Astrocyte deletion of Bmal1 alters daily locomotor activity and cognitive functions via GABA signalling

Circadian rhythms are controlled by a network of clock neurons in the central pacemaker, the suprachiasmatic nucleus (SCN). Core clock genes, such as Bmal1, are expressed in SCN neurons and in other brain cells, such as astrocytes. However, the role of astrocytic clock genes in controlling rhythmic behaviour is unknown. Here we show that ablation of Bmal1 in GLAST-positive astrocytes alters circadian locomotor behaviour and cognition in mice. Specifically, deletion of astrocytic Bmal1 has an impact on the neuronal clock through GABA signalling. Importantly, pharmacological modulation of GABAA-receptor signalling completely rescues the behavioural phenotypes. Our results reveal a crucial role of astrocytic Bmal1 for the coordination of neuronal clocks and propose a new cellular target, astrocytes, for neuropharmacology of transient or chronic perturbation of circadian rhythms, where alteration of astrocytic clock genes might contribute to the impairment of the neurobehavioural outputs such as cognition.

A nimals have an internal timekeeping mechanism to anticipate daily changes associated with the transition of day to night 1 , which is deeply involved in the regulation and maintenance of behavioural and physiological processes 2 . In mammals, the circadian system is organized in a hierarchy of multiple oscillators at organism, cellular and molecular level. At the organism level, the suprachiasmatic nucleus (SCN) is the central pacemaker at the top of the hierarchy, which integrates light information to ultimately regulate rhythms in gene expression, physiology and behaviour. At the cellular level, the SCN is composed of multiple oscillating neurons that are coupled to act as a single circadian unit 3 , leading to coordinated circadian signalling outputs. At the molecular level, the circadian clock consists in the transcription-and translation-based interconnected feedback loops, in which the transcription factors BMAL1 and CLOCK drive the expression of Per and Cry genes, whose products lead to the inhibition of their own transcription 1 . Oscillations in abundance of those core clock proteins in the brain and peripheral tissues drive a cascade of transcriptional output genes that are not involved in the timekeeping mechanism itself, but underlie local behavioural and physiological process 4 .
The SCN and other brain regions are composed of a heterogeneous population of cells, including astrocytes, which have a well-documented role in cooperating with presynaptic and postsynaptic neuronal elements to regulate communication events and behavioural processes. Although recent evidence suggests an involvement of astrocytes in the regulation of circadian rhythms in Drosophila 5,6 and in mammals 7,8 , the role of clock genes in these cells has not been investigated.
It is currently unknown whether the regulation of astrocyte physiology by clock genes might contribute to the maintenance of neuronal rhythmic behaviour at cellular, tissue and organism level. Remarkably, SCN astrocytes express transporters for GABA (g-aminobutyric acid), the principal neurotransmitter in the master pacemaker, and by up taking GABA from the extracellular space also synapses in the SCN function as 'tripartite' synapses 9 . Indeed, the circadian release of gliotransmitters, such as ATP 10 and the rhythmic expression of neurotransmitter transporters was also reported 11 . Identifying such a role for astrocyte clock genes will not only reveal a more complex cellular signalling in the brain than that considered so far but would also have significant implications for therapeutic research on transient perturbations in circadian rhythms (such as jet lag), adverse effects of shift workers and in disorders associated with circadian rhythms dysfunctions.
Here we report that astrocytic BMAL1 impacts the neuronal clock by altering GABAergic signalling. Importantly, this leads to altered circadian locomotor behaviour and to severe cognitive defects in mouse.

Results
Efficient deletion of Bmal1 in SCN astrocytes. To evaluate the role of the core clock gene Bmal1 in astrocytes of adult mice in vivo, we generated a conditional Tamoxifen (TM)-inducible knockout mouse model (Bmal1 flx/flx , Glast-CreER T2 þ / À here referred to as Bmal1cKO), where Cre-recombinase is expressed under the control of the glutamate transporter Glutamate Aspartate Transporter (Glast) promoter, a widely accepted astrocyte-specific gene [12][13][14] . It has been previously shown that adult induction of CreER T2 expressed from the Glast locus targets a large subset (60-80%) of astrocytes, corresponding in their frequency to those that endogenously express GLAST in cortex and striatum 12 . In contrast, Cre-mediated recombination in adult Glast:Cre-IRES-hrGFP mice has been shown to occur only in 0.1% of NEUN-positive neurons 12 , thus confirming the astrocyte-specific recombination.
As the inducible form of Cre (CreER T2 ) cannot be localized by immunocytochemistry 12 , to ascertain the Cre-mediated recombination we crossed Glast-CreER t2 mice with a reporter mouse line, in which red fluorescent reporter Td-TOMATO is driven by CAG promoter, on Cre-mediated recombination of loxP sites. Two months after TM treatment, we analysed the Glast-Cre-mediated recombination and immunoreactivity for the astrocyte-specific markers glial fibrillary acidic protein (GFAP) or S100b in the SCN. Glast-Cre-Td-TOMATO reporter expression was intense ventrally and spread in central and dorsal SCN (Fig. 1a). We found that 49.53% and 46.51% of Td-TOMATOpositive cells co-localized with GFAP or with S100b, respectively, confirming that recombination occurred in astrocytes of the SCN ( Fig. 1a and Supplementary Fig. 1a).
To evaluate the efficiency of Bmal1 recombination in SCN astrocytes, we quantified the co-immunolocalization of BMAL1 and Td-TOMATO with GFAP or S100b in Glast-Cre-Td-Tomato (control) or Bmal1cKO-Td-Tomato animals at Zeitgerber (ZT) 0. Control animals expressed BMAL1 in 61.13% of Td-TOMATOpositive cells, whereas the number of Td-TOMATO cells expressing BMAL1 was reduced by B70% in Bmal1cKO-Td-Tomato mice (paired t-test, P ¼ 0.0008; Fig. 1b,c). Similarly, the percentage of GFAP or S100b-positive astrocytes expressing BMAL1 was significantly reduced by 59.33% and 60.24%, respectively, in the SCN of Bmal1cKO-Td-Tomato mice (P ¼ 0.0001 for GFAP and P ¼ 0.0068 for S100b, paired t-test; Fig. 1d,e and Supplementary Fig. 1b).
Surprisingly, in mutant animals, the percentage of BMAL1positive cells was also reduced by 51% in Td-TOMATO-negative cells (P ¼ 0.02, paired t-test; Fig. 1c), suggesting that deletion of Bmal1 in GLAST-positive astrocytes results in a global downregulation of BMAL1 in the SCN. Given the high expression of Glast-Cre-Td-Tomato by SCN astrocytes and the significant reduction of BMAL1 in the SCN of our mutants, we sought to evaluate circadian locomotor activity and cognition, both neurobehavioural outputs under circadian control, in Bmal1cKO mice.
Altered circadian and cognitive phenotype in Bmal1cKO mice. Two months after TM treatment (Fig. 2a), wheel-running activity of control (Bmal1 flx/flx ) 15 and Bmal1cKO animals was used as an index of SCN circadian function 16,17 . Animals were entrained to a schedule of 12-12 h light-dark (LD) cycle for at least 8 days, before being transferred to constant darkness and then, re-entrained to a new 12-12 h LD cycle.
During the LD condition, the locomotor activity of Bmal1cKO mice was indistinguishable from that of control animals (Fig. 2b left panels and Fig. 2c), showing no differences in the periodicity or in the total average activity ( Supplementary Fig. 2).
On release into constant darkness, control animals exhibited a daily rhythm with a period of B24 h (23.95 ± 0.08). Although this locomotor component was also observed in Bmal1cKO mice (24.04 ± 0.09), 71% of the mutants (5 of 7 animals) exhibited an additional activity component with periods of B12 h as shown by the Lomb-Scargle periodograms (Fig. 2b, right panels), suggesting a bimodal pattern of locomotor activity. There were no differences in the average activity or periodicity among genotypes ( Fig. 2c and Supplementary Fig. 2). However, Bmal1cKO mice significantly delayed their active phase (11.70±2.84 min per day) compared with control animals (6.88 ± 1.44 min; Fig. 2d, left panel).
After constant darkness, Bmal1cKO mice did not require a longer time to adjust their activity to a new LD cycle than control ARTICLE NATURE COMMUNICATIONS | DOI: 10.1038/ncomms14336 animals (Fig. 2b). However, Bmal1cKO mice showed a significant advanced onset (Fig. 2d, middle panel). Although mutant animals also showed an activity offset advance (Fig. 2d, right panel), the active period was significantly reduced in these mice (13.  impact on the clock. However, our results reveal that astrocytic BMAL1 is involved in the proper organization of daily locomotor activity in LD cycles. Moreover, the bimodal behaviour of Bmal1cKO mice, which might reflect the output of two circadian oscillators, suggests a potential role of astrocytes in the coupling of SCN oscillators that govern locomotor activities. Perturbations of circadian rhythms in humans, such as in shift workers and resulting from jet lag 18 , or in constitutive knockout mice for clock genes such as Bmal1 À / À , have been associated with cognitive dysfunction 19 . We therefore evaluated a possible impact of astrocyte BMAL1 on cognition in our mutant mice. For this, the novel object recognition (NOR) test was used to assess short-term memory and long-term memory, by separating the sample and test phases by 1 and 24 h, respectively 20 (Fig. 2e). Bmal1cKO mice exhibited a significant reduction in the discrimination index (DI) as compared with control mice after 1 and 24 h, thus indicating the impairment of both shortand long-term memory ( Fig. 2f and Supplementary Fig. 3). In the spatial object location (SOL) task, only control animals but not Bmal1cKO, showed preferential exploration of the novel location, indicating compromised consolidation and object place recognition memory in mutant mice ( Fig. 2f and Supplementary Fig. 3).
Altogether, our results indicate that the selective ablation of Bmal1 in adult astrocytes is sufficient to alter daily locomotor activity and declarative memory in mice. These phenotypes might be dependent on astrocytic BMAL1 functions affecting gene expression in the SCN and/or rhythmic oscillations in cortical and hippocampal circuits involved on memory. Thus, we analysed whether rhythmic gene expression in the SCN, cortex and hippocampus were preserved in Bmal1cKO. Altered rhythmic gene expression in brain of Bmal1cKO mice. Studies in Drosophila showed that the involvement of astrocytes in the regulation of circadian rhythms 5,6 was mediated by a clock neuron peptide transmitter, pigment dispersing factor 6 , which acts on a receptor similar to that for vasoactive intestinal polypeptide (VIP) in mammals. VIP is an oscillating neuropeptide with a light-dependent rhythm 21 that plays a well-defined role in coupling, synchronizing and phase-shifting rhythms within SCN neurons [22][23][24] . We investigated whether VIP expression was affected in the SCN of Bmal1cKO mice, by immunostaining. In particular, we compared the levels of VIP by immunofluorescence in the SCN of control and Bmal1cKO mice at ZT0 (when its expression peaks 21 ) and ZT12 (Fig. 3a). Consistently with previous reports 21 , VIP expression at ZT12 decreased compared with ZT0 in control mice (Fig. 3a). In contrast, VIP expression was not downregulated in the SCN of Bmal1cKO mice at ZT12 (Fig. 3a). This result suggests repression of VIP by astrocytic BMAL1, leading to altered expression of this neuropeptide in mutant mice. Next, we evaluated rhythmic oscillations of core clock genes in the cortex and hippocampus of Bmal1cKO mice. We found rhythmic expression of Bmal1, Cry1, Per2 and BMAL1 target Dbp 25 in control animals, with phases similar to those observed previously in pituitary gland and cortex [26][27][28] . However, these oscillations were impaired in Bmal1cKO mice ( Fig. 3b and Supplementary Fig. 4a,b). As expected, BMAL1 levels were reduced by 30% in the cortex of Bmal1cKO mice and, notably, rhythmic expression of BMAL1 and PER2 dampened in the mutant mice ( Fig. 3c and Supplementary Figs 4c and 5).
In our mice, BMAL1 depletion occurs in the majority of SCN astrocytes (as revealed by immunostaining of Glast-cre-driven  Td-TOMATO with GFAP and S100b ( Fig. 1 and Supplementary  Fig. 1). However, we observed a significant reduction of BMAL1 in Td-TOMATO-negative SCN cells, a population likely to be embracing neuronal cells as well (Fig. 1c). This global reduction of BMAL1, together with the altered VIP expression in the SCN and the disruption of clock genes oscillation in both the cortex and hippocampus in Bmal1cKO mice, would be difficult to explain without considering the presence of intercellular communication between astrocytes and neurons. This scenario would be consistent with previous reports, indicating that astrocytes are competent circadian oscillators with temperaturecompensated period 29 that can modulate clock gene expression of other cell types such as clock neurons or fibroblasts 30 . Thus, we hypothesized that BMAL1 function in astrocytes might be required to entrain circadian rhythmicity in neurons.
Astrocytic BMAL1 is required for neuronal rhythmicity. Impaired oscillations of core clock genes in the cortex and hippocampus of Bmal1cKO mice ( Fig. 3 and Supplementary Figs 4 and 5) might depend on either a primary effect of astrocytic BMAL1 in the SCN, in local autonomous circuits or in both. As it is extremely difficult to discriminate between these possibilities, we investigated whether BMAL1 functions in astrocytes could affect the oscillations of core clock genes in cortical neurons. For that, we first set up a synchronization assay in astrocyte cultures by a short pulse (2 h) of Dexamethasone 31 (Fig. 4a). As expected, this treatment successfully induced rhythmic oscillation of core clock genes such as Cry1, Per2 and the BMAL1 target, Dbp, in control astrocytes (Astro Dexa, Fig. 4a,b). Next, we co-cultured synchronous astrocytes transfected with scramble small interfering RNA (siRNA) (Fig. 4a,b) or arrhythmic astrocytes (on Bmal1 knockdown, Fig. 4a,b) onto asynchronous primary cortical neurons, in physically separated layers (B1.5 mm) but sharing the same culture media. We found that synchronous astrocytes systematically induced rhythmic expression of Bmal1, Cry1, Per2 and Dbp, as well as CRY1 in neurons (Fig. 4b,c and Supplementary Figs 6 and 7). Interestingly, we also found a significant advance in the acrophase of all the transcripts in synchronous astrocytes co-cultured with neurons, in comparison with synchronized astrocytes in isolated cultures ( Fig. 4b and Supplementary Fig. 6a). In contrast, on Bmal1 knockdown, arrhythmic astrocytes failed to synchronize these transcripts and CRY1 in neurons (Fig. 4b,c and Supplementary Figs 6 and 7). Together, our results demonstrate that, by means of exchanged extracellular factor(s), astrocytic BMAL1 is required to entrain rhythmicity in neurons.
We next sought to identify the extracellular factor that mediates astrocyte to neuron communication. It is widely accepted that astrocytes can dynamically regulate neuronal communication via the uptake of neurotransmitters (for example, GABA and glutamate) or by gliotransmitter release (such as D-serin and ATP) 32 . Among those molecules, glutamate is considered to be a major factor in transducing retinal photic information to the SCN via retinohypothalamic projection 33,34 . On the other hand, GABA, the principal neurotransmitter in the master pacemaker, is released in a daily rhythm within the SCN 35 and rhythmic levels were found in different brain regions 36 . Remarkably, SCN astrocytes express transporters for GABA and by uptaking GABA from the extracellular space also synapses in the SCN function as 'tripartite' synapses 9 , where astrocytes can modulate neuronal activity and communication. Moreover, GABAergic signalling plays a role in tonic inhibition of neurons by modulating a continuous current dependent on extrasynaptic GABAA receptors 37 , whose are involved in memory 38 . Furthermore, astrocytic GABAergic signalling was shown to be involved in tonic inhibition in different brain areas 39 . Thereby, we hypothesized that glutamate and/or GABA might be the extracellular factor(s) mediating astrocyte to neuron communication.
To investigate this hypothesis, we tested whether glutamate or GABA can induce rhythmic oscillations of core clock genes in cortical neurons in vitro. We treated primary cortical neurons with either a 2 h pulse of glutamate (10 mM) or a pulse of GABA (100 mM) and we found that GABA, but not glutamate, was sufficient to entrain their rhythmic expression of Bmal1, Cry1, Per2 and Dbp in these cells (Fig. 5a).
Next, to determine whether GABA signalling can mediate astrocyte-neuron communication, we synchronized astrocytes with a short pulse (2 h) of Dexamethasone and co-cultured them with asynchronous cortical neurons (as in Fig. 4a), in the presence of the GABAA receptor blocker Bicuculline (30 mM). We found that the inhibition of GABAA receptor signalling prevents astrocyte-induced entrainment of clock gene oscillations in neurons (Fig. 5b). Remarkably, rhythmic expression of Per2, Cry1 and Dbp was maintained in astrocytes in the presence of Bicuculline ( Supplementary Fig. 8), suggesting that GABAA receptor signalling is required to entrain rhythmic expression of clock genes in neurons, but not to sustain rhythmicity in astrocytes. Together, these results indicate that GABA, through GABAA receptor signalling, mediates astrocyte to neuron communication.
Impaired GABA uptake on Bmal1 deletion in astrocytes. Given that arrhythmic astrocytes fail to synchronize the neuronal clock (Fig. 4), we hypothesized that regulation of extracellular GABA might be impaired on Bmal1 deletion in astrocytes. To test this hypothesis, we analysed the expression of the GABA transporters (Gats), Gat1 and Gat3, which are localized to astrocytes, in the cortex and SCN of control and Bmal1cKO animals.
We found that neither Gat1 or Gat3 transcripts were oscillating in the cortex (Fig. 6a,b, left panels). However, Bmal1cKO mice showed a significant reduction of Gat1 (at ZT0 and ZT6) and Gat3 (at ZT0) as compared with control mice (Fig. 6a,b, left panels). We confirmed these results by immunostaining of GAT1 and GAT3 in cortex of control and mutant animals at ZT0 (Fig. 6a,b right panels and Supplementary Fig. 9a). As GAT1 is also expressed in axon terminals 40 , to identify astrocyte-specific GAT1, this staining was performed in Bmal1cKO-Td-Tomato mice.
The reduced expression of astrocytic GAT1 and GAT3 in the cortex of Bmal1 cKO mice suggests a potential impairment in the clearance of extracellular GABA released by neurons. To verify this hypothesis we performed a GABA uptake assay in arrhythmic astrocytes on Bmal1 knockdown. Control or arrhythmic cortical astrocytes were treated with different doses of GABA and extracellular GABA levels were determined after 15 min by enzyme-linked immunosorbent assay (Fig. 6c). We found that GABA uptake was severely impaired in arrhythmic astrocytes (Fig. 6c), suggesting that astrocytic BMAL1 is required to avoid accumulation of extracellular GABA.
To further confirm this finding in vivo, we quantified GABA levels in the cerebrospinal fluid (CSF) of control and Bmal1cKO animals. Importantly, we found significantly higher GABA levels in the CSF of Bmal1cKO compared with control animals at ZT6 (Fig. 6d). We performed those measurements by pooling together CSF from several mice and following previously described liquid chromatography-tandem mass spectrometry (LC-MS/MS) protocols 41 . Although we could not find the reference values of ARTICLE the CSF GABA levels in mice, perhaps due to the intrinsic difficulty of the sampling in this animal model, our values are in line with those reported for humans (50-100 ng ml À 1 ) and slightly lower than those reported for rats 42 .
We also investigated whether astrocyte deletion of Bmal1 might lead to an alteration of GATs in the central pacemaker. GAT1 is mostly expressed between the lobes of the SCN and around the third ventricle, whereas GAT3 is expressed evenly in the glial processes of SCN astrocytes 9 . Thus, we quantified the expression of GAT3 in the SCN in control and Bmal1cKO animals at ZT0. In control mice, GAT3 had a higher density in the dorsal and ventrolateral part of the SCN (Fig. 6e and Supplementary Fig. 9b). In contrast, this interregional distribution of GAT3 was lost in the SCN of Bmal1cKO mice and we found a significant reduction of GAT3 intensity in the dorsal part and an increase in the ventral SCN ( Fig. 6e and Supplementary Fig. 9b).
All together, our results reveal that GABA can induce rhythmic oscillations of core clock genes in primary cortical neurons. Moreover, we also found that arrhythmic astrocytes, on Bmal1 knockdown, cannot modulate their uptake of GABA. Indeed, Bmal1cKO mice showed reduced expression of GAT1 and GAT3, and elevated GABA levels in the CSF. Increased GABA in CSF of our mutants might not reflect local GABA levels in different areas of the brain. Indeed, the absolute concentrations of GABA in presynaptic cytosol, in vesicles and in the extrasynaptic space are unknown. However, the affinity constants of extrasynaptic GABA receptors may serve as a rough estimate of background concentrations (0.2-2.5 mM) 43 , and in rat, direct measurements from CSF yielded similar or slightly higher values 42 . Thus, the CSF levels of GABA might reflect the concentrations of this neurotransmitter at the extrasynaptic space.
Our results suggest that BMAL1 in astrocytes might play a fundamental role in maintaining extracellular GABA levels and/or rhythms in a range compatible with neuron synchronization. We postulate that this scenario could be involved in the cognitive impairments that we observed in Bmal1cKO mice. In fact, this would also be in agreement with evidence showing that the arrhythmic SCN of the Siberian hamster leads to an over-inhibition of synaptic circuits involved in memory. Indeed, the cognitive deficits of those animals were completely restored after administration of GABAA receptor antagonists or on SCN removal [44][45][46] .
On the other hand, GABA has been shown to transmit phase information between the ventral and dorsal oscillators of the SCN 47 . This evidence, together with the loss of the interregional distribution of GAT3 expression in the SCN of Bmal1cKO mice, suggests that altered GABA-mediated coupling among the SCN oscillators might underlie the bimodal pattern of locomotor activity of our mutants.
GABAergic antagonists rescue behaviour of Bmal1cKO mice. To verify our hypothesis that altered GABA signalling might lead to the circadian locomotor and declarative memory phenotypes of Bmal1cKO mice, we administered GABAA receptor antagonists Pentylenetetrazole (PTZ) or Picrotoxin (PTX) (0.3 mg kg À 1 per day) to control and Bmal1cKO mice at ZT6, for 10 days (Fig. 7a), following previously reported protocols 45,46,48 . We then analysed the circadian locomotor activity and performed again the cognitive tests, as described above (Fig. 2). Consistent with previous reports 49 , PTZ treatment did not alter neither the average activity nor the periodicity of control mice in any of the lighting conditions (LD, dark-dark (DD) and re-entrainment to a new LD cycle; Fig. 7b,c and Supplementary  Fig. 10). Similarly, we observed no significant differences between PTZ-treated and untreated Bmal1cKO mice. Indeed, the periodicity and the average activity was not different among control and Bmal1cKO animals on treatment with PTZ ( Supplementary Fig. 10).
Remarkably, PTZ treatment rescued the bimodal pattern of locomotor activity of Bmal1cKO mice in DD as shown by the Lomb-Scargle periodograms (Fig. 7b, lower panels). This finding suggests that astrocytic BMAL1 couples the oscillators in the SCN through GABAA receptors. In this context, we hypothesize that the loss of the interregional distribution of GAT3 expression in the SCN of Bmal1cKO mice might alter local GABA levels to uncouple the oscillators in the central pacemaker. Moreover, we found that on PTZ treatment, the activity onset of Bmal1cKO in DD was not different from PTZ-treated control animals (Fig. 7d,  left panel), thus restoring the delayed onset of activity observed in Bmal1cKO (11.70 ± 2.84 min per day for Bmal1cKO mice versus 5.38 ± 0.83 min for PTZ-treated Bmal1cKO). Similarly, the advanced activity onset and offset of Bmal1cKO mice in rLD were not observed on PTZ treatment and were not different from the PTZ-treated control animals (Fig. 7d, middle and right panels).
Declarative memory was also analysed in control and Bmal1cKO mice on PTX or PTZ treatment by subjecting animals to the NOR and novel object location tasks. Interestingly, we found that these treatments also re-established cognitive functions in Bmal1cKO mice to normal levels, whereas, similar to previous reports, did not augment the DI in control animals 45,48 ( Fig. 7e and Supplementary Fig. 11). This result implies that mnemonic deficits observed in Bmal1cKO mice arise from specific abnormalities in declarative memory that are rescued by drug effects within the circuits involved, rather than to some nonspecific effects of the drugs.
We postulate that loss of BMAL1 function in astrocytes results in altered GABA levels, leading to the over-inhibition of the circuits involved in learning and memory and to the uncoupling the SCN oscillators. Consistently, administration of GABAA receptor antagonists restores the circadian locomotor activity and the cognitive functions of Bmal1cKO mice.

Discussion
This study is the first demonstration of a role for astrocytic BMAL1 in the modulation of circadian locomotor behaviour and Remarkably, our study promotes Bmal1cKO mice as a valuable in vivo tool to model human pathologies related to alterations in the circadian system. The well-documented adverse effects of human dysrhythmia on cognition and declarative memory 18 stands in stark contrast with the marginal effect of clock gene knockouts and SCN lesions in rodents. One critical factor that often gets overlooked in translating animal studies to human conditions is the fact that human dysrhythmia occurs, while the SCN circuitry remains intact, both genetically and structurally. Thus, our model is well suited for functional studies of the circadian system, because it allows acute adult disruption of astrocytic BMAL1, while avoiding functional abnormalities or compensations that might occur during development. Our results indicate that deletion of Bmal1 in a sub-population of astrocytes is sufficient to affect globally the neuronal clock in the brain. The anatomical properties of astrocytes are crucial to understand this finding. Indeed, cortical and hippocampal astrocytes are organized in structurally non-overlapping domains in vivo [50][51][52] , where one astrocyte interacts with, an average of, four cortical neuron cell bodies and between 300 and 600 dendrites 50 . Moreover, astrocytes form a syncytium via gap junctions that is required for the circadian patterns of neuronal activity in the SCN 8 . Thus, although gap junctions mediate intercellular coupling among astrocytes, our data indicate that astrocytes regulate neuronal clock by a mechanism involving GABA and GABAA receptor signalling.
Although our results show a global reduction of BMAL1 in the SCN, some of the cells that retain BMAL1 expression might underlie the fact that none of the Bmal1cKO mice analysed showed arrhythmic locomotor behaviour. This finding is therefore consistent with different studies showing that partial deletion of Bmal1 (refs 53,54) or partial lesions of the SCN 17,55-59 did not result in loss of rhythms. Whether deletion of Bmal1 in all SCN astrocytes will drive to complete loss of daily locomotor activity is an unresolved question that is difficult to address. Indeed, astrocytes are widely distributed throughout the nervous system and express brain-region-specific genes. Thus, the selection of regulatory elements to target all astrocytes in vivo is almost impossible with today's tools.
However, based on our results, a potential contribution of intrinsic astrocyte clock to the global timekeeping system to finely coordinate autonomic circadian rhythms and associated neurobehavioural outputs cannot be discarded. In fact, in vitro studies indicate that astrocytes are competent circadian oscillators with a temperature-compensated period 29 , which can modulate clock gene expression of other cells types such as clock neurons or fibroblasts 30 . Indeed, the circadian release of gliotransmitters, such as ATP 10 , and the rhythmic expression of neurotransmitter transporters 11 indicate that astrocyte physiology is regulated by circadian rhythms. Thus, together with those findings, our study suggests that astrocyte rhythmic modulation of neurotransmission might be required for the correct organization of the hierarchy of oscillators at the cellular, tissue and organism level. Consistently, the behavioural phenotype of Bmal1cKO mice was completely rescued, on modulation of GABAA-receptor signalling, demonstrating a physiological regulation of neuronal function rather than a neural degeneration-induced phenotype.
Bmal1cKO mice showed impairments in restricting the locomotor activity to day or night when re-entrained to the new LD cycle. As Glast is expressed in the retinal Müller glia, it might be argued that this could be due to a secondary effect of altered retinal sensory inputs. However, this is unlikely as the circadian locomotor activity of Bmal1cKO and control mice is indistinguishable in LD cycles. On the other hand, it is widely accepted that glutamate signalling plays a crucial role in the transduction of retinal photic information to the SCN 33,34 . As in our mouse model Bmal1 is conditionally deleted by Glast-driven Cre in astrocytes, a potential alteration in glutamate signalling might underlie the impaired light-mediated resetting of the circadian clock of Bmal1cKO mice. However, the phenotype of our mutants was restored on GABAA antagonist treatment, suggesting that altered glutamate signalling might not be the prime cause of this phenotype of Bmal1cKO.
Remarkably, astrocytic BMAL1 might have crucial implications for the synchronization of the internal clock. We found a bimodal pattern of activity when Bmal1cKO mice were released in constant darkness, suggesting uncoupled circadian oscillators governing locomotor activities. The loss of interregional distribution of GAT3 in the ventral and dorsal SCN of our mutants suggest that altered GABAaergic signalling might underlie this phenotype. Consistently, GABA has been shown to transmit phase information between the ventral and dorsal oscillators of the SCN 47 . Indeed, we found that GABAA antagonist treatment restores the bimodal locomotor behaviour of Bmal1cKO mice. However, we cannot discard an involvement of VIP on this phenotype, as it was found that administration of this neuropeptide increase the frequency of GABA inhibitory postsynaptic currents in clock neurons 35 . Consistently, hyperpolarized resting membrane potential of SCN cells was found in the vpac 2 À / À / mice 60 . Moreover, it was recently reported that VIP neurons evokes functional GABAergic responses, through GABAA receptor in the SCN 61 . Thus, we postulate that the contribution of astrocyte BMAL1 to the stable phase of the SCN clock is very likely to be a balance between VIP and GABAA receptor-mediated synchronization.
It was reported that the differential day/night glial coverage of VIP neurons dendrites is important to facilitate entrainment to light and to the pacemaker-resetting mechanism 62,63 . The finding that Bmal1 deletion in GLAST-positive astrocytes alters VIP expression in the SCN is consistent with the intense expression of Td-TOMATO reporter in the ventral SCN, where VIP neurons are located. Similarly, studies in Drosophila showed that conditional, glial-specific genetic manipulations affecting membrane (vesicle) trafficking or calcium signalling lead to circadian arrhythmicity 5,6 . This behavioural phenotype was correlated with effects on a clock neuron peptide transmitter such as pigment dispersing factor 6 , acting on a receptor similar to that for VIP in mammals. As VIP rhythm is driven by the LD cycle and not by the circadian clock, the mechanism by which astrocytes deficient in Bmal1 alters VIP levels in our case remains unknown.
The phenotype of our mice might be dependent on static versus oscillating BMAL1 function in astrocytes. Ultimately, the function of clock proteins is never completely disengaged from their oscillation, as the BMAL1/CLOCK DNA binding shows clear circadian variation 64 . However, a major challenge, which holds true for most of the studies involving genetic manipulations of core clock genes, involves distinguishing the specific importance of circadian oscillation versus the 'static' function of clock genes. Despite this fact, our results together with the elegant studies in Drosophila 5,6 not only emphasizes the conservation of cellular and molecular mechanisms that regulate behaviour in mammals and insects but also demonstrate the capacity for astrocyte-to-neuron signalling in the circadian circuitry.
In conclusion, our study reveals a crucial role of astrocytic BMAL1 for the coordination of neuronal clocks and opens new avenues to understand the physiology of the timekeeping system and to treat or prevent disorders associated with them, based on astrocyte targeted drugs.

Methods
Animals and treatments. Mice were housed with ad libitum access to food and water, and kept on a 12 h (07:00-19:00 h) LD cycle, in a room maintained at 21°C at the animal facility of the Istituto Italiano di Tecnologia, Genoa, Italy. All experiments and procedures were approved by the Italian Ministry of Health (Permit Number 214/2015-PR) and local Animal Use Committee, and were conducted in accordance with the Guide for the Care and Use of Laboratory Animals of the European Community Council Directives and of Italian Ministry of Health.
Six to 8 weeks old Glast-CreERT2; Bmal1 flx/flx (Bmal1cKO) and controls (Bmal1 flxflx ) were treated with TM dissolved in corn oil. Animals received 5 mg per day for 2 days by oral gavage 12 . All behavioural studies and tissue samples were performed or collected, respectively, after 6-8 weeks of TM treatment from both male and female mice, except for the circadian locomotor behaviour where only male mice were analysed.
PTZ and PTX were dissolved in saline (NaCl) and administered to Bmal1cKO (n ¼ 10) and control mice (n ¼ 9) 0.3 mg kg À 1 per day by intraperitoneal injection (injection volume was 0.1 ml per 30 g body weight) at the middle of the light phase (ZT6) ± 1 h. The chronic treatment protocol involved daily doses of PTZ or PTX for 10 days, followed by 2-4 days without treatment before assessment of learning and memory (NOR and novel object location) tests, or to wheel-activity monitoring. This protocol and dosage of PTX or PTZ normalize memory performance of Down syndrome mice models, from 1 week to 2 months (m) after the treatment 48 .
Mice were randomized into the experimental groups based on their body weight.
Mouse and behavioural activity monitoring. Four to 5 months old male Bmal1cKO (n ¼ 7) and control mice (n ¼ 8) were single-housed in cages equipped with running wheels (ENV-044; Med Associates, Inc.). Mice were adapted to the wheel for 3 days in standard LD cycles (12:12 h, lights on at 07:00 h) and the experiment started under these conditions during at least 7 days. Then, mice were released in constant darkness for 3-4 weeks in isolated black cages (developed in collaboration with Tecniplast Spa, Italy) followed by, at least, 8 days under LD cycles (12:12 h, lights on at 07:00 h). Running wheel activity was recorded in 5 min bins by Wheel Manager software (SOF-860; Med Associates, Inc.). The data obtained were analysed with Actogram J.
Cognitive tests. Control and Bmal1cKO mice were housed two to five per cage and were given ad libitum access to food and water. They were kept on a 12 h (07:00-19:00 h) LD cycle in a room maintained at 21°C. All tests were conducted during the light cycle (ZT 3-7). Mice were habituated to handling and transport from the colony room to the behavioural room for 3 days before beginning the tests. After transport to the behavioural room, mice were habituated for 1 h before any test. The investigators were blinded to group allocation during experiments.
For the NOR and novel object location tests, mice were tracked with an overhead FireWire camera (DMK 31AF03-Z2, The Imaging Source) and ANY-maze (Stoelting). All apparatuses and testing chambers were cleaned with 70% ethanol wipes between each animal.
NOR test. The day after habituation to the grey acrylic arena (44 Â 44 cm), mice were exposed to two identical objects for 10 min during the familiarization phase. Object preference was evaluated during this session. Testing occurred 1 and 24 h later in the same arena, to assess short-term memory and long-term memory, respectively. Mice were allowed to explore for 10 min the same arena but one of the familiar objects was changed by a new one. The objects used were different in shape, colour, size and material. Moreover, the position of the objects in the familiarization and test session was counterbalanced between animals. Any investigative behaviour towards (that is, head orientation, sniffing occurring within o1.0 cm) or deliberate contact with an object was used as measure of the exploration and was registered manually by an experimenter blind to genotype and treatment. The exploration time for the Familiar object (Fam or F) and the new object (New or N) during the test phase was recorded. Memory was operationally defined by the percentage of DI for the novel object as the time spent investigating the new object minus the time spent investigating the familiar one in the testing period (%DI ¼ ((NO À FO)/Total Exploration Time) Â 100).
SOL test. The SOL test evaluates spatial memory by measuring the ability of mice to recognize the new location of a familiar object on the basis of the available extra-maze cues. After habituation, mice were exposed to two identical objects for 10 min during the familiarization phase. Object preference was evaluated during this session. Testing occurred 1 h later in the same arena. Mice were allowed to explore for 10 min the same arena but one of the familiar objects was moved to a novel location. Any investigative behaviour towards (that is, head orientation, sniffing occurring within o1.0 cm) or deliberate contact with an object was used as the measure of exploration that was registered manually by an experimenter blind to genotype and treatment. The exploration time for the object in the old location (Old or O) and the object in the new location (New or N) during the test phase was recorded. Memory was operationally defined by the percentage of alternation index for the novel located object as the time spent investigating the object in the new location minus the time spent investigating the object in the old location in the testing period (% Alternation ¼ ((NL À OL)/Total Exploration Time) Â 100).
Each animal was tested in the novel object tasks only once. All experimental groups are thus fully independent. To exclude the possibility that NOR performance might be confounded by a priori spatial or object biases, placement of the novel object was alternated between the left and right corners of the open-field arena.
Determination of GABA levels in CSF by LC-MS/MS. Mouse CSF samples were collected from the cisterna magna following a previously described protocol 65 .
GABA was quantified by ultra performance LC-MS/MS, following the protocol described by Buck et al. 41 . Briefly, GABA was extracted from CSF by precipitation with acetonitrile spiked with deuterated GABA (D6) as internal standard (Sigma Aldrich). Given the intrinsic difficulty of collecting a sufficient amount of CSF from mice, the CSF content of two or three individual animals were pooled into final samples consisting of 7 to 12 ml of CSF. A total number of 14 controls and 10 Bmal1cKO animals were used for this experiment (pooled into 5 and 4 samples, respectively). GABA was then separated by hydrophilic interaction LC using a BEH HILIC 2.1 Â 100 mm column and a short gradient of water in acetonitrile (5 to 40% in 2 min), with both eluents added with formic acid to a final 0.1% v/v. Flow rate was kept at 0.45 ml min À 1 . GABA was quantified on a Xevo TQ-MS instrument operating in electrospray, positive ion mode and following the multiple reaction monitoring transitions as previously described 41 . Both the column and the ultra performance LC-MS/MS systems were purchased from Waters, Inc. (Milford, MA, USA). GABA was quantified using a standard calibration curve prepared by serial dilution in artificial CSF 66 .
The investigators were blinded to group allocation during experiments.
The following day, sections were extensively washed and incubated for 2 h with goat anti-rabbit or anti-mouse Alexa-488 or Alexa-546 secondary antibodies used at 1:1,000 dilution (Invitrogen, A-11034, A32723, A-11003 and A-11010). Slices were then washed, mounted with Prolong Gold and imaged in an inverted laser scanning confocal microscope (TCS SP5 microscope using a Â 20 or Â 40 objective (Leica Microsystems)). Quantification and analysis was performed in ImageJ software (Wayne Rasband, NIH, USA), by outlining the SCN or cortex from the 4,6-diamidino-2-phenylindole-stained image and using this template to measure the relative intensity of the immunostaining for VIP, GAT1 or GAT3. When more than one section was analysed from each animal, the mean of the measures from consecutive sections were used for that individual.
Cell cultures and transfections. Primary monolayer cultures of astrocytes or neurons were established from cerebral cortices of neonatal (P1-P3) or Embryonic day 17 (E17), respectively, Sprague-Dawley rats and maintained at 37°C in a humidified atmosphere of 5% CO 2 . Astrocyte cultures were prepared and maintained as previously described 67,68 . The cultures were maintained at 37°C in a humidified atmosphere of 5% CO 2 for 1 week and thereafter cells were trypsinized and subcultured for the different experiments. These astrocyte-enriched cultures contained 496% astrocytes as indicated by immunofluorescence, with a monoclonal antibody anti-GFAP (Clone 6F2, Dako; dilution 1:1,000).
For primary neuronal cultures, tissue was dissociated by digestion with 10% (v/v) Trypsin for 30 min at 37°C and 200,000 cells were plated onto 20 mm tissue culture dishes coated with 0.1 mg ml À 1 poly-D-lysine. Neurons were grown in Neurobasal medium supplemented with B27, Gmax, 2.5 U ml À 1 Penicillin and 2.5 mg ml À 1 Streptomycin. Three days after plating, and subsequently every 4-5 days, half of the medium was changed. Neuronal cultures were maintained for up to 25 days in vitro before being used for the different experiments.
Astrocytes were synchronized with 100 nM of Dexamethasone for 2 h. The hormone was washed out and used for co-cultured experiments or harvested at different time points for subsequent analyses. Neurons were synchronized with 100 nm of Dexamethasone, GABA (100 mM), glutamate (10 mM) or with vehicle in neuronal conditional medium for 2 h. For the co-culture experiments, astrocytes were plated on coverslips on 12-well dishes and grown until confluence. Cortical neurons were plated on coated 12-well dishes with paraffin feets, to avoid contact between the cultures. Astrocyte grown on coverslips were inverted and resting on the dish containing the neurons sharing the same culture media, and harvested at different time points for subsequent analysis. 1(S), 9(R)-( À )-Bicuculline methiodide was diluted in cultured medium (stock 10 mg ml À 1 ) and cells were treated with 30 mM. Cells treated with vehicle were used as controls.
Primary astrocytes were transiently transfected with Bmal1 siRNAs (ON-TARGET plus smartpool specific for rat Bmal1, Dharmacon) or scramble control (ON-TARGET plus non-targeting siRNAs, Dharmacon). Briefly, primary astrocytes were plated on coverslips on 12-well dishes and when reached 60-80% confluence were transfected with 10 nM of scramble control or Bmal1 siRNAs using the Lipofectamine RNAiMAX Transfection Reagent (Thermo Fisher Scientific) following the suggestions of the manufacturer. Medium was changed 24 h later and on the next day (48 h after transfection) were subjected for consequent experiments.
RNA isolation and quantitative real-time RT-PCR. Cells were harvested at the appropriate time points, with the time of treatment being defined as CT4. For each time point, we prepared samples for the assay in triplicate. Tissues (cortices and hippocampus) were collected for RNA isolation at the ZTs of interest from animals in LD cycles (12-12 h). Total RNA was extracted using TRIzol reagent following the manufacturer's instructions. RNA was further cleaned using an RNeasy Mini Kit. Complementary DNA was obtained by retrotranscription of 0.5 mg of total mRNA using the ImProm-II Reverse Transcription System following the manufacturer's instructions. Real-time reverse transciptase-PCR was done using the ABI PRISM.7900 (Applied Biosystems). For a 15 ml reaction, 9 ng of cDNA template was mixed with the primers to a final concentration of 200 nM and mixed with 7.5 ml of 2 Â QuantiFast SYBR Green PCR Master Mix. The reactions were done in duplicates using the following conditions: 5 min at 95°C followed by 40 cycles of 10 s at 95°C, 30 s at 60°C and 1 min at 70°C. The primers used are available on request. Gapdh or Pgk1 transcripts were used as reference controls.
GABA uptake assay. Primary astrocytes were grown and transfected in 12-multiwell plates with Bmal1 or Scrbl siRNAs. After 48 h, cells were rinsed once with PBS and pre-incubated with 1 ml of Hank's buffered salinesolution for 10 min at room temperature. GABA was added to the medium to a final concentration of 5, 10 and 40 mM, and the cells were incubated for 15 min at room temperature. The medium was collected and GABA levels analysed with the GABA research ELISA kit (Labor Diagnostica Nord GmbH & Co), following the manufacturer's instructions.
Statistical analysis. Data are presented as mean ± s.e.m. and were analysed and graphed using Prism 6 (GraphPad, San Jose, CA, USA). Statistical comparison of two groups was done by Student's two-tailed unpaired t-test or two-way analysis of variance with a post-hoc Bonferroni. Data were checked for normality and equal variances between groups. Po0.05 was considered as statistically significant and the significance is marked by *Po0.05, **Po0.01 and ***Po0.001. Statistical significance of the rhythmic expression was determined by Cosinor analysis (expression data was fit by a nonlinear least-squares regression with the following equation: y ¼ A þ B Â cos [2p Â (t À C)/24], where A is the rhythm-adjusted mean, B is the amplitude of the rhythm, C is the phase given in circadian time representing the time of peak expression and t is the circadian time). The required sample size was calculated based on the similar experiments and analyses carried out previously. The number of animals in each experiment is stated in the respective figure legends. Samples or animals were excluded from the data analysis with pre-established criteria, if they deviated more than 2 s.d. from the group mean.
Data availability. All raw/original relevant data are available upon request.