Subthalamic Nucleus Electrical Stimulation Modulates Calcium Activity of Nigral Astrocytes

Background The substantia nigra pars reticulata (SNr) is a major output nucleus of the basal ganglia, delivering inhibitory efferents to the relay nuclei of the thalamus. Pathological hyperactivity of SNr neurons is known to be responsible for some motor disorders e.g. in Parkinson's disease. One way to restore this pathological activity is to electrically stimulate one of the SNr input, the excitatory subthalamic nucleus (STN), which has emerged as an effective treatment for parkinsonian patients. The neuronal network and signal processing of the basal ganglia are well known but, paradoxically, the role of astrocytes in the regulation of SNr activity has never been studied. Principal Findings In this work, we developed a rat brain slice model to study the influence of spontaneous and induced excitability of afferent nuclei on SNr astrocytes calcium activity. Astrocytes represent the main cellular population in the SNr and display spontaneous calcium activities in basal conditions. Half of this activity is autonomous (i.e. independent of synaptic activity) while the other half is dependent on spontaneous glutamate and GABA release, probably controlled by the pace-maker activity of the pallido-nigral and subthalamo-nigral loops. Modification of the activity of the loops by STN electrical stimulation disrupted this astrocytic calcium excitability through an increase of glutamate and GABA releases. Astrocytic AMPA, mGlu and GABAA receptors were involved in this effect. Significance Astrocytes are now viewed as active components of neural networks but their role depends on the brain structure concerned. In the SNr, evoked activity prevails and autonomous calcium activity is lower than in the cortex or hippocampus. Our data therefore reflect a specific role of SNr astrocytes in sensing the STN-GPe-SNr loops activity and suggest that SNr astrocytes could potentially feedback on SNr neuronal activity. These findings have major implications given the position of SNr in the basal ganglia network.


Introduction
The pars reticulata of the substantia nigra (SNr) is the main output nucleus of the basal ganglia network and conveys the final output signal to the thalamus and brain stem [1]. The GABAergic neurons of the SNr are tonically active and display spontaneous activity (around 10 Hz) that is partially independent of synaptic inputs [2]. Thus, the brain areas receiving inputs from the basal ganglia are, by default, under strong tonic inhibition. Brief changes in firing, often associated with movement, result in the disinhibition or further inhibition of the target nuclei neurons [3]. The SNr may therefore be considered as a gateway in the transmission of information to the motor and cognitive systems. This output nucleus mostly integrates inputs from the striatum, the external segment of globus pallidus (GPe) and the subthalamic nucleus (STN). Among its inputs, the excitatory STN and the inhibitory GPe play a particularly important role in the regulation of SNr activity as they form a feedback system engaged in synchronized bursting and constituting a basal ganglia pacemaker [4]. The STN occupies a crucial position in the functional architecture of the basal ganglia network since it is the only glutamatergic nucleus [5,6]. Abnormal hyperactivity of the STN, consecutive to the loss of dopaminergic cells in the substantia nigra pars compacta, is thought to play a critical role in the expression of Parkinson's disease symptoms [7,8]. High-frequency stimulation of the STN (STN-HFS) is currently the main surgical therapy for treatment of advanced Parkinson's disease. The cellular mechanisms of STN-HFS are still debated but it has been proposed that HFS could inactivate STN neurons [9,10] or change their abnormal firing mode into recurrent bursting activity [11]. This normalization could reduce the hyperactivity of basal ganglia output structures, such as SNr [12,13,14]. STN-HFS may produce these effects by activating the axons of STN neurons, STN afferents or fibers passing close to the stimulation site, leading to both glutamate and GABA release within SNr [15,16,17]. As the final stage of information processing within the basal ganglia occurs mainly in the SNr, studying the regulation of the activity of this nucleus in

Results
Preservation of STN-GPe-SNr pathways in rat sagittal brain slices The pallido-nigral and subthalamo-nigral loops have a crucial influence on the spontaneous activity of SNr neurons and we wondered about their impact on astrocyte activity. The adult rat brain atlas suggested that these nuclei were present on the same sagittal plane, from 2.4 to 2.9 mm laterally to the midline, in adult rat (Fig. 1A) [33]. We performed such sagittal brain slices and introduced lipophilic tracers (DiI crystals) into the SNr or the STN to confirm that the subthalamo-nigral and pallido-nigral pathways were indeed preserved. The dye diffused for four weeks, resulting in retrograde and anterograde labeling of the persistent subthalamo-pallido-nigral pathways. When a crystal was placed in the SNr (Fig. 1B), bundles of labeled fibers could be seen coursing from the SNr towards the STN and probably also towards the GPe beacuse some of these fibers forked close to or within the STN [34] (Fig. 1C). We also detected some labeled soma within the STN (Fig. 1C, inset) meaning that at least some projecting STN neurons were intact. When a crystal was placed in the STN, two pathways running in opposite direction were observed, one in the rostrocaudal direction and the other one (usually more intense) in the caudorostral direction, suggesting connections towards SNr and GPe respectively (not shown). These observations confirmed that the glutamatergic subthalamo-nigral pathway and the GABAergic pallido-nigral pathway were, at least partly, preserved in these slices.
On acute slices, incubation with propidium iodide and Hoechst 33342 showed that 8065% of the SNr cells were still alive 5 hours after slicing (n = 16 slices). Electrophysiological experiments were carried out to record the typical pace-maker activity of nigral GABAergic and dopaminergic neurons [2] in these sagittal slices (n = 15 neurons in a total of 7 slices). Patch-clamp recordings performed in sagittal plane by other groups had already shown that SNr neurons were affected by a complex pattern of excitatory glutamatergic and inhibitory GABAergic synaptic currents [34,35]. Thus, in these conditions, GABAergic pallidal activity, glutamatergic STN activity and/or GABAergic intranigral activity may potentially contribute to the basal calcium activity of astrocytes in the SNr.

Astrocytes constitute the main cell population in the SNr
The SNr contains two types of neurons: GABAergic and dopaminergic neurons [2]. GABAergic neurons predominate (about 90% of SNr neurons, [36]), forming the major output from the basal ganglia [37]. With an average of 17956358 total cells/mm 2 (n = 7 slices from 3 rats, estimated by nuclei staining with TO-PRO 3) throughout the whole SNr, immunohistochemical stereological analysis showed that neurons (as revealed by NeuN immunostaining) were heterogeneously distributed within the SNr and accounted for 1263% of total SNr cells ( Fig. 2A; n = 16 tissue sections from 4 rats). When we quantified the astrocyte population with S100B immunostaining, we found 4263% of S100B-positive cells in the SNr ( Fig. S1A; n = 6 tissue sections from 2 rats). As described in other brain structures [38,39], however, we observed that S100B or GFAP are not coexpressed in all astrocytes within the SNr (e.g. 1565% of S100Bpositive cells were GFAP-positive and 5163% of GFAP-positive cells were S100B-positive, n = 3 slices; Fig. S1B). Thus, this cell population might be underestimated on the basis of immunostainings. We therefore used an alternative strategy to precisely and specifically determine the proportion of astrocytes in the SNr. Intravital sulforhodamine dyes are known to be specifically targeted to the astrocyte population when injected intravenously in rats [40,41]. Sulforhodamines cross the blood brain barrier and astrocytes are stained directly via their endfeet. We then achieved sagittal living brain slices and loaded them with intravital Hoechst 33342. Sulforhodamine 101 (SR101) staining showed astrocytes to be uniformly distributed throughout the SNr (Fig. 2B) with similar morphology when compared to S100B immunostaining (Fig.  S1A). We found that SR101-stained astrocytes accounted for 65.067.2% of total SNr cells (n = 6 slices from 3 rats; 48 fields; Fig. 2C).
A major advantage of SR101 staining is that it can be combined with Fluo-4 AM loading in acute living slices for calcium imaging. Thus, we showed that among these SR101-stained astrocytes, 76.266.6% were loaded with Fluo-4 (n = 6 slices from 2 rats; Fig. 2D). Moreover, in our conditions, 98.062.2% small (less than 10 mm) Fluo-4 loaded cells were astrocytes (SR101 stained cells; n = 6 slices from 2 rats; Fig. 2D), indicating that almost all the cells subsequently recorded were astrocytes. Microscopic analysis of the Fluo-4 labeled slices confirmed that, compared to neurons, the astrocyte cell bodies were smaller and more strongly labeled by Fluo-4, probably due to their greater ability to take up or hydrolyze AM-ester dyes. This allowed us to distinguish between the two cell populations on morphological criteria, as already reported for mouse cortex [42].

Astrocytes display spontaneous calcium activities in physiological conditions
In order to assess the impact of astrocytes on SNr output activity, we first recorded spontaneous calcium activity using calcium imaging on rat sagittal brain slices at the subthalamo-pallido-nigral level. On average, individual movies (field of view around 6006300 mm) included 66.3615.3 astrocytes (n = 54 slices; neurons were excluded on the basis of morphological criteria). In basal conditions, 30.8614.5% of the SNr dye-loaded astrocytes displayed spontaneous calcium activity during a five-minute recording period (Fig. 3A-D; n = 54 slices from 29 animals, 3330 cells). This proportion increased with the recording period (64% astrocytes displayed calcium activity when observed for a 30 minute-period, Fig. 3E) suggesting that most of astrocytes display spontaneous activity. In active astrocytes, the calcium signal had a mean duration of 23.566.8 s with a mean amplitude of 0.1960.06 (DF/F 0 , arbitrary units). It occurred at a frequency of 0.3760.07 event/min (n = 54 slices, 3330 cells). Synchronous activity was observed in 2.6% spontaneous active astrocytes. We assessed the autonomous part of spontaneous calcium activity recorded in astrocytes (i.e. independent of neuronal activity) by incubating slices with tetrodotoxin (TTX, 500 nM) to block neuronal transmission. The effect of TTX was analyzed on a 5 minute recording and compared to a previous recording of calcium activity performed in the same area just before treatment. The inhibitory effect was then compared to a control experiment, performed in the same conditions but without drugs, and normalized. In presence of TTX, the proportion of astrocytes displaying spontaneous activity was significantly lower (52.069.9% inhibition; p,0.001; n = 13 slices from 10 rats, 908 cells; Fig. 3F) indicating that astrocytic calcium activity partially depends on neuronal activity. In the residual active astrocytes, TTX had no effect on calcium activity parameters (such as frequency, duration and amplitude) suggesting that this remaining calcium activity is totally independent of neuronal activity. Moreover, a separate analysis of TTX-sensitive and TTXinsensitive cells performed in the first recording movie (i.e. without TTX) revealed no obvious difference in the pattern of calcium signaling between these two sub-populations. As in basal conditions, no synchrony between active cells was observed in the presence of TTX.
Incubation of slices with thapsigargin (2 mM during 30 minutes; a Ca 2+ store depleting agent) significantly decreased the number of active astrocytes (72.469.1% inhibition; p,0.001; n = 7 slices from 5 rats; 738 cells; Fig. 3F). This suggests that Ca 2+ stores constitute the main route for the astrocytic calcium activity but extracellular Ca 2+ source may also be involved.
These data provide the first evidence of spontaneous calcium activity in the astrocytes of the SNr in basal conditions. Half of this spontaneous activity is autonomous; the other half depends on neuronal activity.

Glutamate and GABA receptors mediate calcium signals in SNr astrocytes
The most widely accepted mechanism for astrocytic Ca 2+ increase is dependent on G protein-coupled metabotropic receptors that leads to release of intracellular InsP 3 -activated Ca 2+ stores [43]. Alternatively, extracellular Ca 2+ may enter the cell through ionotropic receptors (e.g. glutamate receptors) or voltage-gated calcium channels [19]. Given the large amounts of glutamate and GABA released within the SNr, we first characterized the receptors and channels expressed by SNr astrocytic cells, by applying glutamate and GABA (leading to a calcium rise in glial cells since its action on GABA A receptors leads to a membrane depolarization in these cells [44]). Bath applications of glutamate (100 mM) induced a calcium increase in 59.368.8% These agonist-induced calcium responses may reflect a direct effect on astrocytic receptors and/or an indirect effect via neuronal receptors leading to secondary transmitter release. We therefore blocked neuronal transmission with TTX (500 nM) and observed that it did not inhibit the astrocytic calcium response induced by glutamate or GABA suggesting a direct action of these mediators on astrocytic receptors ( Fig. 4B and D; n = 12 and 9 slices respectively; p = 0.32 and 0.21 respectively).
These pharmacological data provide strong evidence for the expression of AMPA, mGlu and GABA A receptors by SNr astrocytes, all of which being potentially involved in calcium signaling in these cells.
Contribution of glutamate and GABA to the spontaneous calcium activity of SNr astrocytes As shown above, the subthalamo-nigral and pallido-nigral pathways were partially preserved in sagittal slices. As neurons are known to be spontaneously active in these nuclei, even in slice preparations [2,45], SNr activity is constantly modulated by the balance between GABA and glutamate, originating from the GPe and STN, respectively. This balance could also potentially influence astrocytic activity within the SNr, as these cells were found to express the required receptors. As only half of the spontaneous calcium activity of astrocytes is autonomous within SNr, we investigated the relative contribution of glutamatergic and GABAergic influences to the non-autonomous part of spontaneous calcium activity. The application of a combination of glutamate receptors inhibitors (10 mM CNQX, 100 mM LAP-3, 50 mM AP-5) inhibited 63.6611.1% active astrocytes within the SNr (Fig. 4E and F; p,0.001; n = 10 slices from 7 rats, 629 cells). Bicuculline 20 mM (BMI) inhibited 47.7618.8% spontaneous active SNr astrocytes (p = 0.003; n = 7 slices from 5 rats, 447 cells).
Thus, like SNr neurons, SNr astrocytes display spontaneous activity that is partially dependent on the release of GABA and glutamate within the SNr.
Modulation of the activity of astrocytes by electrical stimulation of the STN STN-HFS is a remarkable tool for alleviating the motor symptoms of advanced Parkinson's disease but the cellular substrates responsible for these promising results remain a matter of debate. STN-HFS may produce its effects by activating the axons of STN neurons, STN afferents or fibers passing close to the stimulation site, leading to the release of glutamate and/or GABA within the SNr [15,16,17,35]. In normal rats, STN-HFS generates concomitant excitatory-inhibitory synaptic currents in SNr neurons by recruiting efferent and passing fibers [34]. Depending on the amplitude and frequency used, electrical stimulation has been shown to differentially impact the glutamate and/or GABA release within SNr [16,46]. These data raised questions about the effects of STN electrical stimulation on the activity of SNr astrocytes in sagittal slices. We applied bipolar electrical stimulation to the STN during a 5 minute recording period, varying amplitude or frequency parameters. First, using a stimulation frequency of 100 Hz, we varied the amplitude (Fig. 5A). We prior checked that single STN electrical stimulations were effective to induce postsynaptic currents in SNr neurons in our conditions. At 100 mA, STN-HFS triggered a significant decrease in the proportion of active astrocytes in the SNr (70.8610.6%; p = 0.011; n = 9 slices from 7 rats, 514 cells). At 200 mA, a significant increase in the proportion of active astrocytes was observed (153.4642.3%; p = 0.008; n = 15 slices from 12 rats, 1006 cells). This significant increase also occurred for an amplitude of 400 mA (148.2638.2%; p = 0.004; n = 14 slices from 10 rats, 867 cells) but was not observed at an amplitude of 500 mA (108.1618.7%; p = 0.375; n = 8 slices from 6 rats, 619 cells). Thus, like SNr neurons, astrocytes were differentially affected by STN-HFS induced imbalance in glutamate/GABA release.
In a second series of experiments, we set the amplitude at 400 mA and varied the frequency (Fig. 5B). At a low frequency (10 Hz), STN electrical stimulation had no effect on the proportion of active astrocytes (89.3622.5%; p = 0.896; n = 15 slices from 11 rats, 684 cells). Similarly, at 230 Hz, STN electrical stimulation did not induce any significant effect on the number of active astrocytes (114.7644.8%; p = 0.817; n = 5 slices from 4 rats, 183 cells).
Whatever the stimulation parameters used no significant difference in calcium signal duration, frequency or amplitude was observed when electrical stimulation was delivered to the STN. When STN-HFS was applied, the delay of occurrence of the first calcium event in active astrocytes was 110.768.0 s (n = 9 slices, 121 active cells). This delay was not significantly different from the one measured without STN-HFS (106.169.4 s; p = 0.537; n = 9 slices, 94 active cells). No synchronous activity between active astrocytes was highlighted during STN electrical stimulation. Thus, only the percentage of active astrocytes was affected by STN-HFS and these newly active astrocytes were recruited all along the STN-HFS period ( Fig. 5D and E).
At the end of the STN-HFS period, applied at an efficient amplitude (200-400 mA), the proportion of active astrocytes returned to its basal level and no residual effect was observed during a 5 minutes recording post STN-HFS (91.2612.0% residual activity; p = 0,383; n = 7 slices from 6 rats, 369 cells; Fig. 5D and E).
When the electrode was placed outside the STN, just to the side of the internal capsule or in a medial position, no significant response was observed within the SNr. Moreover, when TTX (500 nM) was applied locally in the SNr during STN electrical stimulation (100 Hz, 200 mA), we observed a significant inhibition of the SNr astrocytic calcium activity (24.967.2% remaining active astrocytes; p = 0.001; n = 9 slices, 463 cells; Fig. 5C). This residual effect corresponds to the autonomous part of the astrocytic activity. This implies that the observed effect was specifically due to the stimulation of subthalamo-pallido-nigral projections rather than to electrical field diffusion.
Thus, when the parameters used corresponded to those routinely used in parkinsonian patients (with the amplitude adapted to the size of the rat STN [46,47] i.e. 200 to 400 mA), electrical stimulation of the STN induced significant astrocyte activation in the SNr as testified by the increase in the proportion of astrocytes displaying intracellular calcium variations. This activation did not persist when STN-HFS is switched off.

Involvement of astrocytic AMPA, mGlu and GABA A receptors in the increase of SNr astrocytic calcium activity induced by STN-HFS
In order to assess the implication of GABA and glutamate release in the modulation of the astrocytic calcium response triggered by STN electrical stimulation, we locally inhibited these transmitter receptors and compared the effect of STN electrical stimulation (100 Hz, 200 mA) on calcium response in the SNr without inhibitor application. We investigated whether the calcium response was dependent on glutamate by locally applying glutamate receptor inhibitors in the SNr. CNQX (10 mM) decreased the astrocytic calcium response induced by STN-HFS (31.166.0% remaining active astrocytes; p,0,001; n = 9 slices, 505 cells). LAP-3 (100 mM) also significantly decreased the SNr calcium response (47.567.2% remaining active astrocytes; p = 0.01; n = 8 slices, 352 cells), whereas AP-5 (50 mM) had no significant effect (75.0369.2% remaining active astrocytes; p = 0,121; n = 13 slices, 487 cells). Application of a cocktail of glutamate receptor inhibitors (CNQX 10 mM, LAP-3 100 mM, AP-5 50 mM) to the SNr induced a significant decrease in the number of responding cells (34.264.9% remaining active astrocytes; p,0.001; n = 11 slices, 844 cells), in the same range as CNQX alone. These data suggest that an increase in glutamate release within the SNr during STN electrical stimulation is partly responsible for the astrocytic calcium response.
Local application of bicuculline (BMI, 20 mM) to the SNr significantly decreased the SNr astrocytic calcium response to STN electrical stimulation (27.163.8% remaining active astrocytes; p,0.001; n = 24 slices, 1495 cells; Fig. 5C), indicating that GABA is also involved in the STN-HFS effect on astrocytes within the SNr through GABA A R activation. In the same conditions, 2hydroxysaclofen (100 mM) failed to significantly inhibit the STN electrical stimulation effect (72.6611.6% remaining active astrocytes; p = 0,266; n = 7 slices, 380 cells), confirming that GABA B Rs are not involved in the SNr astrocytic response. Concomitant applications of glutamate receptor inhibitors and BMI had no additive inhibitory effect compared to those obtained with each group of inhibitor (not shown). These results suggest that glutamate and GABA are both involved in the astrocytic calcium response induced by STN-HFS. Astrocytic AMPAR, mGluR and GABA A R are involved in this activity increase.

Discussion
In this study, we used calcium imaging to monitor SNr astrocytic calcium activity and the effects of modulating some of the basal ganglia networks that are known to influence the regulation of the SNr output activity. We took advantage of the sagittal brain slice preparation that preserved the pace-maker activity of the STN-GPe-SNr loops to modify its activity and investigate the impact of STN electrical stimulation on the activity of SNr astrocytes. SNr astrocytes displayed spontaneous calcium activity in basal physiological conditions. About half (48%) of this activity was found to be autonomous (i.e. independent of neuronal activity) while the other half is dependent on the spontaneous release of glutamate and GABA, probably under the control of the STN-SNr and GPe-SNr pathways. Changes of the activity of STN-GPe-SNr loops (induced by STN electrical stimulation) affected this astrocytic calcium activity in a consistent way to the amount of current delivered to the STN. Astrocytic AMPAR, mGluR and GABA A R are involved in the specific response of SNr astrocytes when STN is electrically stimulated. Thus, in physiologically preserved active loops, astrocyte activity is affected by glutamate/GABA release as is SNr neuronal output activity.

Technical considerations and validity of the model
The excitability of astrocytes is mediated by increases in the intracellular Ca 2+ concentration [22]. Calcium imaging provides readout of this cell excitation, monitoring the activities of dozens of cells simultaneously. Its use on brain slices makes it possible to study cells in a physiological context, preserving some networks and environmental background. Astrocytes represent the major cell population in the SNr but their role remains unclear. In our loading conditions, few neurons were loaded in the SNr allowing specific study of the activity of astrocytes. However, we showed that most of the SNr neuronal cells in the brain slice were functional even if they were not loaded.
Sulforhodamine stainings revealed that more than 98% of SNr recorded cells were astrocytes, meaning that involvement of other glial cells may be considered insignificant.
The sagittal slice allows study of the communication between astrocytes and neurons in an integrated loop endowed with spontaneous pace-maker activity. A major point is that we can modulate the SNr activity by electrically stimulating a distant nucleus (e.g. STN). These nuclei (STN, GPe and SNr) are of major importance within the basal ganglia network in physiological or pathological conditions (e.g. Parkinson disease) and are at the heart of the most famous example of a functional restoration of the network activity by STN-HFS. Thus, this model will be very useful for deciphering the nature of neuron-astrocyte communication within the SNr, in physiopathological conditions and under STN-HFS. More often, non physiological or direct stimulations were used in astrocyte studies in other brain areas [48,49,50]. Modulating STN-GPe-SNr loops with STN-HFS allows to interact in a physiological way with SNr astrocytes.

Evoked and autonomous spontaneous astrocytic calcium activity
We showed that only 48% of astrocytic calcium activity in the SNr is autonomous, i.e. independent of neuronal activity. In other brain structures such as the hippocampus [48,51] and somatosensory cortex [52], astrocytic calcium activity is known to be totally independent of neuronal activity and is therefore considered autonomous. Indeed, in these structures, there is no tonic spontaneous release of neurotransmitters at rest. The spontaneous pace-maker activity of STN-GPe-SNr loops and the continual release of glutamate and GABA within the SNr, probably account for the distinctive nature of the spontaneous calcium activity in SNr astrocytes, fitting with their environment and sensing the neurotransmitter released. Thus, the existence of two different modes of spontaneous Ca 2+ signaling within astrocytes, i.e. autonomous and activity-evoked, should have specific physiological roles fitting with astrocyte function in SNr [53].
The evoked part of spontaneous calcium activity is dependent, at least, on glutamate and GABA release within the SNr. GABAergic pallidal activity, glutamatergic STN activity and/or GABAergic intranigral activity may potentially contribute to this evoked basal glial calcium activity. In vivo, neurochemical data have shown that, within intact SNr, the extracellular glutamate concentration is 16 times higher than GABA [54]. Indeed, the proportion of glutamate-responding astrocytes is significantly higher than the proportion of GABA-responding astrocytes (59.368.8% versus 23.568.0%; p,0,001). This heterogeneity in astrocytic sensitivity might depend on their proximity to intranigral axon collaterals, GPe-SNr or STN-SNr projections, as described in hippocampus [55]. The major input to GABAergic SNr neurons arises from STN whereas GABAergic projections from GPe are fewer [56,57]. This particular astrocyte sensitivity will be further explored in pathological situations where the GABA/glutamate balance is disrupted.
Moreover, transmitters such as serotonin, endocannabinoid and dopamine could also be involved in part of this evoked spontaneous activity. We will further explore the contribution of these pathways to neuron-astrocyte communication within SNr.

Imbalance of the STN-GPe-SNr loops through STN electrical stimulation
Given the importance of the GABA/glutamate balance in the SNr output activity [13,14,15], studies on the effect of this balance on glial cell-dependent neuromodulation are of great interest. The efficacy of STN-HFS for alleviating Parkinson's disease symptoms has highlighted the value of treatment strategies bypassing the dopamine system to restore balance in the basal ganglia motor circuitry. In vivo, neurochemical data have shown that electrical stimulation of STN at clinical parameters induces an increase of both glutamate and GABA contents in the SNr of intact rats [15]. These data have been confirmed by electrophysiological studies performed in rats [13,34] and monkeys [58] demonstrating a concomitant excitatory and inhibitory effect of STN-HFS on SNr neuronal activity mediated by glutamate and GABA action. These complex effects are known to depend on current intensity [34,46] and frequency [11,16]. In vivo, a weak current intensity (,60 mA) did not affect extracellular glutamate and GABA levels in SNr of intact rats, [46]. In our model, we showed that a weak intensity (100 mA) significantly decreased the proportion of active astrocytes in the SNr. This could be due to an imposed decrease of the activity pattern of STN neurons as reported in rat brain slices at such a low intensity [11]. On the contrary, when intensities are close to clinical parameters, STN-HFS is known to impose a new activity pattern on STN neurons [11] together with activation of pallidonigral fibers and the intranigral axon collaterals of SNr cells [13] leading to concomitant glutamate and GABA release in SNr during the stimulation period [46]. In agreement with these data, we found that the proportion of active astrocytes significantly increased at 200 and 400 mA. This increase was due to the activation of astrocytic AMPA, mGlu and GABA A receptors. Thus, there is a window of STN-HFS intensity that enhances calcium activity in astrocytes through an increase of GABA and glutamate release. Higher intensity (500 mA) had no effect on the astrocyte calcium activity even though an effect on SNr neuronal activity has been reported in rat brain slices [34].
In the same way, it has been shown that low frequency (10 Hz) STN electrical stimulation did not modify ongoing STN activity [11] and did not affect glutamate and GABA levels within SNr in intact rats [33]. Concordantly, we showed that low frequency (10 Hz -400 mA) STN electrical stimulation had no effect on the proportion of active astrocytes. At 350 Hz, it has been reported that only GABA increased, due to a differential recruitment of STN and GPe fibers [16]. In our model, at 230 Hz (we could not go further without damaging the slice) there is no effect on the proportion of active astrocytes. It is possible that the frequency threshold to recruit only GPe fibers was not reached. It could also be due to a differential distribution of nearby GABAergic projections from GPe vs intranigral axon collaterals or to regionalization of GABA A R within astrocytes leading to a lesser sensitivity of SNr astrocytes to GABA release from GPe.
Altogether these results reveal that modification of GABA/ glutamate release strongly affect not only neurons but also the activity of the astrocytic compartment. Only the proportion of active astrocytes is enhanced suggesting a supplemental recruitment of astrocytes, potentially close to activated synapses. This recruitment is concomitant with STN-HFS and did not persist after the STN-HFS has been switched off. Thus, the additional calcium signaling may activate specific astrocytic pathways suggesting functional consequences on overall activity of SNr.

Functional relevance
Calcium is considered as a key second messenger in astrocyteneuron communication. Astrocytes use a different language, a different way of getting input and output, on a totally different timescale from neurons but superimposed with it. The functional consequence on the information leaving the SNr might be complex to decipher. Modulating the astrocytic calcium activity in a physiological loop would allow the study of its impact on SNr neuronal signaling. Astrocytes are involved in various and ubiquitous functions within the brain, some of them being calcium-dependent [39]. Indeed, astrocytic calcium activity has been involved in short-term and long-term plasticity [59] but a better understanding of calcium dynamics, signaling and gliotransmitter release is necessary to underline calcium as the key second messenger in astrocyte-neuron communication and neuronal plasticity in Parkinson disease. Few studies have focused on the involvement of astrocytes in the basal ganglia network activity. In normal rats, it has been shown that astrocytes, via the uptake of neurotransmitters, increase the strength of filtering operated through striatal medium spiny neurons [60]. In parkinsonian monkeys, a significant expansion of the astrocyte coverage has been described in glutamatergic striatal synapses providing evidence for compensatory changes that affect both neuronal and astrocytic elements in striatum, suggesting an active role in transmitter homeostasis [61]. Optogenetic studies, performed in hemiparkinsonian rats, reported that light stimulation of STN astrocytes inhibited neuronal firing in STN although no motor consequences were observed [62]. Thus, it becomes crucial to fully take into account and investigate the contribution of astrocytes in the physiology and pathophysiology of the basal ganglia output structures. Combining calcium imaging, patch-clamp experiments and physiological modification of the STN-GPe-SNr loop activity will allow us to go further in this field.

Conclusion
The originality of this study lies in its novel approach, complementary to the traditional research route focusing mainly on the influence of neurons within pathophysiological mechanisms. More specifically, it brings new insight concerning the role of astrocytes in deep brain stimulation mechanisms. STN-HFS affecting SNr astrocytic calcium activity suggests a potential functional role for astrocytes in pathophysiological processes concerning this basal ganglia output structure. The glutamate/ GABA balance is already known to be of major importance in the regulation of SNr neuronal activity, mainly through the drastic effects of its deregulation. For the first time, we show that modification of glutamate and GABA release also has a strong impact on astrocyte activity.The actual knowledge on neuronastrocyte communication prods us into hypothesize that astrocytes could be effective partners in the modulation of SNr neuronal activity and in counteracting abnormal neuronal activity within the basal ganglia network. The next step will be to further study the impact of the astrocytic calcium activity on the SNr output neuron activity either in physiological or in dopamine denervation conditions. . Moreover, all animal procedures were carried out in accordance with the French guidelines on the use of living animals in scientific investigations with the approval of the ''Grenoble Institute of Neurosciences ethical committee'' (agreement number 004). Every effort was made to minimize the number of animals used and their suffering during the experimental procedure. Rats were killed by decapitation and their brains were removed rapidly and cut into slices with a vibratome VT1000S (Leica, Wetzlar, Germany) in cold low Ca 2+ -high Mg 2+ artificial cerebral spinal fluid (ACSF) containing: 126 mM NaCl, 2.5 mM KCl, 7 mM MgCl 2 , 0.5 mM CaCl 2 , 1.2 mM NaH 2 PO 4 , 25 mM NaHCO 3 , 11 mM D-glucose, bubbled with 95% O 2 /5% CO 2 . Sagittal slices containing the GPe, the STN and the SNr were placed in ACSF (126 mM NaCl, 2.5 mM KCl, 1.2 mM MgSO 4 , 2.5 mM CaCl 2 , 1.2 mM NaH 2 PO 4 , 25 mM NaHCO 3 , 11 mM D-glucose) saturated with 95% O 2 /5% CO 2 and supplemented with 1 mM sodium pyruvate at room temperature, for a recovery period [63].

Projection tracing
We checked the integrity of the STN-SNr and GPe-SNr pathways on the sagittal slices by injecting DiI into these nuclei. We fixed acute slices in 4% paraformaldehyde solution (Electron Microscopy Science, Hatfield, Pennsylvania, USA) overnight at room temperature. We then placed crystals of DiI (1,19dioctadecyl-3,3,39,39-

Cell viability
In some experiments, cell viability was estimated at various stages of the slicing, loading and recording protocol, by propidium iodide staining (0.75 mM; Sigma-Aldrich) for 1 minute. The slices tested had previously been incubated with Hoechst 33342 (5 mM; Molecular Probes) for 30 minutes in the dark. Images were acquired with an upright compound microscope (Eclipse E600 FN, Nikon, Paris, France) equipped with a water immersion 206 objective (NA 0.5) and a confocal head (confocal C1 head, Nikon, Paris, France).

Sulforhodamine staining
Sulforhodamine 101 sodium salt (Sigma-Aldrich) was injected intravenously at a volume of 100 ml using a concentration of 10 mg/ml in saline solution. Animals were briefly anesthetized using isoflurane in a 70% air, 30% O 2 gas mixture for the intravenous injection (100 ml/50 g body weight). After 2 h, the rats were decapitated and their brains were rapidly removed. Acute sagittal brain slices were sectioned as described above.
When required, slices were loaded with Hoechst 33342 (5 mM; Molecular Probes) for 30 minutes to stain cell nuclei. Stained slices were then placed in a constantly perfused chamber (ACSF bubbled with 95% O 2 /5% CO 2 ) on the stage of a two-photon laser scanning microscope (TPLSM 7 MP, Zeiss, Germany) equipped with a 206 water-immersion objective (NA 1.0; Zeiss) and images were acquired with ZEN 2009 software. Laser excitation at 800-900 nm was carried out with a Ti:Sapphire laser system (Chameleon vision II; Coherent, UK). The excitation wavelengths used for the sulforhodamine dye and Hoechst were respectively 880 nm and 750 nm.
In some experiments, slices were loaded with Fluo-4 AM (5 mM; Molecular Probes) for 30 minutes and placed in a constantly perfused chamber (ACSF bubbled with 95% O 2 /5% CO 2 ) on the stage of an upright compound microscope (Eclipse E600 FN, Nikon, Paris, France) equipped with a water immersion 206 objective (NA 0.5) and a confocal head (confocal C1 head, Nikon, Paris, France). SR101 was excited at 543 nm and emission was filtered with a 605675 nm filter. Fluo-4 AM was excited with an argon laser at 488 nm and emission was filtered with a 515615 nm filter.

Stereological quantification
Acquisitions were performed with a translating platform with motorized crossed roller stages (8 fields of 425 mm6425 mm for each SNr area). Image processing was performed with ImageJ software [66]. Multistack (40 slices at 2 mm spacing) mosaics were reconstructed with the ''3D stitching'' plugin of ImageJ [67]. Quantitative analysis was based on a two step procedure. First, raw images were segmented by adaptive thresholding (Bernsen) to facilitate automated cell detection. The result was then analyzed with the ''3D object counter'' plugin [68] in ImageJ. The results are presented as mean 6 SEM.

Calcium imaging on brain slices
Brain slices were loaded with the calcium indicator dye Fluo-4 by immersion for 30 minutes at 35uC in a bath containing 5 mM Fluo-4 AM (Molecular Probes, Eugene, Oregon, USA), 0.005% Pluronic acid F-127 (Molecular Probes), 0.00025% Cremophor EL (Sigma-Aldrich) and 0.05% DMSO (dimethyl sulfoxide, Sigma-Aldrich) in ACSF. The loading chamber was continuously oxygenated with 95% O 2 /5% CO 2 . Slices were then placed in ACSF saturated with 95% O2/5% CO2 and supplemented with 1 mM sodium pyruvate, at room temperature for at least 45 minutes before imaging. Next, slices were placed in a constantly perfused chamber (ACSF bubbled with 95% O 2 /5% CO 2 ) on the stage of an upright compound microscope (Eclipse E600 FN, Nikon, Paris, France) equipped with a water immersion 206 objective (NA 0.5) and a confocal head (confocal C1 head, Nikon, Paris, France). Experiments were initially performed at both 35uC or at room temperature. No difference in the percentage of activated cells was observed so, for practical reasons, most of the experiments were then performed at room temperature. Excitation was achieved with light at 488 nm from an argon laser and emission was filtered with a 515615 nm filter. Images were acquired with EZ-C1 software (Nikon, Paris, France), in a 12-bit encoded format. Images were taken at 800 ms intervals in a single confocal plane over a period of 5 minutes.
Pharmacological agonists, such as glutamate, NMDA and GABA (Sigma-Aldrich), or inhibitors, such as tetrodotoxin (TTX; Latoxan, Valence, France), thapsigargin (Molecular Probes), CNQX, L(+)-2-amino-3-phosphonopropionic acid (L-AP3), D(2)-2-amino-5-phosphonopentanoic acid (AP-5), bicuculline (BMI) and 2-hydroxysaclofen (Sigma-Aldrich), were perfused in the bath. The inhibitory effects of antagonists on spontaneous calcium activity were assessed by comparison to a first recording of calcium activity performed in the same area. In these conditions, the percentage of residual active cells in a control experiment was not significantly different in the second recording (79.567.4%, p = 0.183, n = 18 slices from 11 rats, 1077 cells) and there was also no significant difference in the calcium signal parameters (amplitude, duration, frequency). The inhibitory effect of all tested antagonists was assessed by comparison with this control experiment and normalization. The evoked effect of glutamate and GABA on astrocyte calcium activity and the inhibitory effect of antagonist on exogenous glutamate and GABA applications were assessed by analyzing the cells on addition of the agonist to the bath and in the 10 images following agonist addition. The inhibitory effects of antagonists on STN-HFS induced activity were compared versus a control group of HFS where two successive stimulations were applied to the slice. In that case, the HFS #2 represented 97.3611.6% of the HFS #1 response (p = 0,570; n = 9 slices from 8 rats, 754 cells). In these experiments, inhibitors were applied locally in order to affect the SNr but not the STN or GPe.

Quantification and statistical analysis
CalSignal software was used to detect cells and measure their intracellular Ca 2+ activity [69]. This software can automatically detect and track hundreds of cells per slice and carries out statistical analysis for numerous parameters, analyzing the fluorescence signal F within each region of interest. F 0 was calculated for each region of interest (ROI) on the recording period or, when an agonist was applied, it was calculated on the period before the drug application. Based on the DF/F 0 ratios, significant fluorescence variations were detected and their amplitude and duration were calculated. These data were exported to SigmaStat software (Systat) for statistical analysis. All results are presented as mean 6 SEM. When inhibitors were applied, each slice was its own control since we compared the effect of the pharmacological agent tested on spontaneous calcium activity with an initial recording in the same area during the same duration before treatment. Statistical analyses were carried out with Mann-Whitney-Wilcoxon tests with significance levels of *, p,0.05; **, p,0.01 and ***, p,0.001.
We performed cross-correlation analysis to estimate synchrony between cells. The cross-correlations between the calcium signals of couple of cells were calculated at time lags varying between 23 and +3 s using the ''crosscoef'' function in Matlab. When the cross-correlation coefficient exceeded 0.7, the cell couples were considered as synchronous [48].

Electrical stimulation procedures
A concentric bipolar stimulation electrode (platinium/iridium; Frederick Haer & Co., USA) was positioned in the middle of the STN within the brain slice. The electrical stimulation consisted of multiple rectangular pulses (0.1 ms duration; 10,100 or 230 Hz) of constant current (100 to 500 mA) applied during the 5 minute recording period. These parameters correspond to those routinely used in parkinsonian patients, but the amplitude was adapted to the size of the rat STN [46,47]. At intensities above 600 mA at 100 Hz or 400 mA at 250 Hz, stimulation resulted in tissue damage (burn and lesion). As for the application of inhibitors described above, the impact of STN electrical stimulation on astrocytic calcium activity parameters was determined by comparison with an initial recording of the same area. The positive or negative effect of STN-HFS was assessed by comparison with the previous control experiment and normalization.

Supporting Information
Figure S1 S100B and GFAP immunostainings in the SNr. (A) Two-photon multistack (9 slices at 2 mm spacing) mosaic reconstruction of S100B immunostaining in the SNr of a fixed brain slice. Antero-posterior (AP) and medio-lateral (ML) orientations are shown in the upper right part of the figure. (B) Confocal images of S100B (green) and GFAP (red) immunostained cells in the SNr of a sagittal section of rat brain. Merged image showing the proportion of S100B and GFAP positive cells (right panel). Nuclei are identified by TO-PRO staining (blue). Arrow indicates a GFAP positive cell that is S100B negative and arrowheads show several S100B-positive cells that are GFAP negative. (TIF)