Astrocytic control of extracellular GABA drives circadian timekeeping in the suprachiasmatic nucleus

Significance Circadian clocks drive daily physiological rhythms that adapt us to day and night. Their disruption by societal pressures (shift-work) or disease (neurodegeneration) carries severe economic and health consequences. The suprachiasmatic nucleus (SCN) is the principal brain clock, coordinating rhythms across the body. It consists of interconnected neurons that use the inhibitory neurotransmitter GABA, alongside astrocyte supporting cells. Using microscopic imaging, pharmacology, and molecular genetic approaches, we show that astrocytes sustain SCN timekeeping by controlling a daily rhythm of GABA levels. Their reduction during the day, by enhanced clearance, provides a circadian window for neuronal activity. This resolves a long-standing paradox of how an exclusively inhibitory network sustains neuronal activity and highlights the critical importance of astrocyte-to-neuron signaling in the clock.

The suprachiasmatic nucleus (SCN) of the hypothalamus is the master circadian (~daily) clock in mammals. It coordinates the timing of distributed peripheral tissue clocks, thereby controlling adaptive behavioral and physiological rhythms. It is synchronized to environmental time by innervation from the retinohypothalamic tract (RHT) (1), but the RHT is dispensable for robust SCN timekeeping. When isolated ex vivo, the SCN sustains precise circadian rhythms of molecular, electrical, and metabolic activity.
At the cellular level, timekeeping pivots around a core transcriptional/translational feedback loop (TTFL), whereby the transcription factors CLOCK and BMAL1 drive expression of Period and Cryptochrome genes, and their encoded proteins, PER1/2, and CRY1/2, subsequently repress that transactivation. This generates self-sustaining cycles of TTFL activity and its dependent transcriptional outputs (2), including ion channels (3). These in turn direct daily rhythms in electrical excitability of SCN neurons (4), peaking in circadian day, with quiescence during circadian night (4)(5)(6). Cell-autonomous timekeeping is augmented by reciprocal circuit-level neuron-to-glia signaling that determines emergent properties of the SCN network: precise high-amplitude oscillation, tightly defined ensemble period and phase, and spatiotemporally complex cellular synchrony (7,8). These network-level mechanisms are so powerful that they sustain robust rhythms with periods ranging between 16 and 42 h in various genetic/pharmacological contexts (9). Neuronal electrical rhythms are integral to this synchronization through appropriately timed neurotransmitter/neuropeptide release: Pharmacological blockade of neuronal firing (10) and synaptic vesicle release (11) desynchronize and damp network oscillations.
SCN neurons are neurochemically heterogeneous due to their complex expression of a wide array of neuropeptides and receptors (12)(13)(14)(15)(16) but are also homogeneous insofar as almost all utilize the inhibitory small neurotransmitter γ-aminobutyric acid (GABA) (17). GABAergic neurotransmission is mediated via ionotropic GABA A or metabotropic GABA B -receptors, and both subtypes are expressed within the SCN (17) together with GABA transporters (GATs) (18). Nevertheless, little is known about the role of GABA in steady-state network timekeeping (17,19). It has been considered as a counterbalance to neuropeptidergic signaling (20), a synchronizing factor (21), a modulator of encoded phase (17), and a network adaptor to seasonal OPEN ACCESS changes in day length (22)(23)(24). Pharmacological and genetic disruption of GABAergic signaling does not, however, disrupt TTFL oscillations and circadian firing of SCN neurons ex vivo (9,25,26). This suggests that although GABA is important in vivo as an SCN output acting on distal targets (via GABAergic synaptic terminals) (26,27), it is largely dispensable at the network level. Indeed, how could a circuit that exclusively uses GABA as a neurotransmitter alongside neuropeptides sustain circadian electrical activity, when increased neuronal firing would inhibit the network via this GABAergic output?
To reconcile this paradox and identify a role for GABA within the SCN network, we first monitored real-time extracellular GABA dynamics ([GABA] e ) in free-running SCN explants. This revealed a counterintuitive daily bulk [GABA] e flux, peaking at night in antiphase to (GABAergic) neuronal activity. This rhythm is not, however, driven by circadian variation in neuronal activity, but is rather driven by daytime GABA uptake via GATs, GAT1 (Slc6a1, mGAT1), and GAT3 (Slc6a11, mGAT4). Furthermore, this rhythmic uptake is controlled by the cell-autonomous circadian clock of astrocytes. This reveals a GABAergic network mechanism whereby active daytime uptake is permissive to neuronal activation and neuropeptide release during circadian day. Thus, the SCN circuit has devolved GABAergic control of its neurons to its astrocytes and their cell-autonomous clock to sustain and encode circadian time.

Extracellular GABA is Rhythmic in the SCN and Peaks during
Circadian Night. To determine the circadian dynamics of [GABA] e , we transduced SCN explants with adeno-associated viral vectors (AAVs) encoding the fluorescent GABA reporter iGABASnFR (28) controlled by the neuronal synapsin (Syn) promoter. Appropriate membrane-targeted fluorescence (SI Appendix, Fig. S1) was observed across the explant 1-wk post-transduction (Fig. 1A). Fluorescence was recorded for ~5 d in tandem with a bioluminescent TTFL reporter, PER2::Luciferase (PER2::LUC), which provided a circadian phase reference: PER2::LUC peaks at circadian time (CT)12, the start of circadian night (Fig. 1A). Imaging revealed rhythms in [GABA] e (Movie S1) with a robust waveform consisting of a broad, flat peak and a sharp trough. The circadian properties of the [GABA] e rhythm were comparable with those of PER2::LUC, with identical periods (Fig. 1B, (Fig. 1F). All three circadian markers had identical periods (Fig. 1G (Fig. 1I). In SCN explants, therefore, the circadian rhythm of [GABA] e sits in antiphase to neuronal rhythmicity, peaking in circadian night.

GABA Transporters Facilitate Ongoing SCN Rhythmicity.
To reconcile the apparently paradoxical observation that neuronal activity (which should drive GABA release) occurs when [GABA] e is lowest and [GABA] e is highest at night when SCN neurons are inactive, we investigated mechanisms that could control GABA flux. First, we explored published single-cell RNA-sequencing (scRNA-seq) data from SCN explants harvested during circadian day (CT7.5) or night (CT15.5) (13,30) to determine the expression of genes encoding the four canonical GATs: GAT1 (Slc6a1, mGAT1), GAT2 (Slc6a13, mGAT3), GAT3 (Slc6a11, mGAT4), and BGT1 (Slc6a12, mGAT2). Data were clustered irrespectively of time of day into three principal cell groups: neurons (defined by Slc32a1, Tubb3, and Celf4 expression), astrocytes (defined by Aldh1l1, Gfap, and Aqp4 expression), and other (defined by exclusion of these markers). Cell clusters were then divided by time-of-day, and relative expression levels of the four Gat genes were evaluated. Genes encoding β-Tubulin 3 (Tubb3) and ALDH1L1 (Aldh1l1) were used as internal controls to confirm specific segregation of neuronal and astrocyte populations (Fig. 3A). This revealed Gat1 and Gat3 as the principal GATs within ex vivo SCN (with little to no Gat2 or Bgt1 expression), and expression of GAT1 and GAT3 across the SCN was confirmed by immunostaining (SI Appendix, Fig. S4A). scRNAseq revealed that Gat1 was expressed by SCN astrocytes and neurons, while Gat3 was specifically enriched in SCN astrocytes (Fig. 3A) (18,31,32). Furthermore, testing for differential expression in astrocytes between the day and night revealed a significant temporal variation in the levels of Gat1 and Gat3, with expression being higher during the circadian day (Fig. 3B). This was confirmed by qPCR (SI Appendix, Fig. S4B). The fact that the transcriptomic data predict that GATs are present at a higher abundance during the circadian day when [GABA] e is low and at a lower abundance when [GABA] e is high highlights them as candidates to generate the observed dynamics in [GABA] e .
To clarify this, we recorded PER2::LUC rhythms from explants treated with a range of doses of the GAT3-specific inhibitor (S)-SNAP 5114 (Fig. 3G). Again, as dose increased, period lengthened (Fig. 3H, one-way ANOVA F(7, 36) = 9.40, P < 0.0001), precision decreased (Fig. 3I, one-way ANOVA F(7, 36) = 14.01, P < 0.0001), and amplitude was suppressed (Fig. 3J, one-way ANOVA F(7, 36) = 56.66, P < 0.0001). These effects were reversible, but in contrast to the immediate restoration following removal of CI966-HCl, the TTFL took several cycles to return to its initial state: amplitude progressively expanded, cycle-to-cycle, post-washout (SI Appendix, Fig. S5 B and C). Importantly, the concentrations of (S)-SNAP 5114 that maximally produced these sustained effects, between 33 and 50 µM, are within 6 to 10 times the IC 50 for GAT3 transporters (IC 50  Vehicle had no effect, but GAT3 inhibition rapidly elevated [GABA] e and suppressed neuronal [Ca 2+ ] i rhythms and PER2::LUC (Fig. 3 K and L). When quantified as the absolute peak-to-trough excursion, this suppression was significant relative to vehicle treatment across all reporters ( The scRNAseq data suggested that the [GABA] e rhythm is generated by a daytime GAT-mediated uptake of GABA from the extracellular space, meaning that GAT activity should account for the trough but not peak of [GABA] e . To test this, we extrapolated the trough and peak levels from baseline recordings forward into treatment intervals (SI Appendix, Fig. S6), enabling us to quantify the relative change in the peak and trough of the [GABA] e rhythm whilst under treatment (Fig. 3N). This revealed a significant change in these metrics under treatment with (S)-SNAP 5114, but not vehicle (repeated-measures two-way ANOVA: treatment-effect F(1, 6) = 7.52, P = 0.03) which was not equal between the peak and the trough of the rhythm (repeated-measures two-way ANOVA: peak/trough-effect F(1, 6) = 10.13, P = 0.019; interaction, F(1, 4) = 9.02, P = 0.040). The flattening of the [GABA] e rhythm was caused by elevation of the circadian trough, but not the peak of the oscillation (Sidak's multiple comparisons test vehicle vs. (S)-SNAP 5114 actual/predicted ratio: peak 1.01 ± 0.01 vs. 1.10 ± 0.04 au, P = 0.25; trough 1.01 ± 0.02 vs. 1.22 ± 0.06 au, P = 0.01) (Fig. 3N). Thus, under free-running conditions, GAT3 in the SCN is responsible for taking up GABA released during the day, presumably to facilitate neuronal firing, which in turn is integral for high-amplitude, precise oscillation across the circuit.

Daytime Uptake of GABA Permits Neuropeptidergic Signaling.
Our model predicts that the daily [GABA] e rhythm, driven principally by daytime up-regulated GAT3 activity, facilitates increased neuronal excitability in the presence of enhanced neuronal GABA release. In the SCN, neuropeptide release is the critical functional consequence of neuronal excitability in sustaining circuitlevel oscillations (35,36). To interrogate the relationship between [GABA] e and neuropeptide release, we recorded extracellular VIP levels ([VIP] e ) via AAV-dependent expression of the fluorescent VIP sensor, GRAB VIP1.0 (Syn.GRAB VIP1.0 ). In this reporter, sensitivity to VIP is conferred by a catalytically dead VPAC2 receptor coupled to circularly permuted EGFP (37). We recorded Syn.GRAB VIP1.0 in SCN explants, phase-registering the signal against PER2::LUC oscillations (Fig. 4A). This revealed pronounced [VIP] e rhythms with properties comparable to the PER2::LUC rhythm (Fig. 4 B-D (Fig. 4F), whereas VIP-null SCN displayed damping [Ca 2+ ] i rhythms. Furthermore, VIP-null SCN displayed no observable [VIP] e rhythm, with levels remaining at the limit of detection, thereby confirming reporter specificity (Fig. 4F). The absence of [VIP] e rhythmicity in VIP-null SCN was reflected in the rhythmicity index (  (Fig. 4H), whereas in WT SCN, the [VIP] e peak was directly in phase with neuronal [Ca 2+ ] i . This is consistent with the phase of peak membrane potential and [Ca 2+ ] i in VIP cells (~CT7) (8) and indicative of daytime neuronal activity driving VIP release. Importantly, this occurs when [GABA] e is at its nadir, consistent with the view that low [GABA] e is permissive for daytime neuronal activity and the dependent neuropeptide release.
To test this, SCN slices expressing GRAB VIP1.0 were treated with (S)-SNAP 5114 to examine the effects of loss of GAT3 function on VIP release in the SCN. Following baseline recording, slices treated with vehicle continued to show rhythmic GRAB VIP1.0 fluorescence (Fig. 4I). In contrast, in slices treated with (S)-SNAP 5114, the GRAB VIP1.0 rhythms were immediately compromised resulting in arrhythmicity (Fig. 4I), alongside a progressive damping of the PER2::LUC TTFL report. This was reflected in an immediate and severe reduction in the rhythmicity index in the [VIP] e report (repeated-measures two-way ANOVA: reporter-effect F(1, 6) = 10.27, P = 0.0185) between baseline and treatment intervals (repeated-measures two-way ANOVA: treatment-effect F(1, 6) = 16.27, P = 0.0069; interaction F(1, 6) = 5.62, P = 0.0555) (Fig. 4J). GAT3 function is therefore necessary to sustain circadian daytime release of the neuropeptide VIP in the SCN.

Astrocytes Initiate Network Rhythmicity through Acute Control of [GABA] e .
Due to the fact that the GAT3 transporter is integral to neuropeptide release from SCN neurons to ensure correct network function and that it appears to be an astrocyte-enriched transporter and circadian-regulated, we next explored whether the circadian clock of astrocytes could, autonomously, control [GABA] e . We have previously used CRY1-complementation targeted to astrocytes to control circadian network function by initiating oscillations in arrhythmic CRY1,2-null SCN (38,39). Using this approach, however, initiation takes >7 d (39). To increase the temporal precision with which the astrocytic TTFL could be controlled and its acute effects observed, we utilized translational switching (ts) of CRY1 protein (40). To validate this approach in astrocytes, CRY1,2-null SCN explants with the PER2::LUC reporter were transduced with tsCRY1 (pCry1.CRY1 (TAG) ::mRuby) (41) alongside the orthogonal tRNA-synthetase machinery targeted to astrocytes via GFAP promoter (GFAP.BFP2-P2A-mMPylS/PylT) (Fig. 5A). Following baseline recording, slices were treated with either vehicle or 10 mM noncanonical amino acid alkyne lysine (AlkK), a dose previously shown to initiate CRY1-translation in SCN explants (40)(41)(42). Vehicle-treated slices remained arrhythmic. In contrast, TTFL oscillation emerged in 10mM AlkK-treated slices (Fig. 5B) with a period (26.95 ± 0.27 h) appropriate for CRY1-driven oscillations. It was initiated within 2 d, faster than when Cre-recombinase is used to express CRY1 in astrocytes (38,39). In all slices tested, 10 mM AlkK initiated SCN-wide rhythms as evidenced by the change in rhythmicity index between baseline and treatment intervals (Fig. 5C, ΔRI: vehicle 0.02 ± 0.02 vs. 10 mM AlkK 0.16 ± 0.04, paired twotailed t test t(10) = 3.69, P = 0.0042). Furthermore, the effect was reversible on withdrawal of AlkK (SI Appendix, Fig. S7).
This approach therefore presented a unique opportunity to observe acute [GABA] e changes as network oscillations were initiated by the cell-autonomous astrocytic TTFL. CRY1,2-null slices were supertransduced with Syn.iGABASnFR to monitor [GABA] e as network rhythms initiated. In these CRY1,2-null SCN slices, similar to the TTFL rhythm, [GABA] e was arrhythmic ( maintained following treatment with vehicle (Movie S4 and Fig. 5F). In contrast, under 10 mM AlkK treatment, not only were bioluminescent rhythms initiated but [GABA] e also started to fluctuate with an initial rapid decrease followed by a rapid increase into a sustained peak (Movie S5 and Fig. 5F, marked R). This sustained elevation of [GABA] e aligned with the nadir of the bioluminescent signal, marking the first circadian night and repressive phase of the TTFL clock (Fig. 5F).
[GABA] e then fell into a sharp trough (Fig. 5F, marked P) in the first circadian day, marked by rising PER2::LUC bioluminescence. Subsequently, [GABA] e rose again to another sustained peak in the second circadian night (>CT12). These dynamics, therefore, align low [GABA] e with the positive arm of the neuronal TTFL (38) and high [GABA] e with the repressive arm. This indicates that, as network timekeeping is initiated, acute changes in [GABA] e driven by the astrocytic TTFL generate network oscillation. They first facilitate repression via neuronal inhibition under high [GABA] e , which is then followed by coordinated facilitation of neuronal activity under low [GABA] e via the permissive uptake of GABA from the extracellular space.
To assess the initiation of TTFL and [GABA] e dynamics quantitively, we measured the coherence of the initiated oscillations by calculating their relative amplitudes, defined as the difference between the peaks and troughs predicted from the dynamics of the PER2::LUC bioluminescence recorded in photomultiplier tubes (PMTs) (Fig. 5B) and the circadian dynamics of [GABA] e in wild-type slices (Fig. 1). We saw that in the high-temporal resolution PMT traces, PER2::LUC reached its first initiated peak 34.5 h (34.6 ± 0.6 h, Fig. 5B) after AlkK treatment. Using this, we identified when the troughs in the PER2::LUC and [GABA] e oscillation that preceded this peak would have fallen (21 h and 26 h posttreatment, respectively) and where the ensuing peak in [GABA] e would have fallen (39.5 h posttreatment) in the normalized data (Fig. 5F). Using these measures, we observed a low relative amplitude of the vehicle-treated oscillation in PER2::LUC and [GABA] e , which increased in both when treated with AlkK ( Fig. 5G)

Discussion
Free-running SCN explants display robust rhythms in [GABA] e , peaking at night in antiphase to neuronal activity (Fig. 1). This observation is counterintuitive because a peak in [GABA] e would be expected in circadian day when SCN GABAergic neurons are active. Equally, GABA is inhibitory, and so, high daytime [GABA] e would be expected to suppress neuronal activity and therefore its own release. The paradox could be resolved if GABA were excitatory, a condition observed during development (43), arising when intracellular chloride levels ([Cl − ] i ) are high. This is achieved via changes in the expression of K-Cl (KCC2, Slc12a5) and Na-K-2Cl (NKCC1, Slc12a2) cotransporters and leads to a reversal of GABA A -receptors on activation (44,45). Indeed, a model whereby SCN neurons may switch GABAergic transmission between excitatory and inhibitory on a circadian basis has been proposed (46) and excitatory GABAergic drive reported in the SCN under certain conditions. There is, however, no consensus on its temporal or spatial patterning (for review see ref. 17). More directly, our data contradict this model, indicating that GABAergic signaling, mediated by GABA A -receptors, forms a suppressive rather than excitatory axis in the SCN (Fig. 2 and SI Appendix, Figs. S2 and S3). In our explants, the SCN neurons and astrocytes demonstrate mature phenotypes, and there is no evidence in our preparations for the activity of NKCC1 (SI Appendix, Fig. S8). GABA is therefore suppressive, and this suppression is presumably driven by extra-synaptic GABA A -receptors which are well suited to sense circadian changes in bulk [GABA] e and drive long-term inhibition during the night due to their high affinity and decreased desensitization in the prolonged presence of GABA (17,47,48). A limitation of our approach, however, is that we have not assessed this mechanism directly via electrophysiological investigation. A wealth of electrophysiological data focusing on the specifics of cellular GABAergic signaling in the SCN are available (17,49), but here our focus was specifically on the global dynamics of [GABA] e .
While we consider GABA to be suppressive within the SCN, via a potentially tonic mechanism, this could have effects beyond simple inhibition of the neurons. In thalamic neurons, a consequence of increased tonic GABAergic currents is to hyperpolarize neurons into the activation range for T-type calcium channels (Cav3.1, 3.2, and/or 3.3) (50), which serves to permit neuronal activity through rebound depolarization (51). Additionally, in hippocampal neurons, it has been shown that activation of tonic GABAergic currents decreases the membrane time-constant leading to a requirement for higher precision of timing of incoming excitation (52). In both these ways, enhanced tonic GABAergic currents alone or in tandem with T-type calcium channels can generate excitation and augment responses to incoming afferent signals to synchronize and promote neuronal activity. T-type channels are expressed within the SCN (53)(54)(55), although in isolated SCN, pharmacological inactivation of T-type calcium channels does not alter the circadian electrophysiological properties of SCN neurons (54). This indicates that activation of T-type calcium channels is not required at network steady state to generate or synchronize circadian cycles of neuronal firing. This does not preclude a role for interaction between tonic GABA and T-type calcium channels, as this could be a mechanism in vivo by which responses to incoming glutamatergic afferents from outside the SCN are augmented by T-type calcium channels during the night. Indeed, pharmacological inactivation of these calcium channels has been shown to block glutamate-induced phase delays during circadian night (53) when [GABA] e is rising (Fig. 1). Our current model therefore predicts that the high level of GABA we report during the night generates tonic GABA currents that act at two levels: first, it supresses neuronal activity when the network is at steady state, and second, it potentially primes the network to respond to appropriately timed extra-SCN afferent signals through a mechanism that involves T-type calcium channels.
In our model, astrocytes, not neurons, control [GABA] e (Fig. 5I). This predicts that GABA uptake during the day (when neurons are most actively releasing GABA) via the GAT3 transporter prevents GABA spill-over from synaptic sites and accounts for the sharp trough in the [GABA] e oscillation (Fig. 1). This uptake is permissive to neuronal activity, which in turn supports electrically sensitive CREs in Per genes, thereby leading to robust high-amplitude TTFL oscillation across the SCN (Fig. 2) (3, 4). Conversely, elevated [GABA] e sustains the nocturnal repressive phase of the TTFL.
How then, does the [GABA] e rhythm sit in antiphase to the activity of its presumed source? Using our published single-cell transcriptomic dataset (13), we found a day-night difference in GAT expression, predominantly on astrocytes ( Fig. 3 and  SI Appendix, Fig. S3). SCN expression of GATs has been reported (18,31,32) and potential circadian variation shown in rat SCN (18) and cortical astrocytes (31). Further, this circadian difference in GAT transcripts has been reported in cortical and hippocampal preparations to be controlled by the circadian transcription factor REV-ERBα via inhibition of the clock-controlled output gene E4BP4 (56), driving transcript levels to peak during the day and reach their nadir during the night, consistent with our proposed model. Our functional tests of GAT1 and GAT3 inhibition revealed that their inactivation suppressed the PER2::LUC oscillation (Fig. 3) because of the consequent rising levels of GABA A -receptor activation (Fig. 2). Furthermore, GAT3 inhibition produced stronger effects on application and during washout ( Fig. 3 and SI Appendix, Fig. S5), likely due to the fact that GAT3 has a higher affinity for GABA compared to GAT1 (57) allowing GAT3 to lower GABA levels more efficiently in the absence of GAT1 function. GAT3 is therefore the stronger candidate as the principal transporter for control of [GABA] e and network function in the SCN via daytime GABA uptake. This critical daytime uptake of GABA was confirmed by direct imaging of the [GABA] e oscillation under treatment with the GAT3-specific inhibitor (S)-SNAP 5114, where loss of rhythmicity in the GABA report is caused by an acute elevation of the trough without effects on peak level (Fig. 3). Daytime removal of GABA from the extracellular milieu is therefore necessary for neuronal activity, and activity-dependent neuropeptide release, which reinforces circuit-level rhythms (Fig. 4) accounting for the loss of PER2::LUC amplitude. Consistent with our model, whereby a low level of [GABA] e is permissive to timekeeping, sustained loss of GABAergic release from neurons by genetic targeting of the vesicular transporter VGAT does not markedly influence SCN molecular rhythmicity (26), even though it compromises transmission of behaviorally relevant cues at distal extra-SCN targets. Equally, TTFL function in the SCN is sustained during blockade of GABA A -and/or GABA B -receptors ( Fig. 2 and refs. 9 and 25), although GABA A blockade [(+)-bicuculline] elevates both the peak and trough of the oscillation (Fig. 2), indicating that GABA fine-tunes SCN electrical activity within a certain dynamic range that complements TTFL function (26,27). The nocturnal rise in [GABA] e is important because it curtails the permissive low-GABA state. If that low-GABA state is prevented by GAT3 blockade, not only is the neuronal activity rhythm (monitored by [Ca 2+ ] i ) damped but the consequent release of neuropeptides is obstructed.
VIP is a principal SCN neuropeptide, essential for circadian neuronal synchrony and high-precision oscillation (58)(59)(60), driving the cell-autonomous TTFL of cells expressing its receptor, VPAC2, by kinase-dependent signaling to CRE-elements in the Per genes (7,61). A VIP-GABA counterbalance mechanism has been proposed to regulate the SCN TTFL (20,31) and GAT-mediated GABA uptake suggested as a conduit for circadian astrocyte-to-neuron information transfer in coculture in vitro models (31). Indeed, pharmacological disruption of GABAergic signaling counteracts some of the deficits of VIP deficiency in isolated SCN (20), adding to the notion that opposing GABAergic and neuropeptidergic cues set the dynamic range for normal TTFL function. Using a VIP-specific reporter (37), we mapped peak VIP release to the middle of circadian day (CT6.97), consistent with previous recordings and the known activity cycle of VIP neurons (8,36) and showed that VIP and GABA are ordinarily maintained in a distinct temporal segregation in the SCN, held in antiphase by GABA uptake by astrocytic transporters. Compromising this uptake rapidly damped rhythmic VIP release, revealing explicitly the interplay between [GABA] e and [VIP] e in circadian time and showing that control of [GABA] e not only directs the cell-autonomous TTFL but also circuit-level SCN timekeeping.
Nighttime release of GABA is compatible with our previously proposed astrocyte-to-neuron signaling axis, whereby nocturnally released astrocytic glutamate depolarizes neurons expressing NR2C-containing NMDA receptors, increasing action potential-independent GABA release from these neurons (6). The effectiveness of this synaptic release would be augmented by nocturnal inactivation of GABA uptake, allowing for enhanced spill-over of GABA into extra-synaptic sites. A further intriguing aspect is that the metabolism of GABA and glutamate is intimately interlinked via the GABA/glutamate-glutamine shuttling between astrocytes and neurons (62), which may point to some metabolic coupling between the daytime active neurons and nighttime active astrocytes. Indeed, pharmacological disruption of astrocytic metabolism compromises the ongoing network oscillation (39). In the SCN, synthesis of glutamate is dependent on astrocytic metabolism, positioning them as the metabolic source (6). However, GABA metabolism is more complex, and it has been shown in other brain areas that both neurons [which express GAD65/67 (63,64)] and astrocytes are capable of synthesizing GABA, potentially via alternative mechanisms (65). Release of glutamate into the SCN extracellular space has been tracked to nonvesicular release from the astrocytes (6,38), but whether the source of the nighttime GABA is solely from the neurons or also arises from the astrocytes by similar mechanisms remains unknown. Nevertheless, in these complementary, coordinated ways, it is evident that astrocytes can suppress the SCN neuronal network during the night directly by manipulating the properties of their extracellular milieu.
Evidence of this reciprocal astrocyte-to-neuron signaling is provided by the reemergence of TTFL rhythms following the removal of GAT3 inhibition. On washout, the TTFL amplitude did not return immediately, but rather, it grew steadily, cycle-to-cycle (SI Appendix, Fig. S5 B and C). This progressive recovery following a switch from high [GABA] e to low [GABA] e can be explained by the mutual dependence of neuronal and astrocytic TTFLs. In the prolonged high [GABA] e state, loss of neuronal rhythmicity likely also impairs astrocytic rhythms secondarily, and following transfer to the low [GABA] e state, it requires several cycles of mutual reinforcement to establish full spectrum oscillations in both cell types. A comparable reestablishment of TTFL cycles occurs following washout of cycloheximide (suspending cell-autonomous TTFLs) or tetrodotoxin (suspending neuronal signaling) (10,66,67). In the case of GAT3 inhibition, these dynamics are not, however, driven purely by release of GABAergic inhibition to the neurons because washout of muscimol does not produce the same delayed increase in amplitude (SI Appendix, Fig. S2). The building of amplitude upon washout of GAT3 therefore represents a gradual reconfiguration of the mechanisms to control GABA uptake and release across neurons and astrocytes following chronic inactivation of an astrocyte-specific transporter. We therefore interpret this as the SCN network iteratively rebalancing itself as the neurons and astrocytes are recoupled on a cycle-to-cycle basis. Comparable washout dynamics are seen following disruption of astrocyte-to-neuron signaling when NR2C-receptors or glutamate uptake are inhibited and then restored (6), which could reflect mutual recoupling of signaling and/or metabolic processes.
Our model proposes that daily rhythms in GABA uptake, driven principally by astrocytes, control circadian dynamics of bulk GABA in the extracellular space. This uptake is dependent on a functional SCN clock, as [GABA] e is arrhythmic in CRY1,2-null explants, consistent with previous observations of arrhythmic astrocytic TTFL and [Ca 2+ ] i and extracellular glutamate (38,39,68). In order to identify causal events in a repetitive, cyclical system such as the SCN, where changes on one cycle may be caused by events some cycles previously, it is necessary to have precise temporal control of the pertinent components. We therefore used tsCRY1-expression specifically in astrocytes to activate their cell-autonomous TTFL and thereby test our model by following the sequence of events (Fig. 5) as circadian rhythmicity was initiated, de novo, in arrhythmic CRY1,2-null SCN (40)(41)(42). This revealed precise patterning of coherent [GABA] e dynamics occurring over the first 48h, with changes in [GABA] e simultaneous with those in the TTFL. Furthermore, from their initiation stage, these rhythms matched the relative phasing in wild-type SCN. Thus, in the absence of a functional clock in the rest of the SCN, astrocytes can impose their cell-autonomous circadian state onto the SCN network by initiating circadian cycles of [GABA] e which are necessary to entrain the "clockless" neurons. We have, therefore, revealed a different level of astrocyte-to-neuron communication that controls SCN network dynamics. Having established this mode of communication, it is now imperative to explore whether further signaling axes support this and to determine the mechanisms by which neurons reciprocate this signaling to astrocytes. A particularly striking observation from our results is that despite neurons being the (presumed) source of GABA in the SCN network, circadian control of this signaling is devolved to the astrocytes. Is this, therefore, a more general mechanism whereby astrocytes regulate other GABAergic circuits in the brain across different time-scales?

Materials and Methods
Detailed description of materials and methods is provided in SI Appendix along with a summary of analysis techniques and R scripts to repeat scRNA-seq analysis of publicly available datasets.
Animals were used in accordance with the UK Animals (Scientific Procedures) Act of 1986, with local ethical approval (MRC LMB AWERB). SCN explants were made from postnatal day 10 (P10) to P12 of either sex and cultured as an organotypic explant via the interface method. Slices cut at 300 µm were rested for at least a week before further experimentation or AAV transduction. AAV transductions were carried out by applying AAV particles in PBS as a 1 µL droplet to the surface of the SCN explant. In the case of serial transductions, these were made 2 to 3 d apart and explants were left for at least 1 wk following the final transduction before the medium was changed. Full details of the animal genotypes and AAVs used are detailed in SI Appendix.
PER2::LUC bioluminescence was recorded in PMTs or a LumiCycle (Actimetrics), while multiplexed bioluminescence and fluorescence imaging was carried out on LV200 Bioluminescence Imaging Systems (Olympus). Where slices were manipulated pharmacologically, compounds were added directly to the medium. For confocal imaging, SCN explants were fixed in 4% PFA before subsequent processing for immunohistochemistry and mounting. Full details of the recording procedures and parameters for live bioluminescent and fluorescent recordings and confocal fixed imaging are included in SI Appendix.
For qPCR analysis, RNA was extracted from explants (phase-mapped via their PER2::LUC oscillation) using a Direct-Zol RNA microprep kit column-based kit (Zymo Research) and reverse transcribed to cDNA using the QuantiTect Reverse Transcription Kit (Qiagen). qPCR was carried out using the QuantiNova SYBR Green PCR Kit (Qiagen), and all samples and the standard curve were run in triplicate alongside no template controls on a Techne PrimePro 48 real-time PCR machine (Techne). All primers, cycling parameters, and associated information are included in SI Appendix. Data, Materials, and Software Availability. All study data are included in the article and/or supporting information. Previously published data were used for this work [NCBI Gene Expression Omnibus with the accession number: GSE167927 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE167927)] (30).