Activity-dependent switch of GABAergic inhibition into glutamatergic excitation in astrocyte-neuron networks

Interneurons are critical for proper neural network function and can activate Ca2+ signaling in astrocytes. However, the impact of the interneuron-astrocyte signaling into neuronal network operation remains unknown. Using the simplest hippocampal Astrocyte-Neuron network, i.e., GABAergic interneuron, pyramidal neuron, single CA3-CA1 glutamatergic synapse, and astrocytes, we found that interneuron-astrocyte signaling dynamically affected excitatory neurotransmission in an activity- and time-dependent manner, and determined the sign (inhibition vs potentiation) of the GABA-mediated effects. While synaptic inhibition was mediated by GABAA receptors, potentiation involved astrocyte GABAB receptors, astrocytic glutamate release, and presynaptic metabotropic glutamate receptors. Using conditional astrocyte-specific GABAB receptor (Gabbr1) knockout mice, we confirmed the glial source of the interneuron-induced potentiation, and demonstrated the involvement of astrocytes in hippocampal theta and gamma oscillations in vivo. Therefore, astrocytes decode interneuron activity and transform inhibitory into excitatory signals, contributing to the emergence of novel network properties resulting from the interneuron-astrocyte interplay. DOI: http://dx.doi.org/10.7554/eLife.20362.001

We then stimulated the interneuron with longer depolarizing pulses (700 ms) to elicit bursts of APs that were followed by the SC stimulus ( Figure 1B). In 33 out of 64 pairs, this stimulation paradigm enhanced the synaptic efficacy (from 9.48 ± 0.74 pA to 13.89 ± 1.09 pA after interneuron stimulation; 148.80 ± 5.86% of basal; n = 33; p<0.001; paired t test) due to an increase of the success rate (from 0.42 ± 0.04 to 0.60 ± 0.04 after interneuron stimulation; 147.62 ± 4.94% of basal; p<0.001; paired t test), without affecting the synaptic potency (from 18.79 ± 1.42 pA to 19.78 ± 1.51 pA after interneuron stimulation; 105.40 ± 4.03% of basal; p=0.37; paired t test) ( Figure 1D,E). The increase of success rate and the constancy of synaptic potency indicate a presynaptic mechanism, as supported by the decrease in the PPF ratio (from 1.76 ± 0.10 to 1.48 ± 0.09; n = 33; p=0.04; paired t test) ( Figure 1F). The interneuron-mediated synaptic potentiation was observed for synapses showing a wide-range of the success rate ( Figure 1G; n = 33; p<0.001; paired t test), indicating that the capability of synaptic modulation was unrelated with their initial values of transmitter release.
Both regulatory phenomena were mediated by GABAergic signaling, but while synaptic inhibition was abolished by picrotoxin (GABA A receptor antagonist; 50 mM) without affecting the synaptic enhancement ( Figure 1E), the synaptic potentiation was impaired by CGP55845 (GABA B receptor antagonist; 5 mM) without modifying inhibition ( Figure 1E). Therefore, interneuron single APs inhibited CA3-CA1 synapses through activation of presynaptic GABA A receptors, whereas high interneuron firing rates surprisingly potentiated excitatory synaptic transmission through activation of GABA B receptors. The degree of synaptic potentiation was activity-dependent, varying with the interneuron number of APs according to the Hill equation ( Figure 1H), but was independent of the interneuron subtype stimulated (Figure 1-figure supplement 1D,E). In order to test the cell-type influence to the interneuron-induced synaptic potentiation, parvoalbumin-positive (PV + ) interneuron activity was evaluated. Pairs of CA1 pyramidal neuron and PV + -interneurons were recorded from PV- EPSCs from the total number of stimuli); and synaptic potency (i.e., mean EPSC amplitude excluding failures) (bin width, 33 s) before and after pairing SC stimuli with single (C) red; n = 10) or bursts (D) blue; n = 33) of interneuron APs. Horizontal bars indicate the time of pairing. (E) Relative changes of synaptic parameters from basal (black) elicited by pairing SC stimuli with single (red) or bursts of interneuron APs (blue), in control (n = 10 and 33, for red and blue, respectively), picrotoxin (PTX; n = 6 and 7, for red and blue, respectively), and CGP55845 (n = 5 and 9, for red and blue, respectively). (F) Paired pulse ratio (PPR). (G) Success rate from synapses shown in D, and averaged values (black) in control and after IN burst stimulation (n = 33). The IN-mediated effects were independent on their initial values. (H) Relative potentiation of synaptic efficacy vs. interneuron firing activity (n ! 4 for each data point; Hill equation fitting, R 2 = 0.9955). *p<0.05, **p<0.01, #p<0.001; paired t test. See also tdTomato transgenic mice (Figure 1-figure supplement 2), showing a marked potentiation of excitatory synaptic transmission after PV + -cell burst stimulation (success rate: 146.02 ± 17.86% of basal; n = 9; p=0.002; paired t test; Figure 1-figure supplement 2C,D), with analogous features to those synapses from unlabeled interneuron recordings ( Figure 1D,E), supporting that interneuron potentiation of excitatory synaptic transmission might be a broad phenomenon involving different interneuron subtypes.
We next investigated whether inhibition and potentiation of excitatory transmission induced by interneuronal activity were segregated processes to a particular set of synapses or could concur at the same synapse. We found that~17% of recorded synapses (6 out of 36 recorded pairs) showed both inhibition and potentiation of EPSCs when single and bursts of interneuron APs were consecutively evoked (Figure 2A,B). These synapses showed similar properties to those that expressed inhibition and potentiation independently (see Figures 1C,D and 2C), indicating that the mechanisms of synaptic inhibition can coexist with those responsible for synaptic potentiation at the same synapses, and the final outcome is regulated by the interneuron firing rate.

Interaction between interneuron and astrocyte signaling dynamically controls excitatory synaptic transmission
We next studied the dynamic interplay between interneuron, astrocyte activity and excitatory transmission to determine whether interneuron-mediated inhibition and potentiation could occur under particular conditions mimicking physiological hippocampal activity. Interneuron activity plays key roles in the appearance of certain oscillatory network activities, such as theta (3-12 Hz) and gamma frequency bands (25-100 Hz), which have relevant functions in coding neural information (Lisman and Jensen, 2013). Therefore, we stimulated the interneurons to elicit firing patterns that simulate those involved in these rhythms. Bursts of interneuron APs were elicited by depolarizing pulses (166 ms delivered at 3 Hz for 30 s) while continuously stimulating SC at 6 Hz. Thus, SC-evoked EPSCs were phase-locked at the interneuron depolarization (up-EPSC) and resting levels (down-EPSC) ( Figure 5A, inset). The analysis of the responses showed that interneuron stimulation bursts evoked: (1) an overall potentiation of the synaptic efficacy mediated by an enhancement of the success rate at both up-EPSCs and down-EPSCs (success rate: 132.89 ± 5.32% and 150.23 ± 6.22% of basal, respectively; n = 13; p<0.001 vs. basal; paired t test) that lasted throughout the stimulus period, and (2) fast dynamic changes of the synaptic efficacy and success rate within each up/down cycle (i.e., depolarized or resting state of the interneuron) ( Figure 5A), that is, up-EPSC values were significantly lower than those of immediately succeeding down-EPSCs (mean relative success rate difference: 17.34%; p=0.04; paired t test) ( Figure 5B). No significant changes in the synaptic potency (p=0.29; paired t test; Figure 5B) were observed. Consistent with the mechanisms described above, the overall INevoked potentiation was unaffected by picrotoxin (down-EPSC, n = 11; p<0.001; paired t test), but blocked by CGP55485 (down-EPSC, n = 11; p=0.97; paired t test) or MPEP+LY367385 (down-EPSC, n = 8; p=0.53; paired t test) ( Figure 5A,B), and absent in Ip3r2 -/mice (down-EPSC, n = 7; p=0.06; paired t test) (Figure 5-figure supplement 1), indicating that it requires GABA B receptor activation, astrocyte Ca 2+ signaling, and mGluR activation. In contrast, differences in synaptic efficacy and the success rate in up-EPSC vs. down-EPSC were abolished by picrotoxin (mean relative success rate difference: 0.40%; n = 11; p=0.79; paired t test) ( Figure 5A,B) but were still present in the rest of conditions. Indeed, mean relative differences for the success rate were 27.72% with CGP55845 (n = 11; p=0.013; paired t test) and 19.30% with MPEP+LY367385 (n = 8; p=0.04; paired t test) ( Figure 5B), and 13.33% in Ip3r2 -/mice (n = 7; p=0.01; paired t test; Figure 4B), indicating that the reduced synaptic transmission during up-EPSCs relative to down-EPSCs was due to the GABA A -mediated inhibition. Similar results were observed by identical interneuron activation when SC were stimulated at 3 Hz, independently phase-locked at either the up or down-state of interneuron activity ( showing that synaptic stimulation per se did not account for the potentiation observed and that interneuron-astrocyte signaling was necessary to induce the enhancement of excitatory synaptic transmission. In summary, bursts of interneuron activity induced a dynamic modulation of CA3-CA1 synaptic transmission consisting on an overall steady potentiation superimposed with faster transitions within each cycle of up and down interneuron states. While the latter and faster effect is due to GABA A -mediated inhibition of transmitter release, the former and sustained modulation is mediated by astrocytic GABA B receptors, which stimulate the Ca 2+ -dependent release of glutamate that activate presynaptic mGluRs at the excitatory terminals.  Because GABA B receptors are present in neurons, both pre-and postsynaptically (Vigot et al., 2006), and in astrocytes (Meier et al., 2008), to further support the astrocytic function of GABA B receptors, we generated genetically modified mice with conditional ablation of GABA B receptors specifically in astrocytes (GB1-cKO mice). Deletion of the receptor subunit gene Gabbr1 is sufficient to completely abolish GABA B receptor function, even when the GABBR2 subunit is expressed as well. We crossed GLAST-CreERT2 knockin mice (Mori et al., 2006) with Gabbr1 fl/fl mice (Haller et al., 2004) ( Figure 6A). Mice with astrocyte-specific GABA B receptor ablation were investigated 6 to 8 weeks after induction of gene recombination by intraperitoneal injection of tamoxifen. qRT-PCR of genomic DNA revealed 28.5% of Gabbr1 gene recombination (n = 7 for both GB1-cKO and control mice; p=0.003; unpaired t test; Figure 6B), a number reflecting the percentage of astrocytes in the hippocampus. Due to the high abundance of neuronal expression, the mRNA levels were not significantly reduced in GB1-cKO mice (n = 3 for both GB1-cKO and control mice; p=0.24; unpaired t test; Figure 6B). Quantifying the co-localization of astroglial GLAST and Gabbr1 immunolabels (Cordelières and Bolte, 2014), however, demonstrated a significant reduction of Gabbr1 at the protein level (9 sections from 2 GB1-cKO, 11 sections from two control mice; 2-sided; p<0.001; unpaired t test; Figure 6B), although the spatial resolution in single optical sections of a confocal laser-scanning microscope might underestimate the Gabbr1 removal from the fine astrocyte processes.
We next evaluated the neuronal responses evoked by baclofen in CA1 pyramidal cells from wildtype and GB1-cKO mice in the presence of TTX (1 mM). Local application of baclofen (2 mM, 10 s) induced outward currents either in wild-type (13.73 ± 2.12 pA; n = 8 baclofen puffs; three neurons) and GB1-cKO neurons (10.76 ± 1.0 pA; n = 9 baclofen puffs; three neurons; p=0.398; Wilcoxon ranksum test), which were both abolished by CGP55845 ( Figure 6F). These results indicate that the selective deletion of the receptor subunit gene Gabbr1 in astrocytes did not affect the neuronal GABA B receptor activity. Therefore, these data indicate that GABA B -induced Ca 2+ elevations in astrocytes are required for the interneuron-evoked potentiation of excitatory synaptic transmission. In order to evaluate the impact of astrocytic GABA B receptor expression deficiency to the coding properties of cortical networks, we performed in vivo recordings in anesthetized GB1-cKO mice and wild-type littermates. We examined the basic properties of local field potentials (LFP) in the dorsal hippocampus in resting conditions (Figure 7-figure supplement 1) and after whisker stimulation ( Figure 7A). Analysis of the LFP responses and power spectrum revealed that sensory stimulation boosted the theta and gamma components in control littermates, and significant changes occurred in those oscillatory neuronal responses in mice with down-regulated expression of astrocytic GABA B receptors ( Figure 7B,C). Both theta band (peak 4-8 Hz; p=0.007; unpaired t test) and low gamma band activities (peak 30-50 Hz; p=0.042; Wilcoxon rank-sum test) were partially reduced in GB1-cKO mice ( Figure 7C), indicating a significant role of astrocytic GABA B receptors in these network oscillations, which are related to cognitive and behavioral tasks (Buzsáki and Moser, 2013). Thetagamma coupling is thought to have an important function regulating hippocampal-cortical and subcortical communication during learning, episodic memory and recall tasks (Hanslmayr et al., 2016;Tort et al., 2009). We therefore analyzed the cross-frequency coupling between theta phase and gamma amplitude in resting and after whisker stimulation ( Figure 7D). We found that the magnitude of phase-amplitude coupling, measured as phase-locking value (PLV), was enhanced in control littermates after sensory stimulation (PVL= from 0.30 ± 0.03 to 0.50 ± 0.07; n = 24 epochs; p=0.017; paired t test; Figure 7E) as reported previously in behaving rodents (Csicsvari et al., 2003;Tort et al., 2009). In contrast, the coupling between theta and gamma oscillations was disrupted in GB1-cKO mice, which did not show changes in PVL index after whisker stimulation (PVL= from 0.39 ± 0.03 to 0.42 ± 0.04; n = 36 epochs; p=0.62; paired t test; Figure 7E), indicating that ablation of GABA B receptors in astrocytes weaken either hippocampal theta and gamma rhythms as well as their coupling. Additionally, the impact of astrocyte-GABAergic signaling was revealed under stimulus driven-hippocampal activity conditions (i.e., whisker stimulation), but no differences were observed in resting conditions (Figure 7-figure supplement 1), suggesting a state-dependent astrocyte neuromodulation of hippocampal rhythms in vivo. In summary, astrocytic GABA B receptors are involved in an oscillatory brain activity in vivo, contributing to theta and low gamma waves in stimulus-driven conditions.

Discussion
The operation of neuronal networks crucially depends of a fast time course of signaling in inhibitory interneurons. Our results show that GABAergic signaling dynamically impacts excitatory transmission in an activity-and time-dependent manner that is controlled by astrocytes. While excitatory Figure 6 continued localization are largely missing in the GB1-cKO mice. Scale bars, 30 mm; inset, 15 mm. (B) Top left, qRT-PCR of hippocampal genomic DNA reveals a reduction of the Gabbr1 fl/fl alleles by 28.5% (p=0.003; unpaired t test), representing the percentage of astrocytes in the hippocampus. Top right, quantification of Gabbr1 mRNA levels by qRT-PCR does not show a reduction of the Gabbr1 message (p=0.25; unpaired t test), as expected by the high levels of neuronal versus glial expression. Quantification of the immunolabels for Glast and Gabbr1 were determined by the ImageJ plugin JACoP v.2 by calculating the overlap (p<0.001; unpaired t test; bottom left) and Mander's M2 coefficients (p<0.001; unpaired t test; bottom right), both indicate a significant reduction of astroglial Gabbr1 expression (9 sections from 2 GB1-cKO and 11 sections from two control mice). (C) Confocal image from CA1 region of GB1-cKO mice showing the endogenous expression of GCaMP3 in astrocytes lacking Gabbr1 (in green). Scale bar, 60 mm. (D) Intracellular Ca 2+ signals induced by local agonist application of ATP (blue) and baclofen (green) from four representative astrocytes in wild-type and GB1-cKO mice, and astrocyte Ca 2+ transient probability over time induced by agonist stimulation or after evoking bursts of interneuron APs (white). Arrow, green and blue squares denote ATP, baclofen, or interneuron stimulation. Scale bar, 100%, 15 s. (E) Ca 2+ transient probability index after astrocyte stimulation in wild-type (n = 35 astrocytes from seven slices; p<0.001; Wilcoxon rank-sum test) and GB1-cKO mice. GB1-cKO astrocytes failed to increase intracellular Ca 2+ in response to baclofen (n = 88 astrocytes from eight slices; p=0.12; Wilcoxon rank-sum test), and IN stimulation (n = 71 astrocytes from eight slices; p=0.14; Wilcoxon rank-sum test), but were activated by ATP (n = 82 astrocytes from eight slices; p<0.001; Wilcoxon ranksum test). (F) Baclofen-evoked currents in CA1 pyramidal neurons from wild-type and GB1-cKO mice before and after CGP55845 application (Wild-type: from 13.73 ± 2.12 to 2.09 ± 1.22 pA, before and after CGP55845; n = 7; paired t test; p<0.001; GB1-cKO: 10.76 ± 1.0 to 2.85 ± 0.89 pA; before and after CGP55845; n = 8; p<0.001). (G) Synaptic responses evoked by minimal stimulation (15 consecutive stimuli; gray traces), and averaged EPSCs (50 consecutive stimuli; black traces) before and after evoking bursts of interneuron APs in wild-type and GB1-cKO mice. (H) Relative changes of synaptic parameters from basal (black bars) induced by bursts of interneuron APs (white bars; n = 5; p<0.001; paired t test), and after CGP55845 application (n = 3; p=0.38; paired t test) in wild-type and GB1-cKO mice (n = 4; p=0.38; paired t test). **p<0.01; # p<0.001. DOI: 10.7554/eLife.20362.012 neurotransmission was inhibited by interneuron single APs through activation of GABA A receptors (characterized by fast kinetics) (Farrant and Nusser, 2005), it was enhanced by bursts of interneuron APs through additional and concurrent slower mechanisms (i.e., astrocyte GABA B receptor activation, G-protein-mediated intracellular Ca 2+ mobilization, astrocytic glutamate release, and presynaptic group I mGluR activation) that persisted after the interneuron burst (i.e., during the interneuron down-state). Therefore, different patterns of interneuron activity determine diverse consequences on  Representative LFP recordings and corresponding analysis of theta-phase and gamma-amplitude relation for control littermate (left) and GB1-cKO (right) mice. The raw signals (black) were high-pass filtered (1st row; grey and red, respectively) and then computed to extract the theta phase (second row) and gamma envelope (third row; grey and red, respectively) for control and GB1-cKO mice. (C) Normalized LFP power spectrum analysis for theta (4-8 Hz) and gamma frequencies (30-50 Hz, low gamma; 70-90 Hz, high gamma) in control littermate (n = 6) and GB1-cKO mice (n = 6) after whisker stimulation. Inset, Relative power changes for GB1-cKO and control littermate mice (theta band, p=0.007; unpaired t test; low gamma, p=0.042; Wilcoxon rank-sum test; and high gamma oscillations, p=0.738; Wilcoxon rank-sum test).
(D) Gamma-amplitude modulation by theta-phase for wild-type (left) and GB1-cKO mice (right), before (control) and after whisker stimulation (post stim). Note the enhancement of theta-gamma coupling in wild-type after stimulus that does not appeared in GB1-cKO mice. (E) Normalized Phase-Lock Value (PLV), either in control and after stimulation for wild-type (p=0.017; paired t test), and GB1-cKO mice (p=0.62; paired t test). See also excitatory synaptic transmission by direct interneuron-principal cell mechanisms (fast inhibition) and by novel interneuron-astrocyte signaling mechanisms (slow potentiation). In the latter mechanisms, astrocytes decode interneuron activity and transform inhibitory GABAergic signals into excitatory glutamatergic signals that enhance synaptic transmission. In addition, we found a subset of synapses that express both phenomena, switching from inhibition to slow potentiation according to the inhibitory GABAergic tone. Thus, the existence of new mechanisms to fine-tune the output of local synapses by astrocyte activity might contribute to control the excitation-inhibition balance at the hippocampal circuits. GABA B receptors have been proposed to inhibit transmitter release and modulate plasticity via presynaptic and postsynaptic mechanisms (Ulrich and Bettler, 2007). Thus, the activation of GABA B autoreceptors requires strong stimulus intensities, consistent with their distant location from the release sites and probably requires pooling of synaptically released GABA to be activated (Ulrich and Bettler, 2007). Here, we show that the activation of interneruons evoked synaptic inhibition independent of GABA B signaling, because it was present after their blockage with CGP55845 but was abolished by the GABA A receptor antagonist picrotoxin (Figures 1, 4 and 5). Although, GABA spillover and activation of presynaptic GABA B cannot be totally excluded, the pharmacological data and results obtained from Ip3r2 -/mice, where GABA B receptors in neurons and astrocytes are intact, and the new transgenic mouse lacking GABA B Rs specifically in astrocytes ( Figure 6) indicate that GABA B autoreceptors do not contribute to the observed inhibition of synaptic transmission.
The activity of hippocampal interneurons has been shown to activate astrocytes that induce a long-lasting enhancement of inhibitory synaptic transmission through glutamate release and activation of kainate receptors in inhibitory terminals (Kang et al., 1998). Our results indicate that the glutamate released from astrocytes stimulated by interneurons can also access mGluRs in excitatory synapses to transiently enhance their synaptic efficacy, suggesting that a single gliotransmitter may have multiple effects depending on the site of action. In addition to glutamate, hippocampal astrocytes may also release ATP, which is converted to adenosine that depress synaptic transmission through activation of A1 adenosine receptors (Andersson et al., 2007;Chen et al., 2013;Pascual et al., 2005;Serrano et al., 2006). This mechanism has been proposed to be triggered by high frequency stimulation of SC that sequentially activates interneurons and astrocytes leading to the heterosynaptic depression in the hippocampal CA1 region (Andersson et al., 2007;Serrano et al., 2006). Then, astrocytes immersed in the same circuit may be stimulated by interneurons to release different gliotransmitters (i.e., glutamate or ATP) that influence synaptic transmission in different forms. The specific circumstances leading to the release of particular gliotransmitters are unknown. While we directly stimulated interneurons to elicit specifically identified firing patterns, the activity evoked in interneurons during high frequency stimulation of SC was unknown. Since our results indicate that astrocytes decode interneuron activity, it is possible that the different regulatory effects were due to differences in the astrocyte signaling evoked by interneurons. This represents an additional example of the importance of the context-specificity of signaling in the reciprocal communication between neurons and astrocytes (for a detailed discussion, see ). Because the molecular signaling governing interneuron-astrocyte mediated effects shown here were studied in juvenile animals, and considering the developmental receptor expression profiles (Sun et al., 2013), whether these molecular pathways and complex features of interneuron-astrocyte signaling are conserved in the adult brain need further attention. In addition, present data cannot discard that residual Ca 2+ events might occur in the fine process of the GB1-cKO astrocytes, as they have been found for Ip3r2 -/mice (Srinivasan et al., 2015); however, the existence of those events would not be sufficient to induce the synaptic potentiation observed after interneuron stimulation ( Figure 3E,F, and Figure 6G,H). Thus, these data suggest that although the astrocyte-synaptic interactions might primarily take place at the astrocyte processes, the synaptic plasticity induced by interneuron-astrocyte communication is a highly regulated phenomenon that requires the active contribution of astrocyte somatic Ca 2+ signaling.
Numerous functions of interneurons crucially depend on the fast and temporally precise conversion of an excitatory synaptic input into an inhibitory synaptic output. As a result, interneurons provide 'phasic' inhibition to the neuronal network, which is involved in the emergence of fast brain rhythms (Csicsvari et al., 2003) and synaptic plasticity during critical periods of circuit formation (Hensch, 2005) that jointly contribute to the maturation of cognitive functions. Here, we imposed to interneurons an oscillatory activity of 3 Hz with intra-faster oscillations of 35-50 Hz ( Figure 5A). That rhythm, which is similar to physiological slow delta/theta (2-8 Hz) and low frequency gamma (25-40 Hz) bands found in the hippocampus, evoked fast episodes of inhibition nested within a long lasting enhancement of excitatory transmission. To unambiguously determine the locus and contribution of GABA B receptors to the synaptic transmission and network coding properties, and considering that GABA B receptors are ubiquitous, we took advantage of a conditional astrocyte-specific GABA B knockout mouse. The data shown here indicate that GB1-cKO mice showed a down-regulated GABA B -astrocyte Ca 2+ signaling and the absence of the interneuron-mediated potentiation of excitatory synaptic transmission. In vivo recordings in anesthetized mice remarkably displayed changes in the hippocampal oscillatory activity pattern. We found that theta and gamma oscillations that are associated with cognitive processes, as well as the coupling between these rhythms were compromised in mice lacking GABA B -astrocyte signaling, indicating a critical role of astrocyte signaling in higher-order information coding. Thus, recent studies have shown how astrocyte activity can impact the state of neuronal circuits by regulating the generation of neuronal UP states (Poskanzer and Yuste, 2011), and it has been related to brain rhythms (Poskanzer and Yuste, 2016), such as slow cortical oscillations (<1 Hz) associated with nonrapid eye movement (NREM) sleep (Fellin et al., 2009). Disruption of astrocytic vesicular release has been found crucial for gamma oscillatory hippocampal activity with significant impact in recognition memory tasks (Lee et al., 2014). Therefore, giving the important role of interneuron activity and hippocampal oscillations in coding neural information, such as the control of the theta phase-modulation of gamma power that correlates with memory performance (Buzsáki and Moser, 2013;Lisman and Jensen, 2013), the signaling between interneurons and astrocytes, which provides novel properties to the interneuron effects on excitatory synapses within the network, adds more complexity to neuronal information processing.
Additionally, the GABA B receptor pharmacological blockade has been shown to enhance cognitive task performance by activating hippocampal theta and gamma rhythms in behaving rats (Leung and Shen, 2007); in contrast, a recent study shows that GABA B deletion in glutamatergic terminals disrupts the acquisition and learning of hippocampal tasks, demonstrating their contribution to learning-dependent synaptic changes and network dynamics (Jurado-Parras et al., 2016). Here, present data from astrocyte Gabbr1 knockout mice show a partial but significant decrease of stimulus-induced theta-gamma oscillations and coupling, and highlight the intricate roles of GABA B receptors in regulating the neural network operation considering their specific cellular targets. Together, these data suggest that astrocytes might be directly related to critical brain rhythms and their cognitive functions (Lee et al., 2014). Considering that the promoter used GLAST can be expressed by progenitor cells in adult brain (Slezak et al., 2007), we cannot rule out a partial contribution of derived cells from those progenitors to the in vivo observed responses. Some evidence have shown that acute brain slices might undergo hypoxic conditions causing reactive changes in astrocytes (Takano et al., 2014), and a downregulation of GABA A receptor expression (Zonouzi et al., 2015); however, since the study of molecular individualities of the interneuron-astrocyte signaling show limitations that need to be explored ex vivo, these associated alterations and their potential influence cannot not be excluded from the observed responses.
The current view for the mechanisms underlying brain diseases is largely based on neuronal dysfunctions, but increasing evidence suggests that also disturbances of astrocyte-neuron interactions are related to brain disorders (Seifert et al., 2006;Takano et al., 2009). Because alterations of the excitatory/inhibitory balance might underlie different brain states and diseases, such as epileptic activity, schizophrenia, and mood disorders, present results showing astrocytic contribution to the excitatory drive, i.e., transformation of inhibitory signals into an excitatory enhancement, indicate that interneuron-astrocyte signaling might be involved in the excitatory/inhibitory unbalance present in particular brain states.
Taking together, present findings reveal novel and unexpected consequences of interneuron signaling in neuronal network activity through stimulation of astrocytes. Astrocytes decode the temporal activity of neurons and transform neuronal signals to impact circuit function through novel mechanisms based on different signaling and time scales. Thus, interneuron activation of astrocytes through the control of oscillatory activity is directly involved in the coding of neuronal circuits and their functional properties, suggesting that brain the function results from the dynamic interplay of Astrocyte-Neuron networks.

Materials and methods
All the procedures for handling and sacrificing animals followed the European Commission guidelines for the welfare of experimental animals (2010/63/EU), US National Institutes of Health and the Institutional Animal Care and Use Committee at the University of Minnesota (USA). The use of astrocyte-specific GABBR1 knockout mice was approved by the Saarland state´s 'Landesamt fü r Gesundheit und Verbraucherschutz' in Saarbrü cken/Germany (animal license number 72/2010). Animals of both genders were used, and were housed in standard laboratory cages with ad libitum access to food and water, under a 12:12 hr dark-light cycle in temperature-controlled rooms.

Electrophysiology
Electrophysiological recordings from interneurons, CA1 pyramidal neurons and astrocytes were made using the whole-cell configuration of the patch-clamp technique. Patch electrodes had resistances of 3-10 MW when filled with the internal solution that contained (in mM): K-Gluconate 135, KCl 10, HEPES 10, MgCl 2 1, ATP-Na 2 2 (pH = 7.3, with KOH). Recordings were obtained with PC-ONE amplifiers (Dagan Corporation, Minneapolis, MN, USA). Fast and slow whole-cell capacitances were neutralized and series resistance was compensated ( » 70%). Recordings were rejected when the access resistance increased >20% during the experiment. Recordings from CA1 pyramidal neurons were performed in voltage-clamp conditions and the membrane potential was held at À70 mV to record Schaffer collateral (SC) afferents-evoked EPSCs. CA1 interneurons were recorded in current-clamp conditions. Signals were fed to a Pentium-based PC through a DigiData1440 interface board (Molecular Devices, Sunnyvale, CA, USA). The pCLAMP 10 software (Molecular Devices) was used for stimulus generation, data display, acquisition, storage and analysis. Experiments were performed at room temperature (22-24˚C). For astrocyte network loading, the holding potential was À80 mV. BAPTA (40 mM) and biocytin (0.1%) intracellular astrocyte filling was performed for 20-30 min (internal solution contained (in mM): BAPTA-K 4 40, NaCl, 8, MgCl 2 1, HEPES 10, GTP-tris salt 0.4, ATP-Na 2 2; pH = 7.3, with KOH.
Slices were then fixed and biocytin was revealed by Alexa488-Streptavidin ( Figure 3D), showing the wide area covered by the intracellular biocytin loading, and confirming the broad downregulation of Ca 2+ signals by BAPTA intracellular filling astrocytes (cf. [Poskanzer and Yuste, 2011;Serrano et al., 2006]).
Baclofen (2 mM) was locally applied through a micropipette (10 s duration) in the presence of TTX (1 mM) to induce GABA B -mediated currents in CA1 pyramidal neurons (holding potential set to À30 mV) from astrocyte-specific GABA B receptor knockout mice (GB1-cKO) and control littermate mice.
Minimal stimulation was achieved using theta capillaries (2-5 mm tip diameter) filled with ACSF, and placed in the stratum radiatum to stimulate SC afferents. Single pulses (250 ms duration) or paired pulses (50 ms interval) were delivered at 0.5 Hz by stimulator S-900 (Dagan). The stimulus intensity (1-15 mA) was adjusted to meet the conditions that putatively stimulate a single or very few synapses (Navarrete and Araque, 2010;Navarrete et al., 2012;Perea and Araque, 2007), and was unchanged for the entire experiment. The recordings that did not meet these criteria and synapses that did not show amplitude stability of EPSCs were rejected. The synaptic parameters analyzed were: synaptic efficacy (mean EPSC peak amplitude of all evoked responses, including failures), synaptic potency (mean EPSC peak amplitude of successful responses when failures are excluded), the success rate of neurotransmitter release (calculated as the ratio between the number of effective EPSCs divided by the total number of stimuli), and paired-pulse ratio (PPR, 50 ms pulse interval) (Fernández de Sevilla et al., 2002;Stevens and Wang, 1994). The responses and failures were identified by visual inspection and PPF was quantified as second EPSC/1 st EPSC. Basal values were recorded 10 min before the stimulus (e.g., Figure 1-figure supplement 1). Data points represent the mean value of 15 consecutive EPSCs unless indicated and were plotted over time (e.g., Figure 1B).
Interneuron single action potentials (APs) or bursts of APs were evoked by either 15 ms or 700 ms depolarizing pulses (200-300 pA), respectively, that were applied 10 ms before the SC stimulation. Single APs and SC stimuli were paired for 3 min every 4 s. Protocol of pairing bursts of APs and SC stimuli was delivered 3 times every 2 s. To study the dynamic interplay between interneuronastrocyte activity and excitatory synaptic transmission, bursts of APs were applied to interneurons by repetitive depolarizations (3 Hz, 30 s). SC stimuli were then phase-locked at either the interneuron depolarization (up-EPSC) or the resting state (down-EPSC; protocol shown in Figure 5A). In Figure 5, Figure 5-figure supplements 1 and 2 each data point represents the simple moving average (Jadhav et al., 2012;O'Connor et al., 2005) of 15 consecutive EPSCs and were plotted over time. Bar graphs represent the mean value of the EPSCs at the 20-30 s periods after starting the protocol (e.g., Figure 5B).

Calcium imaging
Ca 2+ levels in astrocytes were monitored by fluorescence microscopy using the Ca 2+ indicator Fluo-4-AM. Slices were incubated with Fluo-4-AM (2-5 mL of 2 mM dye were dropped over the hippocampus, attaining a final concentration of 2-10 mM and 0.01% of pluronic) for 20-30 min at room temperature. In order to confirm the specific recording of Ca 2+ signals in astrocytes, animals were injected intraperitoneally with sulforhodamine 101 (SR101; 100 mg/kg) 2 hr before sacrificed. In these conditions, astrocytes were specifically loaded with SR101 (e.g., Figure 3A) (Nimmerjahn et al., 2004;Perez-Alvarez et al., 2014). Additionally, astrocytes were confirmed by their electrophysiological properties (Araque et al., 2002;Nimmerjahn et al., 2004). Astrocytes were then imaged using a CCD camera (ORCA-235; Hamamatsu, Japan) attached to the microscope (Olympus BX51WI). Cells were illuminated during 100-500 ms with a xenon lamp at 490 nm using a monochromator Polychrome V (TILL Photonics, Grä felfing, Germany), and images were acquired every 0.5-1 s. The monochromator and the camera were controlled and synchronized by the IPLab software that was also used for quantitative epifluorescence measurements. Analysis of astrocyte Ca 2+ levels were restricted to the region of the cell body and Ca 2+ variations were estimated as changes in the fluorescence signal over the baseline (4F/F 0 ). The astrocyte Ca 2+ signal was quantified from the probability of occurrence of a Ca 2+ elevation (termed as Ca 2+ transient), calculated as the number of Ca 2+ transient grouped in 5 s bins recorded from the astrocytes in the field of view (6-12 astrocytes per analyzed region) (Navarrete and Araque, 2010), and mean values were obtained by averaging different experiments. To test the effects of interneuron activity on Ca 2+transient probability under different conditions, the respective mean basal (15 s before the stimulus) and maximum Ca 2+ transient probability (recorded 15 s after interneuron stimulation) from !5 slices per condition were averaged and compared (e.g., Figure 3B). Ca 2+ responses from different slices were normalized calculating the 'Ca 2+ transient probability index' as: [(Ca 2+ transient probability after stimulus) À (Ca 2+ transient probability before stimulus)] / [(Ca 2+ transient probability after stimulus) + (Ca 2+ transient probability before stimulus)] (e.g., Figure 3C).
In some experiments the genetically encoded Ca 2+ indicator GCaMP3 specifically expressed in astrocytes was used to monitor Ca 2+ signaling in the GB1-cKO mice by using confocal microscopy (Olympus FV300) ( Figures 6C,D), and analyzed as described. Local application of ATP (10 mM) and baclofen (10 mM) were delivered by 5 s duration pressure pulses through a micropipette.
Conditional, astrocyte-specific GABA B receptor knockout mice (GB1-cKO mice) Functional GABA B receptor ablation was investigated in conditional, astrocyte-specific GABA B receptor knockout mice (GB1-cKO), generated by crossbreeding Gabbr1 fl/fl (MGI:3512742) (Haller et al., 2004) with GLAST-CreERT2 knockin mice (MGI:3830051) (Mori et al., 2006). In some of the experiments, mice with astrocytes-specific expression of the genetically encoded Ca 2+ indicator GCaMP3 were used. For that purpose R26-lsl-GCaMP3 mice (JAX #014538) (Paukert et al., 2014) were crossbred to GB1-cKO and control mice. The selective deletion of the receptor subunit GABBR1 is sufficient to completely block functional GABA B receptor activity . To induce DNA recombination in GLAST-CreERT2xGABA B fl/fl or GLAST-CreERT2xR26-lsl-GCaMP3 mice (Paukert et al., 2014), tamoxifen (10 mg/ml corn oil, Sigma, St. Louis, USA) was intraperitoneally injected into 3-week-old mice on three consecutive days (100 mg/kg per body weight). 21 days after the first injection, mice were started to be analyzed. All mouse lines were maintained in the C57BL/6N background.

Immunohistochemistry
The animals were anesthetized with Ketamine/Rompun (1.4% ketamine, 0.2 xylazin, 0.9% NaCl; 5 ml/ kg per body weight) and intracardially perfused with ice cold ACFS and subsequently with 4% paraformaldehyde (PFA) in 0.1 M phosphate buffer (pH 7.4). The brain was removed, dissected into the two hemispheres, and post fixed for 4-6 hr in 4% PFA in 0.1 M phosphate buffer (pH 7.4) at 4˚C. The fixed brain tissue was cut in phosphate buffered saline (PBS) into sagittal sections (50-70 mm thickness) at a Leica VT1000S vibrato (Leica, Nussloch, Germany). These sections were collected in 24-well tissue culture plates containing PBS. Vibratome sections were incubated for one hour in blocking buffer (0.3% Triton X-100, 5% horse serum in PBS) at RT (room temperature). The primary antibodies were diluted in the blocking solution and the sections were incubated overnight at 4˚C. As marker for GABAergic interneurons mouse anti-GAD67 (RRID:AB_2278725; 1:500) was used. For astrocyte labeling the following antibodies were used: chicken anti-GFAP (RRID:AB_921444; 1:1000), and rabbit anti-GLAST (RRID:AB_304334; 1:250). GABA B receptors were stained with a guinea pig anti-GABABr1 (RRID:AB_1587048; 1:500). The slices were washed three times for ten min each in 1xPBS. The secondary antibody was diluted in the secondary antibody buffer (2% horse serum in PBS) and incubated for 2 hr at room temperature. Secondary antibodies were donkey anti-mouse, anti-goat, and anti-rabbit (1:2000) conjugated with Alexa488, Alexa546, Alexa555, Alexa633, and purchased from Invitrogen (Thermo Fisher Scientific Inc). The sections were finally washed for 3 times with 1xPBS (10 min) and mounted in Aqua polymount (Polysciences). For astrocytic network labeling, after biocytin filling slices were fixed in 4% PFA in 0.1 PBS (pH 7.4) at 4˚C. Biocytin was visualized with Alexa488-Streptavidin (RRID:AB_2315383; 1:500) applied in the staining protocol described above for 48 hr.

Microscopic analysis and quantification
Confocal images were recorded by laser scanning microscopy (LSM 710, Zeiss, Carl Zeiss AG, Jena) using a 40x objective (Plan-Aprochomat 40x/1, 4 Oil DIC (UV) VIS-IR M 27). For excitation of fluorescent dyes, a Lasos Argon laser (454 nm to 514 nm) and a Helium-Neon laser (543 nm, 633 nm) were used. Z-stacks of images were taken at 0.5 mm intervals and processed with ImageJ using the JACoP v2.0 colocalization plugin (Cordelières and Bolte, 2014). In brief, the deletion of Gabbr1 was determined as a reduction of its immunolabel within the respective channel of the astroglial glutamate transporter GLAST. The overlap coefficient and Mander's coefficient M2 were determined. Although both coefficients have their unique limitations, both indicated a significant and astrocytespecific reduction of Gabbr1 and were plotted in Figure 6B (9 sections from 2 GB1-cKO and 11 sections from two control mice; 2-sided; unpaired t test). The analysis of co-localization probably underestimates the Gabbr1 removal, since the spatial resolution in single optical sections is less than the size of the fine astrocyte processes contacting presynaptic terminals that are Gabbr1-positive as well.

Quantitative real-time PCR (qRT-PCR)
Levels of messenger RNA (mRNA) and genomic DNA were detected by reverse transcriptase PCR. Hippocampi of 7 GB1-cKO and seven control mice (seven weeks old) were removed from the skull, homogenized (Precellys homogenizer, peqlab, Erlangen, Germany) and divided for RNA extraction with RNeasy mini kit (QIAGEN, Hilden, The Netherlands) as well as for DNA analysis RNA/DNA ALL Prep-Kit (QIAGEN, Hilden, The Netherlands). Successful gene recombination was determined by quantifying the loss of the loxP flanked gene region. Primers were located closely upstream and downstream of the 5' loxP site. Control and cKO were homozygously floxed for the Gabbr1 locus (Gabbr1 fl/fl ); controls were wild type for the GLAST locus (GLAST +/+ ) and GB1-cKOs were heterozygous for the CreERT2 transgene in the GLAST locus (GLAST CreERT2/+ ). Since only non-recombined alleles were amplified, reduction of the respective PCR signal indicated successful recombination. Values (4CT) of GB1-cKO animals were normalized to the mean 4CT values of control animals. For quantification of the PCR products, the fluorescent dye EvaGreen (Axon) was used. PCR runs were performed using the CFX96 Real-Time PCR Detection System (BioRad). All reactions were carried out in triplicates. Neuregulin one type III (NrgIII) and b-actin were used as endogenous gene controls. Data normalization and analysis were performed with the qbase+ real time PCR data analysis software (Biogazelle) based on the DDCT-method.
Primer sequences for CDNA analysis by qRT-PCR were as follows (in 5' to 3' direction): ATPase forward GGA TCT GCT GGC CCC ATA C; ATPase reversed CTT TCC AAC GCC AGC ACC T, b-Actin forward CTT CCT CCC TGG AGA AGA GC; b-Actin reversed ATG CCA CAG GAT TCC ATA CC; Gabbr1 forward CGA AGC ATT TCC AAC ATG AC; Gabbr1 reversed CAA GGC CCA GAT AGC ATC ATA. Primer sequences for genomic DNA were as follows: NRGIII forward GTG TGC GGA GAA GGA GAA AAC T; NRGIII reversed AGG CAC AGA GAG GAA TTC ATT TCT TA; b-Actin forward CTG CTC TTT CCC AGA CGA GG; b-Actin reversed AAG GCC ACT TAT CAC CAG CC; Gabbr1 forward CAG TCG ACA AGC TTA GTG GAT CC, Gabbr1 reversed TCC TCG ACT GCA GAA TTC CTG.
In vivo recordings GB1-cKO mice and wild-type littermates (12-16 weeks) were placed in a stereotaxic frame under urethane anesthesia (1.8 g/kg, intraperitoneal injection), constantly monitored for body temperature and breathing rate, and kept warm with a heating pad. Electrodes were placed stereotaxically according to the atlas (Paxinos and Franklin, 2012). Local field potentials (LFP) were recorded through stainless steel macroelectrodes (1 MW) placed in the CA1 layer (AP, À2; L, 1.4; V, 1.1 mm from Bregma) and amplified (Differential AC Amplifier Model 1700, A-M System), bandpass filtered between 0.1 Hz and 500 kHz, and digitized at 100 kHz (PowerLab 4/25 T and LabChart, ADInstruments) running in a PC for direct visualization and storage. Then, two nichrome stimulating electrodes (Isolated Pulse Stimulator Model 2100, A-M Systems) were placed in the vibrissae. After stabilization and basal activity recordings, an electrical stimulus (10 Hz, 10 s duration at 10 V) was applied to vibrissae. Three stimuli were applied with an interstimulus period of !5 min.
Six epochs (five second bins) during one minute in basal conditions were analysed. Also, the first 10 s, divided in 5 sec-bins, starting at the end of each stimulus were selected. Epoch was stored in a new file and converted to an adequate format to perform the spectral analyses (Clampfit 10.2, MDS Analytical Technologies). Spectral analyses for each bin were assessed by fast Fourier transformation through the Hamming window with 50% overlap, obtaining the power density (V 2 .Hz À1 ) with a spectral resolution of 0.38 Hz, from 0.38 to 100.3 Hz. Since animals had different levels of baseline power density, the power values for each frequency were normalized as a percentage of the total power density recorded before computing group results. After normalization six epochs were averaged for basal and post-stimuli condition and compared between control littermate and GB1-cKO mice. We selected the following frequency bands: theta, 4-8 Hz; low gamma, 30-50 Hz and high gamma, 70-90 Hz.
For phase-amplitude coupling (PAC) analysis each 5 s bin was converted to text format to perform the computation trough MATLAB (The MathWorks, Inc.). The process was performed by a custom-made script on MATLAB (https://github.com/abdel84/). Raw signal was decimated to a sample rate of 1 kHz, then an elliptical filter was applied to remove frequencies below 3 Hz and two additional bandpass filters for both Theta (4-8 Hz) and Gamma (30-80 Hz) bands. The Theta phase and the Gamma amplitude, respectively, were extracted and computed to obtain their time series using the standard Hilbert transform as described previously (Tort et al., 2010) and to obtain the Phase-Locking Value (PLV). This index represents the degree to which the Gamma amplitude is comodulated with the Theta phase and ranges between 0 and 1, with higher values indicating stronger PAC interactions (Tort et al., 2010). To calculate the mean vector of PLV, circular statistics analysis was performed by using CircStat toolbox 19 and then normalized by Fisher's Z Transformation, to apply regular statistical analysis: z' = 0.5 [ln (1+r) À ln (1À r)].

Statistical analysis
The normality test was performed before applying statistical comparisons, which were made using non parametric Wilcoxon Rank-sum Test and parametric Student's t tests as deemed appropriate. Two-tailed, unpaired or paired t test was used for comparisons unless indicated. Data are expressed as mean ± standard error of the mean (SEM). When a statistical test was used, the precise two-sided P value and the test employed are reported in the text and/or figure legends. Statistical differences were established with p<0.05 (*), p<0.01 (**), and p<0.001 (***, #). Blind experiments were not performed in the study but the same criteria were applied to all allocated groups for comparisons. Randomization was not employed. The sample size in whole-cell recording experiments was based on the values previously found sufficient to detect significant changes in hippocampal synaptic strength in past studies from the lab. For in vivo recordings an N of 3 repetitions of stimuli were applied, and independent recordings were summarized from six animals per condition, which provided sufficient statistical power while trying to minimize the number of animals sacrificed. P) to GP; Juan de la Cierva Program (MINECO, JCI-2011-09144 andIJCI-2014-19136)  The funders had no role in study design, data collection and interpretation, or the decision to submit the work for publication.