21069, Santa Cruz, RRID:AB_2107124). After several washes in PBS, the sections were incubated with the appropriate secondary antibody (biotinylated IgG) and ABC complex, followed by DAB (0.2 mg/mL) staining. Sections were then mounted onto slides, dehydrated and coverslipped. Image from the GPe were acquired at 10 × on an Olympus IX71 microscope (Olympus, Japan). The numbers of PV, HuCD and FoxP2 immunoreactive neurons in the GPe (AP = −0.2 ~ −0.5 mm from bregma) were counted offline using Image-Pro Plus. The average numbers of PV, HuCD, and FoxP2 immunoreactive neurons in each hemisphere were then analyzed statistically as described below.

To map topographical projections of striatal A_2AR neurons, Adora2a-Cre mice, injected with CMV-lox-stop-hrGFP-AAV, were anesthetized with chloral hydrate (10%, 360 mg/kg). Brain tissues were prepared as described above. Sagittal or coronal sections of the striatum and GPe were incubated overnight at 4°C in PBS containing 3–5% normal donkey serum (v/v), 0.25% Triton X-100 (v/v), and a mixture of the following primary antibodies: rabbit anti-hrGFP (1:4000, catalog number: 240142, Agilent, RRID:AB_10644103), goat anti-A_2AR (1:1000, catalog number: sc-7504, Santa Cruz, RRID:AB_2273960), and rabbit anti-mCherry (1:8000, Clontech), goat anti-PV (1:3000, Swant). After several washes in PBS, the sections were incubated with Alexa Fluor-conjugated IgG antibody (Invitrogen) at RT for 2 hr. The sections were then incubated in PBS containing DAPI (1:3000, Sigma-Aldrich) for 10 min. Finally, sections were washed in PBS and coverslipped with Fluoromout-G^TM (Southern Biotech). Control staining with 0.01 M PBS instead of the primary antisera indicated a complete lack of non-specific staining regardless of the detection methods used. Fluorescence images were collected using a Leica confocal system or Olympus IX71 microscope.

Under deep anesthesia with chloral hydrate (10%, 360 mg/kg), Adora2a-Cre mice injected with AAV-CMV-lox-stop-hrGFP, were briefly perfused transcardially with 6–8 mL saline followed by 100 mL ice cold fixative containing 4% paraformaldehyde, 0.5% glutaraldehyde, and 15% saturated picric acid in 0.1 M PB (pH 7.4). Immediately after perfusion-fixation, brains were removed and postfixed in 4% paraformaldehyde for an additional 2 hr at 4°C.

Brain samples containing the striatum and GPe were cut into coronal sections (40 μm 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).

Three weeks after injection of hSyn-DIO-ChR2-mCherry-AAV in Adora2a-Cre mice, slices from the striatum or GPe were prepared for electrophysiological recordings. Mice were anesthetized with chloral hydrate and transcardially perfused with ice-cold cutting artificial cerebrospinal fluid (ACSF) containing (in mM): 213 sucrose, 2.5 KCl, 1.25 NaH₂PO₄, 26 NaHCO₃, 10 glucose, 2 Na-Pyruvate, 0.4 ascorbic acid, 3 MgSO₄, 0.1 CaCl₂ (pH 7.3 when carbogenated with 95% O₂ and 5% CO₂). The brains were quickly removed and cut in coronal slices (300 μm thick) in ice-cold cutting ACSF using a vibrating microtome (VT 1200S, Leica). Slices containing the striatum or GPe were transferred to recording ACSF containing (in mM): 126 NaCl, 2.5 KCl, 1.25 NaH₂PO₄, 1 MgSO₄, 2 CaCl₂, 26 NaHCO₃, and 25 Glucose. Slices were incubated at 32°C for 30 min and subsequently maintained at RT for 30 min before recording.

During recording, slices were submerged in a low-volume (~170 μL) recording chamber and continuously superfused at 2 mL/min with warm (30–32°C) ACSF containing 25 μM d-(-)−2-amino-5-phosphonopentanoic acid (d-APV) and 5 μM 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 (40×/0.8 W) and an IR-sensitive CCD camera (IR1000, DAGE MTI). Recording electrodes (3–5 MΩ) 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_2AR neurons expressing mCherry. Recordings were conducted in the cell-attached 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 μm core, Thorlabs, Newton, USA) coupled to 470 nm diode-pumped 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 μM) on these responses were recorded at a holding potential of −70 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.

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.

The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.

We thank Dr. Miao He (Institutes of Brain Science, Fudan University) for kindly supplying us with Pvalb-Cre mice. We thank Dr. Jun Lu of Harvard Medical School for helpful discussions. We also thank Dr. Jin-Lian Li of the Fourth Military Medical University for help with the immunoelectron microscopy.

Animal experimentation: All animal studies were performed in accordance with protocols approved by the Committee on the Ethics of Animal Experiments of Fudan University Shanghai Medical College (permit number: 20110307-049). Every effort was made to minimize the number of animals used and any pain and discomfort experienced by the subjects.

© 2017, Yuan et al.

This article is distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use and redistribution provided that the original author and source are credited.